This action might not be possible to undo. Are you sure you want to continue?
Celso von Randow
Prof. dr. A. A. M. Holtslag Hoogleraar Meteorologie, Wageningen Universiteit Dr. B. Kruijt Universitair Docent / Onderzoeker, Alterra Wageningen Universiteit en Researchcentrum
Promotiecommissie: Prof. dr. P. Kabat – Wageningen Universiteit Prof. dr. A. J. Dolman – Vrije Universiteit Amsterdam Prof. dr. B. J. J. M. van den Hurk – Universiteit Utrecht Dr. L. D. A. Sá – Museu Paraense Emílio Goeldi, Belém, Brazil
Dit onderzoek is uitgevoerd binnen de onderzoeksschool “Buys Ballot Onderzoeks-School”.
On turbulent exchange processes over Amazonian forest
Celso von Randow
Proefschrift ter verkrijging van de graad van doctor op gezag van de rector magnificus van Wageningen Universiteit, Prof. dr. M.J. Kropff, in het openbaar te verdedigen op woensdag 30 mei 2007 des namiddags te 13:30 in de Aula.
With summaries in English and Dutch. C. Wageningen University. PhD thesis. Wageningen. 2007. On turbulent exchange processes over Amazonian forest..Von Randow. The Netherlands. ISBN: 978-90-8504-640-0 iv .
who dreamt about this. even before I knew what is a dream v .To Roberto von Randow.
Table of Contents Abstract Samenvatting Chapter 1 . Wavelet transforms i 1 3 5 5 6 8 13 18 21 22 24 25 28 28 33 37 41 45 51 55 56 60 62 62 63 . Radiation balance 2. Determination of power spectra and cospectra low frequency ends of the turbulent variables 3. Sites description 2. Conclusions Chapter 3 .1.5 Research questions and thesis organization Chapter 2 .1. General Introduction 1.4.4. Introduction 18.104.22.168.5.3. Net Ecosystem Exchange of CO2 2. Scales of motions in the atmospheric boundary layer 1.3.4. The challenge of measuring fluxes in Amazonian forests 1. 3.2.3 Overview of experimental sites 1.Scale variability of atmospheric surface layer fluxes of energy and carbon over a tropical rain forest in South-West Amazonia. Introduction 22.214.171.124. Sensible and latent heat fluxes 126.96.36.199.4.2.2. Results and discussion 2. Description of measurements and instruments 2. Scale variability analysis theoretical elements 3.1.Comparative measurements and seasonal variations in energy and carbon exchange over forest and pasture in South-West Amazonia. Site description and deployed instruments 3.3. Flux data processing and energy balance closure 2.3. Meteorological conditions and soil moisture storage 2. 2.Introduction 1.4.
Exploring eddy covariance and scintillometer measurements in an Amazonian rain forest 5.4. Introduction 4.4. Standard deviations scaled by dissipation velocity 4.2. Conclusions Appendix A Chapter 5 .3. Footprint estimation ii 2 65 66 67 68 71 73 76 79 80 83 84 86 86 88 90 91 92 93 94 95 95 99 101 107 110 112 4. Scale variances 3. Theoretical background 4.7.4. Results 4.3. ε and CT2 4. Wavelet statistics 3.2. Estimation of ε and CT 4.5.2. Data screening 4.2.4. Lower frequency time series analyses and Taylor’s hypothesis range of validity 3. Scale fluxes 188.8.131.52.184.108.40.206.3. Temperature-humidity correlation and Bowen ratio scale dependence 3.1 Fluxes using scintillometry 5. Site description and data processing 4. Flux-variance method – MOS scaling 220.127.116.11.2.Low-frequency modulation of the atmospheric surface layer over amazonian rain forest and its implication for similarity relationships 4.3 Roughness sublayer height 115 116 118 118 120 .1.1.1. Scale heat flux and mean horizontal wind speed relationship 3. Variances and fluxes separated in classes 3. Scaling on ‘dissipation velocity’ 4. Monin-Obukhov scaling and the flux-variance method 4. Summary and conclusions Chapter 4 .4.2.6. Low frequencies and correlation coefficients 4.4. Discussion 18.104.22.168.2.6. Results and discussion 22.214.171.124. Relations between spectra.3.1. Theoretical background 5.2.2. Introduction 5.1.
5. Results 5.2 Perspective and recommendations References Acknowledgements About the author 121 121 123 125 132 134 137 137 141 143 159 163 iii .2.4. Experimental description 5.3.5. Correcting for Tower Vibrations 5.1 Summary 6.6. Discussion 5.1 Measurements and site description 5. Conclusions Chapter 6 .3.Summary and Perspectives 6.3.
The influence of low-frequency motions on similarity and correlations between turbulent signals and the implications for the application of Monin-Obukhov similarity is investigated.Abstract This thesis deals with turbulent exchange processes of heat. and in these cases the surface layer is different from the textbook descriptions. Additionally. the results suggest that one single eddy covariance system is not able to capture quasi-steady large-eddies that significantly contribute to surface-atmosphere exchange over Amazonian forest. Apart from possible horizontal flux divergence at heterogeneous terrain. The latter quantity decreases when the influence of low-frequency motions in the surface layer is high. and the averaging time scale can be reduced.5. these factors may explain the failure to close the surface energy balance in complex terrain. Largest contributions to measured fluxes occur in turbulent scales (structures with length scales up to 1000 m and time scales up to 15 min). it is studied how turbulence statistics depend on different time scale and spatial scale classes. the use of Large Aperture Scintillometry (LAS) over Amazonia is explored. it is not necessary to use a long time period to sample a number of eddies. however. can contribute with up to 30 % to the total exchange under weak wind conditions. The results show that the EC fluxes are often lower than the LAS. For the estimation of heat fluxes by the flux-variance method it is found that reasonable results only occur when the correlation coefficient between vertical wind and temperature (rwT) is above 0. decomposing the turbulence signals using multi-resolution (wavelets). As the LAS provides a measurement that represents a weighted spatial average of the turbulent eddies along the path. humidity and carbon dioxide over Amazonian forests. The atmospheric boundary layer over Amazonia frequently contains slowly-moving large eddies induced by strong convective motions or local circulations related to the heterogeneity of the surface. To better understand their influence. Low-frequency motions (larger eddies and mesoscale motions). 1 . The difference increases with increasing non-stationarity conditions and decreasing correlation coefficient In general. in comparison with the eddy covariance (EC) method.
In deze gevallen verschilt de oppervlaktelaag van de standaarddefinities. Voor het schatten van warmtefluxen met de flux-variantie methode vonden we dat dit alleen goed werkt als de correlatiecoëfficiënt van vertikale windsnelheid en temperatuur (rwT) groter is dan 0. die zich meestal uitstrekken van het landoppervlak tot aan de top van de grenslaag. en op de consequenties daarvan voor de toepasbaarheid van Monin-Obhukov similariteits theorie. 3 .5. goed te beschrijven. Over het algemeen lijken de resultaten uit te wijzen date en enkel eddy-covariantie systeem niet in staat is om bijna-stationaire grote wervels. Deze wervels. Daarnaast is de toepasbaarheid voor Amazonebos getest van Large Aperture Scintillometry (LAS). zijn lange middelingtijden om een minimum aantal wervels te registreren niet nodig. bestuderen we hoe de statistische kenmerken van turbulentie nabij het Amazone-landoppervlak varieren met de ruimtelijke. De resultaten van deze vergelijking laten zien dat de EC fluxen vaak lager zijn dan die van de LAS. in vergelijking met de eddy-covariantie (EC) methode. De grootste bijdragen aan turbulente uitwisseling (fluxen) zoals die vaak gemeten wordt vinden plaats in lengteschalen van 1000 m of tijdsschalen van 15 minuten. Dit verschil neemt toe naarmate de turbulentie minder stationair is en de correlatiecoefficient rwT afneemt. Dit doen we door gemeten tijdsseries van turbulentie (variatie van windsnelheden) te door middel van wavelet analyse te ontbinden. worden veroorzaakt door krachtige convectieve bewegingen of door heterogeniteit in het landschap. Meer in het bijzonder bestuderen we de invloed van laag-frequente turbulentie op de gelijkvormigheid en correlatie tussen turbulente signalen. Afgezien van de horizontale flux-divergentie die in heterogeen terrein optreedt. een alternatieve techniek om warmtefluxen te meten. bij lage windsnelheden. tot 30% bijdragen aan de totale uitwisseling. is het waarschijnlijk dat dit aspect mede verantwoordelijk is voor het feit dat de energiebalans in complex terrein meestal niet sluitend is. Omdat LAS een ruimtelijk gewogen gemiddelde meting geeft van de turbulente wervels langs het 1-5 km lange meetpad van dit instrument. zich langzaam voortbewegende wervels. Deze grootheid neemt af wanneer de bijdrage van laagfrequente bewegingen in de oppervlaktelaag groot is. Om het belang van deze structuren beter te begrijpen. die in de uitwisseling tussen Amazonebos en de atmosfeer een groet rol spelen. Maar laagfrequente luchtbeweging (wervels groter dan 1000 m en meso-schaal circulaties) kunnen.en tijds-schalen waarop ze worden bepaald.Samenvatting In dit proefschrift bestuderen we turbulente uitwisselingsprocessen tussen de atmosfeer en Amazonebossen. De luchtbeweging in de atmosferische grenslaag boven Amazonia wordt gekenmerkt door een relatief groot aandeel van grote.
interdisciplinary research program led by Brazil. ecological. For several years now. the preferred methodology to measure surface fluxes has been the eddy covariance (EC) technique. they have also raised awareness of the limitations to the application of the eddy covariance method over complex surfaces such as tropical tall forests and heterogeneous terrain. Eddy covariance measurements have significantly improved our understanding of the response of the forest ecosystem to environmental conditions. biogeochemical and hydrological functioning of Amazonia. high time-resolution integral exchange data for the whole ecosystem. energy and momentum.Chapter 1 Introduction 1. the impact of land use change on these functions. 2002. and the interactions between Amazonia and the Earth system (Avissar and Nobre. known as the Planetary Boundary Layer. Gash et al. as it provides direct. designed to study the climatological.g. through the exchange of mass (e. However. The LBA is a multinational. water and CO2). Measuring EC fluxes at micrometeorological towers over rain forests. General Introduction The Earth’s surface is in constant interaction with the lower part of the atmosphere. This interaction plays a key role in the climate system and in the hydrological and biogeochemical cycles.. the core of many studies of surface-atmosphere interaction is to quantify the surface fluxes over various types of land cover. In order to understand and predict these cycles in the climate system. 2004). savanna and grasslands has been a key objective of the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA). The basis of the EC method is to quantify the surface-atmosphere exchange by measuring variations of the vertical 5 .1.
The reduction of variations at intermediate scales observed by Van der Hoven (1957) is commonly referred to as the spectral gap (Stull. as well as parameters that characterize the turbulence in the atmospheric surface layer. 1988). We begin by reviewing the characteristics of turbulent motions near the surface.2. 1. Scales of motions in the atmospheric boundary layer The intensity of flow variations in the atmospheric boundary layer varies at spatial and temporal scales ranging along a wide spectrum. trying to improve our understanding of the processes that underlie the variations at these timescales. These scales are not independent: smaller spatial variations usually have a short lifetime and larger systems last longer. we investigate how fluxes. Over complex surfaces. In this thesis we study aspects of the influence of these slowly-moving eddies and local circulations on turbulent exchange over relatively complex sites in Amazonia. however. vary as a function of averaging time scales from seconds to hours. 1957. and at spatial scales of millimeters to a few kilometers.windspeed and the atmospheric quantity of interest (scalars).1 (Van der Hoven. Although this textbook spectrum has 6 . It is assumed that the averaged product of temporal fluctuations in the wind and scalars measured at one point is equivalent to the spatially averaged exchange by turbulent eddies carried with the mean flow. that weaken the validity of these assumptions. In general. Stull. from a tower extending above the vegetation. Turbulent variations occur at time scales that range from seconds to a few hours. frequently the lower portion of the boundary layer (surface layer) can be affected by slowly-passing large eddies. A schematic (textbook) representation of the intensity of variations at different scales (turbulent energy spectrum) in the atmosphere near the ground is shown in figure 1. such as most LBA sites. or local circulations due to topographical or surface cover heterogeneity. 1988). It is also usually assumed that the flow field is horizontally homogeneous so that horizontal flux divergences and advection are neglected. Variations of several kilometers and time scales of days or weeks are associated with the passage of fronts and weather systems.
spoiling a clear scale separation and making it a difficult regime to properly describe (Stull. and turbulent fluctuations. (ii) the inertial subrange. with larger cumulus clouds acting like large eddies with time scales on the order of an hour. In practice. 1988). Almost all theoretical and experimental studies of atmospheric turbulence rely on the separation between larger scale phenomena and turbulent motions by the spectral gap. where the bulk of the turbulent energy is produced by buoyancy and shear of the flow. and (iii) the dissipation range.1. which are treated statistically. we also identify in figure 1. but handed down from larger to smaller scales. Often. many studies report that the spectral gap often cannot be observed. Schematic of energy spectrum near the ground.1 the three regions relevant to turbulent flows: (i) the production range. where the turbulent energy is neither produced nor dissipated. Considering the spectral region of turbulent scales. where turbulence is finally dissipated and its energy is converted to internal energy (heat).1. which are treated deterministically. as for example. 7 . the spectral gap provides a means of defining a cut-off timescale to separate the mean (background) motions. showing idealized separation of synoptic scale processes and turbulent processes. however. Synoptic scales Spectral gap Turbulent scales Spectral energy Production Inertial range Dissipation Timescales Weeks – 1 day ~1h 10 min – seconds wavenumber . there is no spectral gap. These might enhance the spectrum around these scales.been generally accepted in the past. frequency Figure 1. which suggests that a more appropriate general form of the spectrum would follow the dashed line in figure 1.
Finnigan et al.. Moreover. 8 . such as wind components and scalars. 1997). however.. (ii) contributions from updrafts and downdrafts associated with strong convective activity. Figure 1. Low-frequency motions can modulate the turbulent fluxes at relatively long timescales. Sakai et al. As for longer time scales the frequencies are lower. boundary-layer flow variations at scales of the order of an hour or more have been commonly referred to as low-frequency motions. These sites are particularly prone to conditions of low wind speed and deep convective boundary layers above. into mean and fluctuating (turbulent) parts. coupled with slow changes in scalars (Finnigan et al.. (iii) elongation of eddies in the downwind direction and formation of ‘roll’ vortices that are quasi-permanent in position (LeMone 1973). (iv) slow wind direction variations in complex terrain. 2001. The sources of low-frequency motions are not yet well understood. Recent re-evaluations of the optimum periods for averaging (e.2 shows a schematic map with their locations. We give a brief overview of the experimental sites in this section. deep convective boundary layers and greater measurement heights are frequent. compared to the usual averaging times used by the flux community. This is particularly the case in tropical forests. 1994. 1. 2003).In the past decades the micrometeorology community has almost universally adopted time periods of between 10 and 60 minutes as the averaging interval over which to separate variables. with typical characteristics of variable topography and contrasting vegetation near the edge of forest and deforested areas.. the heterogeneity of the landscape is large..g. 2003).. where days characterized with low wind speed. Sun et al.3 Overview of experimental sites In this thesis we analyze the surface fluxes and aspects of scale variability of turbulent motions with data from three sites in Amazonia. Factors that can enhance variability of the flow at timescales of 1 hour or more (where we otherwise would have a spectral gap) include the following : (i) local circulations induced by heterogeneity of the terrain (Mahrt et al. suggest that at many sites longer averaging periods may be appropriate.
Since early 1999 until the end of 2002. the site has relatively strong seasonal rainfall variability.Manaus K34 FNS Rebio Jaru Figure 1. in the Brazilian state of Rondônia. Rebio Jaru Rebio Jaru is a terra firme forest reserve located about 100 km north of Ji-Paraná. Schematic map of South America. in south west Amazonia. 9 . The forest has a mean height of about 35 m. a. but some of the higher trees (‘emergents’) reach up to 45 m. and K34. in central Amazonia. and an extra mast built on top of the tower held the eddy covariance instrumentation at 62. Located in the south west periphery of the Amazon Basin. including detailed mesoscale characterization of cloud and rain processes (Silva Dias et al.. several field campaigns were made at the site. with the location of the experimental sites Rebio Jaru and FNS.2. A micrometeorological tower installed in this site was 60 m tall.7 m above the ground.
the fetch is mainly pristine rain forest for several km. VOCs and impact on atmospheric chemistry (Andreae et al. Although the forest reserve is protected by the Brazilian Environmental Protection Agency (IBAMA). in the south-west of Brazilian Amazonia. The right panel shows a zoom in the surrounding area of Rebio Jaru site. showing a peculiar ‘fish-bone’ pattern of deforestation in the region. In the predominant wind direction. 10 .. The arrow indicates the predominant wind direction. Satellite images of the central part of Rondonia state. heat. a long-term monitoring project was conducted. water and CO2 and meteorological conditions. continuously measuring surface fluxes of radiation. apart from a small very recently deforested area north of the tower (figure 1. The left panel depicts a broader region.3).3 shows an image of the central part of Rondonia state. The proximity of deforested areas on the other side of the Ji-Paraná river (western boundary of the forest reserve) makes this an interesting site for the study of the influence of low-frequency local circulations on turbulent processes. surrounding areas have been substantially deforested in the past 30 years. Figure 1. Figure 1. with alternating patches of forest and degraded land. 2002). with the fish-bone pattern of recent deforestation (left) and a zoom of the surrounding area of Rebio Jaru site (right). This monitoring program provided the 4-year long dataset analyzed in this thesis. Apart from the campaigns.3. These studies are presented in Chapters 2 and 3.2002) and studies of trace gases. Areas covered with forest are depicted in dark color and degraded areas in light colors.
Compared to the Rebio Jaru site. (2002) presents the general characteristics of this site and the long-term measurements of surface fluxes. since mid 1999. when low-frequency motions are present (chapter 4). at K34 the variable topography can also influence the flow field creating the important low-frequency variations (see next section). Average vegetation height is 30 m near the tower. the forests in Central Amazonia have almost no disturbance or deforestation. Fluxes are measured at a height of approximately 53 m above the ground. 2004). In the panels on the left side of the figure.. The region is a vast area of dense upland broadleaf tropical vegetation. separated by steep slopes and broad. Compared to areas closer to the southern and eastern borders of Amazonia. The landscape pattern of plateaus and valleys in the region also motivated us to explore the application of an alternative method to estimate the heat fluxes: the scintillometer 11 . Araujo et al. the K34 site is relatively homogeneous. in the Brazilian state of Amazonas. often waterlogged. In this region. similar to the Rebio Jaru site. with a representation of the topography. due to the contrast between forest and deforested areas in that landscape. valleys. the plateau areas are depicted in light grey and the valleys in dark grey. Manaus K34 K34 is an experimental site situated also in a terra firme forest reserve. a long-term monitoring micrometeorological tower installed at the K34 site measures the heat. It is expected that local circulations will be more prominent at Rebio Jaru. Figure 1. As part of the LBA project. Although the landscape of the region is covered with pristine forests. we also investigated how the characteristics of the flow at K34 differ from the expected ‘textbook’ surface layer. from about 20 to 45 m. water and CO2 fluxes.4 shows a detailed image of this landscape. located in central part of Amazonia. it consists of a mosaic of well-drained dissected plateaus. but varies considerably. With the objective of studying the influence of the low-frequency motions in the structure of surface layer turbulence.b. the LBA K34 site was established at about 50 km north of Manaus. One example is a situation where large convective cells persist for hours locked onto fixed positions in the landscape (Malhi et al. along with conventional meteorological variables. Nonetheless.
12 . also as part of this PhD study.4. and a Large Aperture Scintillometer (LAS) was installed. overseeing the valley between the two towers. method. By providing a measurement related to a spatial average. we explore the application of the LAS in comparison with EC to analyze the effects of low-frequency motions on overall flux measurements in the K34 site (chapter 5). 1.X Manaus – K34 Data (UTM *) Figure 1. For this. In this thesis. Panel on the bottom of left side shows a representation of the landscape topography around the two towers installed at the site. relative to EC. Detailed images of the K34 site in Central Amazonia. the LAS may more appropriately sample the effect of a larger number of turbulent eddies in a shorter time interval. and this measure can be related to a weighted spatial average of the heat flux (Hill. 1992. a new 45-m-tall tower was built on a plateau1 km north of the main EC tower (see detail in fig. De Bruin. A scintillometer is an instrument that measures the amount of scintillations in the air by emitting a light beam from one tower to the other. 2002).N Elevation Manaus LAS tower ta (U TM*) 200 150 100 50 1 km 9000 9500 10000 EC tower 11500 11000 10500 10500 11000 11500 10000 12000 UTM - Y Da 13500 13000 12500 12000 UTM .4).
as the dominant species.) Stapf. The FNS is located about 85 km away from Rebio Jaru (see indication in figure 1. heat. 1.c. since 1991.4. The area was first deforested by fire in 1977 and. data collected from the FNS cattle ranch is also presented in this thesis. in Amazonia it is very unlikely to find a location with homogeneous terrain. the idealized conditions of clear scale separation between the turbulent and larger scale motions rarely happen. The atmospheric boundary layer over Amazonian forest is often characterized by the presence of strong convective systems that strongly influence the structure of small scale turbulence near the surface. To illustrate the challenge of measuring surface fluxes in Amazonian sites. Moreover. has consisted of a homogeneous sward of perennial grass.3). to study the different functioning of the two vegetation types and how these variables change with the drastic reduction in rainfall during the dry season in the region (chapter 2). with Brachiaria brizantha (A.5 shows a representation of the scale variability observed during daytime at the three locations cited in the previous section: (a) Rebio Jaru forest. as in many natural landscapes worldwide. There is much insight to be gained from the long dataset collected at the Rebio Jaru and FNS sites. The challenge of measuring fluxes in Amazonian forests The motivation for this study comes from the fact that. Fazenda Nossa Senhora (FNS) In addition to the two forest sites. in Amazonia. figure 1. water and CO2 fluxes. (b) K34 forest. and (c) Fazenda Nossa 13 . energy and CO2 fluxes have been continuosly measured at a 4 m tower in the site. Garstang and Fitzjarrald (1999) characterize these conditions as a disturbed state of the classical boundary layer. with peculiar inter-scale links. Rich. and we use the comparative measurements of radiation balance components. Since early 1999.
5c). The data plotted in this figure were generated using multi-resolution decomposition.5b).g. Contribution of motions at different scales to the sensible heat flux over Rebio Jaru forest.5a). Comparing the three sites. it is observed that the contribution from the largest scales (lower frequencies) is more important over the forest sites than over grassland. 1988. motions on time scales beyond 15 min (corresponding to scales of 1 km or more) strongly contribute to the fluxes. Rebio Jaru 40 Scale sensible heat flux [W. Note that these contributions are also highly variable and can be of either sign. The thick line represents binned averages and the dots are individual observations. Then figure 1. pg. 14 . 1997). 5-6). Stull.5 presents on the x-axes both the temporal scales and the spatial scales of motions.Senhora (FNS). and on the y-axes the contribution of motions at a particular scale to the heat flux. but much less at the FNS (fig 1. a type of wavelet transform (see e. Howell and Mahrt.m ] 30 Surface type: Forest Measurement height: ~ 63 m -2 20 10 0 -10 length scale (m) 0.g. Especially in Rebio Jaru (fig 1. Significant low-frequency contributions are also observed at the K34 forest site (fig 1. to project the turbulent signals onto temporal scales and assess the contribution of motions of different scales to the ‘total’ covariances.5a.1 1 10 100 time scale 15 m 30 m 1m 1s 1h in ec in in -20 1000 10000 Figure 1. where the latter are estimated using the average wind velocities and the assumption of Taylor’s hypothesis (see e.
1 1 10 100 time scale in mi n 30 mi n 1m 1s 1h ec -20 1000 10000 Figure 1.5a.1 1 10 100 time scale 1m 15 m 30 m 1s 1h in in ec in -20 1000 10000 Figure 1. but for the K34 forest. Same as 1.m ] -2 Surface type: Grassland Measurement height: ~ 4 m 30 20 10 0 -10 length scale (m) 0. but for the FNS grassland.5a.5c.5b. Same as 1.m ] -2 30 Surface type: Forest Measurement height: ~ 52 m 20 10 0 -10 length scale (m) 0. 15 15 .K34 40 Scale sensible heat flux [W. Fazenda Nossa Senhora (FNS) 40 Scale sensible heat flux [W.
6. can largely increase the uncertainty in flux estimations. the cut-off timescale should include 16 . because it is uncertain whether the mechanisms creating the imbalance affect the CO2 exchange in a similar way. due to the relatively high measurement heights and factors described in the previous section.. relative to the measured available energy. it is important to extend the record intervals to include the active contribution of low frequencies. On one hand. the equation that describes this partition can be written as Rn = H + λE + G + S (1. why at these sites the energy fluxes seem to be systematically underestimated. A lack of energy balance closure reduces the usefulness of eddy covariance data for model validation and parameter calibrations. λE is the latent heat flux (energy used for evaporation E. so this would be an appropriate choice of cutoff scale to separate means and turbulent fluctuations. energy used for evaporation and energy used to heat the soil and the vegetation biomass. for Rebio Jaru (fig 1. In many studies the measured heat fluxes (H + λE) are underestimated by about 10-30 % relative to estimates of the available energy represented by the remaining terms in equation 1. when no spectral gap exists.1) where Rn is net radiation. At the surface. however arbitrarily extending the averaging interval.6b) forest sites. with λ being the latent heat of vaporization). This energy ‘imbalance’ can also be seen in figure 1. 1998). G is the ground heat flux and S is the energy stored in biomass and canopy space.Over the grass vegetation at FNS. Clearly. 2004).6a) and K34 (fig 1. At the forest sites. the thermodynamic energy available through the net radiation balance is partitioned into sensible heat flux. It also raises concern in carbon budget studies. at least partly. however. The complications related to the influence of low-frequency motions revolve around whether these variations in long timescales are “locally meaningful” or represent features of the wider landscape that are not related to the local surface (Malhi et al. These complications might explain. the low-frequency contributions are much more prominent. In a common form. H is the sensible heat flux. the choice of the cut-off averaging timescale to define mean and turbulent fluctuation parts of turbulent variables may become arbitrary. it is apparent that most of the turbulent transport occurs at time scales up to 15 min. When there is no clear spectral gap defining a separation between turbulent and larger-scale processes.1 (Rn – G – S). by increasing random errors and non-stationarity (Mahrt.
this is still at very premature stages.84x 2 R = 0. all scales that carry a significant amount of mass and energy flux.89 2 400 400 200 200 0 -200 -200 0 200 400 600 -2 0 800 1000 -200 -200 0 200 400 600 -2 800 1000 Available energy (W m ) Available energy (W m ) Figure 1.6. Better understanding of how the low-frequency motions affect the structure and scales of turbulent processes is therefore crucial for the reduction of uncertainties of flux estimates and for the improvement of future models of turbulent transport. in W m ) Rebio Jaru 1:1 1000 Sum of heat fluxes (H + λE. Ideally. 17 . Sum of heat fluxes against available energy measured at (a) Rebio Jaru and (b) K34 forest sites in Amazonia.74x R = 0. On the other hand. however. these models should include the effects of low-frequency modulation (by larger-scale motions). in W m ) 1:1 K34 -2 -2 800 800 y = 0.(a) (b) 1000 Sum of heat fluxes (H + λE.91 600 600 y = 0. the basis of nearly all models of surface-atmosphere interactions relates the fluxes to the local mean wind shear and temperature stratification.
that is the small-scale part of the spectrum. A criterion to separate the contribution from ‘turbulent’ and ‘mesoscale’ classes is presented. scintillometry) in comparison with a point measurement. Chapter 3 describes in more detail the characteristics of scale dependence of fluxes observed in Rebio Jaru. to assess the contribution of motions of different scales to the total variances and covariances. based on the variability observed at each scale. and quantify these fluxes in the surface layer over Amazonian forests? • How do the low-frequency motions affect the structure and scales of turbulence in the surface layer? • What are the advantages of using a spatially integrating technique (e. 18 .5 Research questions and thesis organization The large variability and influence of low-frequency motions in the surface layer and the lack of energy balance observed in Amazonia motivated the studies presented in this thesis. Chapter 2 discusses the energy balance closure problem at Rebio Jaru and possible sources of low-frequency motions and uncertainty in flux measurements. for example) of the two types of surface. we decompose the turbulent signals on classes of temporal scales using wavelets. in the presence of low-frequency motions? These questions are investigated in the following chapters.1. of the energy partition between sensible and latent heat fluxes.g. and of carbon fluxes. the comparative long-term flux data collected at Rebio Jaru and FNS grassland provide valuable information of temporal trends (seasonal or inter-annual variations. The results of detailed comparisons of radiation components. are presented. even subject to uncertainties in the absolute accuracy. the main research questions investigated are the following: • How can we separate low-frequency fluxes from ‘regular’ turbulent processes. With the objective of separating contributions from turbulent and larger-scale processes. measured at the two sites. In summary. Nevertheless.
In Chapter 5 we present the measurements of a Large Aperture Scintillometer (LAS) in comparison with measurements of the eddy covariance system. 2004. It is suggested that the spatial averaging feature of LAS measurements provides a better sampling of slowly moving eddies. also analyzing the conditions of low-frequencies modulation and non-stationarity. Finally Chapter 6 gives the overall summary and conclusions of this thesis. but also properties from outer layers and larger scale processes can play a role. some overlap occurs in methods and experimental sites description. 19 . In this chapter we explore the application of the fluxvariance method. 2006). Although MOS is based on empirical relationships derived from local properties of flows over uniform surfaces.In Chapter 4 we study how the low-frequency motions affect the structure and scales of turbulence in the surface layer over K34 forest. As such. in comparison with the point measurement of the EC. based on using the dissipation rate as a scaling parameter (McNaughton. which is based on MOS scaling parameters. where not only the local properties. compared to expected behaviour from studies at uniform surfaces. The chapters of this thesis are integral copies of articles published in (Chapters 2 to 4) or submitted to (chapter 5) relevant peer-reviewed journals. and what is the implication for the application of Monin-Obukhov Similarity (MOS). and an alternative scaling approach. it is widely used in flux estimations and models in complex terrain.
B. compared to the forest.Chapter 2 Comparative measurements and seasonal variations in energy and carbon exchange over forest and pasture in South-West Amazonia. as part of the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA). The night-time respiration in the pasture is also reduced compared to the forest. since the growth of the vegetation is constantly renewed. During the dry seasons. Gash. doi 10.1007/s00704-004-0041-z. Zanchi. this causes an average reduction of 13. Theoretical and Applied Climatology. von Randow.H. Silva. R. 5–26. J.J. ____________________ This chapter is published as C. For the radiation balance. P. compared to the forest. the soil moisture storage profiles down to 3. M. M. P. the reflected short wave radiation increases by about 55 % when changing from forest to pasture. the combined effect is a 19 – 67 % higher daily uptake of CO2 in the pasture. Kruijt. Elbers. The data were collected from February 1999 to September 2002. de Oliveira. although precipitation and specific humidity are greatly reduced. compared to the large reductions in evaporation and photosynthesis observed in the pasture. seasonal changes observed in the energy partition and CO2 fluxes in the forest are small. 78. Cardoso.4 m indicate that the forest vegetation continues to withdraw water from deep layers in the soil.C.L.A. As the reduction in the nocturnal respiration is larger than the reduction in the daytime uptake. with averages 44 % and 57 % lower in the wet and dry seasons.7 % in long wave radiation loss. Abstract Comparative measurements of radiation flux components and turbulent fluxes of energy and CO2 are made at two sites in South West Amazonia: one in a tropical forest reserve and one in a pasture. Waterloo. as the cattle remove the biomass. Daytime CO2 fluxes are 20 – 28 % lower (in absolute values) in the pasture compared to the forest. For this reason. the evaporative fraction (λE/Rn) at the pasture is 17 % lower than at the forest.L. This high uptake of CO2 in the pasture site is not surprising. F.O. Manzi.G Hodnett. F. respectively. In the wet season. 21 . Combined with an increase of 4.3 % in net radiation in the pasture. A. Kabat. This difference increases to 24 % during the dry season.J. 2004. B. J.
The fluxes are being measured using the eddy covariance method. Miller et al. 1999. 1995. Garratt.g. 1998. Araujo et al. most of these long term flux studies focus on primary forest. water and carbon fluxes from Amazonian forest . which provides additional motivation for comparative studies of the fluxes in different regions of Amazonia. as it provides direct. (2004) have recently highlighted some interesting differences in the seasonal variability of carbon exchange across Amazonia. (1990) and Grace et al. 1991. Vourlitis et al. high timeresolution data representing the energy and gas exchange for the whole ecosystem. The forest generates a major part of global land surface evaporation (Choudhury et al.. (1995). (1984) and Fitzjarrald et al. The Bowen ratio (β). More recently. and of carbon exchange of Fan et al. Several measurement projects are now underway measuring the spatial and inter-annual variability in energy. These studies have significantly improved the understanding of the response of the forest ecosystem to environmental conditions.. is a critical influence on the hydrological cycle.. Keller et al. through its role in boundary layer development. Viterbo and Beljaars.. 2004.1.. Keller et al.. weather and climate. 1993) which have shown that the performance of atmospheric models depends on an accurate representation of these surface 22 . (1988). 2004). 1998).. the ratio of the sensible (H) and latent (λE) heat fluxes. 2002. However. 2001). The eddy covariance technique was used in campaign mode by the pioneering studies of Amazonian forest evapotranspiration of Shuttleworth et al. and little attention is paid to the impacts of changes in surface vegetation cover. This influence is emphasised by various modelling studies (e. it has been used by the flux community to study seasonal response and annual budgets of canopy carbon assimilation and forest respiration (Malhi et al. and plays a prominent role in the atmospheric carbon balance (Malhi et al. Dickinson et al.2.the objective is to calibrate basin-wide modelling schemes so that accurate estimates and predictions of these exchanges can be made. over several square kilometres. To assess the effects of these changes requires comparative measurements over different vegetation covers. Introduction An increasing number of observations and model results show that Amazonian forest has a key influence on the regional and global climate system.
We discuss the differences in the short wave and long wave radiation. estimating the long wave radiation balance from the residual of the net radiation... Sá et al.g. 23 . One deficiency of these studies is that they only had independent measurements of the short wave components of the radiation balance. Measurements of short and long wave radiation.. (1998) have studied the fluxes from Amazonian pasture there has been relatively little consideration of the possible impacts of land use changes. about 80 km away. in February 1999 two of the former ABRACOS field sites in the Brazilian state of Rondônia were reactivated. although Wright et al. a few studies with short term measurements (e... In Amazonia. such as a reduction of up to 20 % in precipitation and an increase of 2 oC on surface temperature (Nobre et al. 1984.. Other comparative studies show that net radiation over forest areas is higher than in pasture areas. and recently with two one-year datasets. an area of little disturbed rainforest vegetation.. Gash et al. The main results of ABRACOS indicate large potential impacts on regional climate. one in the Biological Reserve of Jaru (henceforth referred to as Rebio Jaru). 1996). climatological and hydrological processes. (1992. Culf et al.processes. 1996). 1996) and Grace et al. in the energy partition between sensible and latent heat fluxes and in the carbon fluxes at the two sites. 1998). energy and carbon balances over forest and deforested areas in Amazonia. 2002) and eastern Amazonia (Rocha et al. heat. a large grassland ranch. 2004) have discussed the energy partition and the controls on the seasonal variation of these fluxes. Conversion of tropical forest in Amazonia to pasture and agricultural plantations may lead to impacts on the regional ecological.. in central (Malhi et al. 1996). due to differences in the reflected solar radiation (albedo) and in the long wave radiation balance (Bastable et al. as part of the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA).1993. water and CO2 fluxes were made almost continuously from February 1999 to September 2002 at these two sites. of the radiation. Recognising the need to better understand the surface processes. Shuttleworth et al. The climatic and hydrological effects of possible Amazonian deforestation were the main subject of the Anglo-Brazilian Amazonian Climate Observation Study (ABRACOS. This paper highlights the differences and seasonal variability in the surface energy and carbon exchange at these sites with their two contrasting vegetation covers: tall tropical forest and short grassland. and one in the Fazenda Nossa Senhora ranch (FNS). both in the wet and dry seasons. These studies have been focused on forest areas.
has consisted of a homogeneous sward of perennial grass. about 50 km from Ji-Paraná. An extra mast of 2. at 10 º 45’ S and 62 º 22’ W. in the remaining sectors the undisturbed fetch is shorter. a 5 m tower set up in 1991 was used during ABRACOS.706' S. at the height of 145 m A. 2002). with altitude varying between 100 and 150 metres above sea level. but also due to a few hills near the tower.L.S. At this site. 293 m above sea level. Rondônia. Andreae et al. 2002) and LBA/EUSTACH (European Studies on Trace gases and Atmospheric Chemistry. The canopy has a mean height of 35 m. however we should keep in mind that the heterogeneity may cause important disturbances in the wind and scalar fields. The pasture site is located in the cattle ranch Fazenda Nossa Senhora (FNS).2. Nevertheless.). This area was first deforested by fire in 1977. Sites description Rebio Jaru is a terra firme forest area located about 100 km north of Ji-Paraná. as the dominant species. At the end of 1998. since 1991.2. Andreae et al. The River Ji-Paraná forms the western boundary of the reserve. On the other side of the river the rain forest has been progressively cleared during the last 25 years. of the order of 1 km. The closest hill is about 2 km northeast of the tower. Rich. However some of the higher trees reach heights up to 45 m. we can consider the fetch to be predominantly undisturbed rain forest. the predominant wind direction. In summary.) Stapf. (2002) give details of the vegetation at this site. especially close to its north-western border. It consists of around 268000 ha of undisturbed tropical forest and is located between 10 o 05’ S and 10 o 19’ S and 61 o 35’ W and 61 o 57’ W. as part of two subprojects: LBA/WETAMC (first LBA major wet season Atmospheric Mesoscale Campaign. where it is relatively vulnerable to invasions by landless people.. The surface heterogeneity of the area is observed not only because of contrasting deforested areas close by.027' W. However.7 m was built later at the top of this tower. with Brachiaria brizantha (A.. are very small. the few disturbed areas in the sector from northwest clockwise to south-southeast. In early 1999. 61 o 56. and. another tower (8 m tall) was built 70 m away from the original 24 . Although the forest reserve is protected by the Brazilian Environmental Protection Agency (IBAMA). a 60 m tower was built at this site (10 o 4. in recent years it has been suffering some small scale slash and burn activities. compared to the extensive areas of pristine forest. Silva Dias et al. Brazil. near the town of Ouro Preto D’Oeste.
(1996). Didcot Instr. which may 25 . the maximum depth of measurement varied from 2. This led to an important difference in the soil moisture storage behaviour at the two sites. creating saturated conditions. a few cold fronts sometimes penetrate into the far north of the South American continent during June-July. causing the so-called “friagem” events. While the predominant climate is equatorial. The southern hemisphere summer is the rainiest period in the region. The instruments. 2. in the forest. However. with monthly totals over 200 mm.2 m depth intervals to a maximum depth of 3. Description of measurements and instruments Several measurements were made continuously at both sites from the beginning of the activity in early 1999. There is sufficient uniform fetch for 1-2 km in all directions. warm and moist. In contrast during the dry season from June to August. to support the LBA measurements. the precipitation varies strongly with seasons. Co.0 m to 3. Automatic weather stations (AWS) composed of commercial sensors were installed on the towers and provided averaged measurements of the usual weather variables every half hour. and the average temperature is almost constant throughout the year.one. when relatively low temperatures may be observed. Measurements were made at 0.1.6 m.3. Measurements of soil moisture were also made weekly at both sites. A separate mast of 4 m was also set up to support the flux instruments. Rondônia state. because of the presence of hard weathered bedrock in the profile. In the pasture the measurements could be made to a depth of 3.6 m in 6 access tubes (locations) in the pasture and in 8 tubes in the forest. in the south west part of Amazonia does not suffer large influences either of the sea or of topography. data acquisition and power supply systems are nearly identical at both sites and are briefly described in this section..6 m in all tubes. it is common to observe several weeks without rain. water draining from the profile in the wet season ponds on the underlying bedrock. using a neutron probe (Model IH. but in the forest. UK). Details of the vegetation at both the Rebio Jaru and FNS sites are given by McWilliam et al. A full list of the measurements and instruments used on both sites is given in Table 2.
following the methodology described by Moncrieff et al. Although the signals from the IRGA are corrected for analyser cell temperature and pressure.1 s response time) infrared gas analyser (IRGA. Very little drift in the calibration values was observed. Such saturated conditions rarely occurred in the pasture. the effect on the calculated fluxes is almost negligible. usually smaller than 1 % in the span of the IRGA. USA). LICOR. To prevent dust entering the sample tubing. (1997) and Aubinet et al. additional CO2 and H2O concentrations inside and above the canopy were measured at six heights up the tower. 5 m in length. using a membrane air pump (KNF. USA). This IRGA was stable because the single analysis cell was thermostatically controlled and zeroed on a half-hourly basis with a chemically scrubbed air circuit. Germany) at a flow rate of about 7 L min-1. Both sites were equipped with an eddy covariance system. in the first one and a half year period of measurements. Generally.4 Hz for later off-line flux calculations. (1997). using a slow response infrared gas analyser (CIRAS SC. 26 . UK). using the Alteddy software (Elbers. LI-6262. In this setup. The distance between the inlet tube and the centre of the sonic anemometer transducer array is about 20 cm. Alteddy was written in FORTRAN language and can be adapted to a number of different hardware configurations and software options. PP Systems. For our sites. from an inlet near the sonic to the IRGA. similar in design to the systems described by Moncrieff et al. The air is drawn through a 4 mm internal diameter Teflon tube. the program was set to compensate the time delay of the IRGA signals and include corrections for instrument responses and damping of fluctuations through the IRGA tube. the instruments were recalibrated at intervals of about 2 months. Gelman. (2000).extend to within 1 m of the surface during very wet periods. these corrections are small and represent an almost negligible uncertainty factor to the final values (Kruijt et al. These signals plus temperature and wind velocities measured by the sonic were then recorded at a rate of 10. 2004). The ponded water drains downslope towards the Ji-Paraná river. Gill Instruments. At the forest site. Therefore. cross sensitivity and band broadening. composed of a three axis sonic anemometer (Solent 1012R2. 1998).. These are closed-path systems. the H2O and CO2 mixing ratio analogue signals output by the IRGA were fed into the sonic anemometer in-built A/D converter. air filters of 1 μm pore were used (ACRO 50. UK) and a fast-response (0. Since the calibrations need to be done in the field and as these are relatively remote sites (especially the forest site). zero and span calibrations are frequently necessary.
0 m (depth) 0.3 m 62.5 m 8. PT100 resistors** Vaisala thermohygrometer (HMP35A) Cup anemometers Vector A100R Wind vane Vector (W200P) Rain gauge EM ARG-100 Infrared sensor Heimann (KT15) Barometer Vaisala (PTB100A) PT100 resistors Infrared gas analyser PP Systems (CIRAS SC) Flux plates Hukseflux (SH1) Soil thermometers IMAG-DLO (MCM101) FDR sensors IMAG-DLO (MCM101) Neutron probe Heights FNS (pasture) 6. 0. 0. 35. 1.1. 25. 5.3 m* 25.0 m (depth) 0.0.05.5 m -1 and 5 cm (depth) 0.7.5 m 6.0. 0. water vapour and CO2 concentrations are installed only in the forest site. 34.05.0 m (depth) Every 20 cm down to 3. 0.01.6. temperature.126.96.36.199 m 6. 1.3. 0.Table 2. 1.7 m 60.15. 0.3.4 Hz) Eddy covariance system (Gill Sonic Anemometer and LI-COR 6262 IRGA) * Height above canopy top (~ 35 m) ** Vertical profiles of temperature.2. instruments and measurement heights for the automatic weather station and eddy covariance instrumentation installed on Rebio Jaru (forest) and FNS (pasture) sites in Rondônia.3 m* 19.0 m 61.0 m (depth) Every 20 cm down to 3.05.7. 0. 0.0. 45. manufacturer (model) Pyranometers Kipp & Zonen (CM21) Pyrgeometers Kipp & Zonen (CG1) Quantum sensor LI-COR (LI-190SZ) Vaisala thermohygrometer (HMP35A). 2. 45.0.3 m 9.5 m 9. wind speed.6 m (depth) 188.8.131.52.0 m Rebio Jaru (forest) 19.05 m 1 and 10 cm (depth) 0. 0. Meteorological variables Used Instrument.4.3 m 184.108.40.206 m 5.3 m 60.1.3 m 9m 0.3 m 8. 25. 0. 0.1 m 40 m 54.3 m 60.6 m (depth) 4. 45.4. 0.01. H2O and CO2 concentration (10. *** Soil moisture profiles with neutron probe measured once a week 27 .7 m Incident and reflected short wave radiation Incident and emitted long wave radiation Photosynthetically Active Radiation (PAR) Air temperature** Relative humidity Wind speed** Wind direction Rainfall Surface radiative temperature Atmospheric pressure Temperature of pyrgeometers Vertical profile of CO2 and water vapour concentration** Soil heat flux Soil temperature profile Soil moisture profile Soil moisture profile with Neutron probe *** High frequency measurements of 3-D wind speed. 35.7. List of measurements. 15.0 m 8. 0.05. 1. 25.6 m* 60.
In the pasture site. a lightning strike on the tower damaged most of the sensors.4. weather proof and low-power design. Results and discussion 2. water vapour and CO2 were estimated using the eddy covariance method.4. In the forest site. (2002) give a more detailed description of these systems. The AWS data coverage is generally very good. The set-up is very similar to the one used by another LBA forest site: the “K34” site.The whole set of instruments and data acquisition systems is low power consuming and is powered by batteries. respectively. acting as a filter for low frequencies. The data coverage during the period 2001 – 2002 is much better.1. near the city of Manaus. Especially during the first two years. with fluxes available for 62 % of the time in the forest and only 42 % in the pasture. in the forest and pasture sites. 2. Also. The large heterogeneity of the terrain is therefore likely to add large uncertainties associated with low 28 . 1988). two rotations were applied to align the coordinate frame with the mean streamlines and to force the mean vertical component ( w ) to zero (McMillen. Despite the very careful. The fluctuations of the variables were calculated by subtracting 30 minute block average values (or up to 8 hours in a low frequency study) from the instantaneous measurements. there have been several breakdowns in different parts of the system. leading to temporary or permanent replacement of some sensors and causing a few gaps in the dataset. Flux data processing and energy balance closure The fluxes of sensible heat. covering 92 – 99 % of the time in both sites. a faulty sonic could only be replaced several months after the detection of the problem. Araujo et al. For these reasons the flux data coverage during 1999 – 2000 was limited. The averaging and rotation period used defines the main scales of motion that contribute to the calculated transport of the scalars. recharged using solar panels. about 86 % and 84 %. serious problems happened at both sites.
These authors examined the effect of the angle of attack (the angle that the wind vector makes with the horizontal axis of the sonic) distribution on sonic anemometer-based flux measurements.74x R2 = 0. A large imbalance is shown in this figure. for a forest site. 29 .frequency contributions to the fluxes. in W m -2) Rebio Jaru 1:1 800 600 y = 0. The energy balance closure over the pasture site (not shown) is usually better.1 includes the angle of attack dependent calibration described by Gash and Dolman (2003) and Van der Molen et al.89 400 200 0 -200 -200 0 200 400 600 800 1000 Available energy (W m -2) Figure 2. where Rn is the net radiation and G is either the soil heat flux.1. Sum of sensible and latent heat fluxes plotted against the available energy for half-hourly measurements made during 1999.1. but still it is not always achieved (the sum of the fluxes in the pasture range from 80 to 110 % of available energy). or the sum of the soil heat flux and the heat storage in the canopy air space and biomass. 2. (2004). For example. as shown in fig. more than 50 % of the daytime fluxes 1000 Sum of heat fluxes (H + λE. Gash and Dolman (2003) found that. in the pasture. The available energy is calculated by subtracting the soil heat flux and the heat storage in canopy air space and biomass from the net radiation. the energy balance closure at the forest site is very poor. in the forest. The term A is calculated as A = Rn – G. The analysis shown in fig. 2. calculated using the parameterisation proposed by Moore and Fisch (1986). This figure presents hourly values of the sum of the sensible (H) and latent (λE) heat fluxes against the “available” energy A.
2000)..2 % and the latent heat fluxes by 7. 2004). Finnigan et al. In summary. Kruijt et al. 2003. von Randow et al. Van der Molen et al.. 2001. Finnigan. apart from errors related to instrument limitations. the recalibration increases the sensible heat fluxes by 8. 1998. 1998. Lee. 2000). Sakai et al. This lack of energy closure reduces the usefulness of the eddy covariance data for model validation or parameter calibrations. and (iv) especially for carbon budget studies.1 %. Several other studies have reported similar discrepancies between measured fluxes and available energy (Twine et al. There has been much recent discussion about the uncertainties and possible errors in flux measurements using the eddy covariance technique over complex surfaces (Mahrt. On average. reviewing the sources of uncertainties on all the processing steps of calculations of fluxes from the raw turbulent data. (ii) non-stationary conditions and disturbed boundary layers associated with strong convective events present in Amazonia. such as: (i) mesoscale circulation induced by horizontal heterogeneity of the terrain. there is still a shortfall of 26 % in the energy balance closure at the Rebio Jaru site and the reasons are unclear. 1999. the horizontal advection of scalars in stable conditions. Over the Rebio Jaru forest. there is the additional problem of poor nighttime mixing leading to potential advective 30 . We applied these corrections to a few days of Rebio Jaru data. For carbon dioxide fluxes. to some extent. (2004) measured the response of two types of sonic anemometer in a wind tunnel experiment and derived angle of attack dependent corrections. also pointed out the high sensitivity of the estimations to the treatment of low frequencies and non-horizontal flow in the region. which might not be taken into account if the fluxes were calculated using the usual short-averaging periods procedures. several distinct factors that contribute to the variability of surface exchange processes add complications to the estimation of the fluxes in the region by the eddy covariance method. (iii) the effects of slow wind direction changes in complex topography terrain that are likely to modulate the stream flows. Kruijt et al. All these problems might. explain the apparent underestimation of heat fluxes (lack of energy balance closure) and very large observed annual carbon uptake rates (usually not in agreement with ecological expectations. (2002) showed that a substantial amount of the exchanges of mass and energy between the forest canopy and the atmosphere occur due to mesoscale motions. A careful analysis of the possible sources of uncertainty is thus necessary. Even after correcting for the angle of attack on the sonic anemometer. (2004).were carried by eddies with angles of attack that were outside the manufacturer’s recommended operating envelope. Malhi and Grace.
with the energy balance closure increasing as the averaging time is increased from 15 minutes to 2 hours. since the uncertainties in the absolute accuracy of the flux measurements are thought not to vary too much between seasons. To analyse whether that is the reason the energy fluxes at the forest are apparently under-measured we performed a study reprocessing the data collected over the forest site for 48 days augmenting the averaging time periods up to 8 hours. This suggests that the rotations should only be applied on a long time scale basis. even subject to possible large uncertainties in the absolute accuracy.2 shows the variation of the energy balance closure (represented as the ratio of the sum of heat fluxes to the available energy) with the period used to calculate the mean and fluctuating parts of the turbulent fluxes. acts as a high pass filter for the covariances. Figure 2. while further increase of the period out to 8 hours has only a small effect on closure. The result is similar to the analysis for data collected near Manaus (central Amazonia) shown in figure 14 of Finnigan et al. the eddy covariance method is considered a powerful tool when used for analyses of temporal trends (seasonal or inter-annual variations.7 0. Nevertheless. 31 .2 Ratio of heat fluxes to available energy at Rebio Jaru for different averaging and rotation periods. such that the contributions from fluctuations over periods longer than the averaging period are lost.8 0.9 0. (2003).6 0 100 200 300 400 500 600 Averaging period (min) Figure 2. This effect is not likely to significantly affect energy fluxes as these are generally very small during nighttime. (2003) showed that the procedure to rotate the coordinate frame to be aligned with the streamlines. Finnigan et al.losses. at Rebio Jaru. for instance). However the main difference is that. the energy 1 energy balance closure 0.
the contrasting surface vegetation covers may induce mesoscale circulations that modulate the turbulent transports (von Randow et al. (2000). even when the averaging and coordinate rotation is extended so that all the low frequency contribution to the vertical transport is captured. the terrain is more complex: not only is the terrain not flat (there are a few hills in the surroundings of the tower. (ii) a significant amount of the energy is transported horizontally over the region by local circulations causing horizontal flux divergences. In that sense. deep moist convection allows very large convective motions to develop within the boundary layer. the heat fluxes at the soil surface and the energy storage in the biomass and canopy air space.. we can expect substantial low frequency contributions to the covariances. but there are also large deforested areas to the south and west of the reserve. For this reason. therefore. at a range of time scales. the heat fluxes can be adjusted in two ways: either simply 32 . We conclude that the lack of energy balance closure at our sites is the result of one or both the following reasons: (i) transports on time scales longer than 8 hours are still significantly contributing to the total exchange. by independently measuring the net radiation. rather than on the methodological issues. Regardless of the predominant wind direction from the north and eastern sectors. at both sites the high measurement heights allow these slow motions to impact on the measured fluxes. Unfortunately. 2003).balance closure does not reach the 100 % level. however. Both at Manaus and at Rebio Jaru.. As discussed by Twine et al. A new experiment is needed to specifically investigate the spatial variability at this site. steady horizontal flux divergences may still contribute to the total balance and these cannot be estimated from measurements made on a single tower (Finnigan et al. Sufficiently far above the ground. Hereafter we concentrate on comparative measurements over the two contrasting types of surface. As a result of slow wind direction changes there still may be a low frequency component that we are not able to capture using short time rotation scales. At Rebio Jaru. it is simply not possible to estimate this component from measurements made on a single tower. This experiment should include a detailed investigation of the interactions of flow with the region topography and vegetation cover. 2002).
Meteorological conditions and soil moisture storage This and subsequent sections present the results of several aspects of the data. but the differences are very small.2. but λE is likely to be more subject to uncertainties in the measuring device (in our case. 2. a clear drop in specific humidity and a drastic reduction in rainfall during the dry seasons is observed.3. The mean air temperature shows some variability between months.. instead of just one weakly defensible method. Surprisingly. is more appropriate when it is likely that the underestimation of the fluxes are caused not by the instrument limitations. probably related to strong precipitation situations. through almost the entire period of precipitation over pasture compared to forest areas. The alternative method of adjusting both fluxes. keeping the measured Bowen ratio.4. and in June-July. ranging only between 22 and 27 ºC and it is not easy to identify a clear seasonal pattern. 2000) and we have also decided to adjust the fluxes in both ways. at both sites. A few low temperature events seem to drive the averages down in some years in January-February. Monthly averages of air temperature and specific humidity and monthly totals of rainfall measured over the pasture (FNS) and forest (Rebio Jaru) sites during the period February 1999 to September 2002 are shown in fig. Comparisons are made between the measurements over the pasture and forest sites and between wet and dry season periods. 2. Applying the first procedure is justifiable if it is assumed that H is accurately measured. both approaches are used (Twine et al. This approach provides a range of values for the fluxes.discarding the latent heat measurements and estimating this component as the residual of the energy balance or adjusting both the sensible and latent heat flux maintaining the ratio between them (Bowen ratio) as measured by the eddy covariance system. due to the influence of the so-called “friagens” (the cold fronts that can penetrate far north in the continent during these months). when the current data are compared with 33 . In the literature. On the other hand. the closed-path IRGA) than H (calculated only from the sonic measurements). but because of a failure to capture low frequency transport or advection.
4 shows monthly average values of specific humidity deficit in g kg-1. 1996). from February 1999 until September 2002. over the forest and pasture sites. while in the pasture the average values are 13. Figure 2. This figure also highlights the strong seasonality effect on air humidity at both sites. Although previous studies indicate a reduction in the previous data collected during ABRACOS (Hodnett et al. than in the pasture.. with average values ranging from 15.4 and 16. it clearly shows that the deficit is consistently greater at the pasture site than over the forest. specific humidity (inverted triangles) and monthly totals of precipitation (columns) measured over the forest (represented by closed symbols) and the pasture (represented by open symbols). the rainfall amounts were much higher in the forest. Monthly averages of air temperature (circles). the specific humidity deficit is also of interest.0 g kg-1 in the dry and wet seasons. In addition to the specific humidity.3. 34 . observations. respectively. As expected. where the measurements indicated about 1400 – 2000 mm of rainfall per year.Air temperature ( C) and specific humidity (g kg ) -1 28 26 24 22 20 18 16 14 12 10 Forest Pasture 700 600 Precipitation (mm) 500 400 300 200 100 0 o Jan/99 Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Jul/02 Jan/03 Figure 2. It is not possible to say whether this large difference is real or an artefact of inaccurate measurements in one or both of the sites. where annual amounts ranged from 2000 to 2400 mm. the differences are much higher in this current data.5 g kg-1 in the wet seasons.8 g kg-1 in the dry seasons to 17. The specific humidity is also always higher in the forest area.
Monthly averages of specific humidity deficit measured over the forest (closed circles) and the pasture (open circles).2 m and 2 – 3. for both forest and pasture (It should be noted that the storage values are the equivalent depth of water in the profile. and were mainly due to drainage. The data for both forest and pasture show a very pronounced seasonal cycle. the largest decreases of storage in the lower profile occurred at the end of the wet season and in the early dry season.4 m in the soil profile. The storage in the 2 – 3.Specific humidity deficit (g kg ) -1 12 10 8 6 4 2 0 Jan/99 Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Forest Pasture Jul/02 Jan/03 Figure 2. This is not seen in the pasture.5 shows the storage of water for the layers from 0 . during July 2001. indicating that the profile was saturated. in the upper and lower layers shown. in precipitation. and by lateral drainage. In the lower profile. January 2001 and January 2002 was the result of the saturated conditions extending upward to less than 1 m below the surface. the seasonal change was about 110 mm in the pasture compared to about 290 mm in the forest. In the pasture. based on the soil moisture determined by drying the soil in an oven at 105 ° C. but the seasonal changes are very much larger in the forest than the pasture. the rate of moisture loss from the 35 . Figure 2. The values give no indication of water availability). the rate of moisture storage change in the forest is very rapid compared to that in the pasture. more importantly. The drastic reduction in humidity and. The very high storage in the 0 – 2 m layer in March 2000. because the profile is losing water by root uptake (to supply transpiration).4. In the pasture. At the start of the dry season. from February 1999 until September 2002. has impacts on the soil water storage behaviour.4 m layer in the forest remained constant in the wet season.
The loss rates for the lower profile show a large contrast. with more uptake in 1999 and 2002 compared to in the other dry seasons. the rates of moisture loss for the same period were 1. This indicates that the limit of water availability was being reached. upper profile was 2.5. In the forest. This can be seen by comparing the forest curves in fig. the minimum storage reached in each of the four dry seasons shown was very similar.5b).5a and 2.0 m Pasture Forest (b) 2. 2. with little root uptake from below 2 m in the pasture. the minimum storage reached varied between the four seasons. giving a total of 3.64 mm d-1 respectively. The rate of loss from the upper layer in the pasture was higher than that of the forest in this period.4 m Pasture Forest Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Jul/02 Jan/03 Figure 2. but 36 .0 . In the pasture. The storage in the lower layer decreases during all dry seasons (fig.4 m deep. For the forest.26 mm d-1 compared to only 0.2 m and (b) 2 – 3. compared to the forest. but this probably reflects the fact that the forest had already used a greater proportion of the available water from this layer. 2. for the layers from (a) 0 . Storage of water in the soil at forest and pasture sites.46 mm d-1.5b.3.Storage of water in soil (mm) 800 600 400 200 800 600 400 200 Jan/99 (a) 0 .45 mm d-1 from the lower profile.82 mm d-1 and 1.2. In the forest it was of note that uptake in the lower layer ceased soon after the storage in the upper layer had increased following rainfall inputs. in the upper layer.
Incident long wave radiation (square symbols on fig. and incident (Lin) and emitted (Lout) terrestrial radiation. whenever available within the 4 years studied. and also between wet and dry season periods. comparing the measurements over the two sites (comparing the curves with closed and open symbols). data between January and March were used. two composites representing wet and dry season periods were calculated. 2. From these figures we can compare the behaviour of each component over the two types of surface and during the wet and dry seasons.6. provided by the wet and dry season composites.6d. From the same figures it can also be seen that the solar radiation reflected by the pasture vegetation is higher than the forest. Since the data coverage is usually very good for all the radiation components. 2.6c and 2. For the wet season composites. 2. To assess these. usually in the beginning of September.1 ) which is the summation of short wave and long wave radiation components: incident (Sin) and reflected (Sout) solar radiation.as soon as an increase is observed in the upper layer (fig 2.6d) is more or less similar over the two 37 . are presented in fig. 2. it is likely that these composites are good representatives of the average conditions during wet and dry season periods. The average long wave components. as shown in fig.6c and 2.6a and 2. All of the components. incident solar radiation is slightly higher over the forest than in the pasture. shown in fig.4.6b.3. especially the upward ones may present differences over different vegetation covers. As the rainfall inputs continue through the wet season the water storage is largely increased in both layers. also present some interesting features. a levelling in the lower layer is noted. For the dry season composites. 2. The daily patterns of the radiation components. averaging the measurements each half hour for all of the components. data between July and September were used.5a). First. Radiation balance The net radiation (or radiation budget) at the surface is described by: Rn = (Sin – Sout) + (Lin – Lout) ( 2.
Measurements at forest are represented with closed symbols and at pasture. shows marked differences between the two types of surface. circles). The outgoing terrestrial radiation. In both seasons. Bottom panels: same as top panels. squares) and emitted (Lout..Wet season (Jan-Mar) 800 Dry season (Jul-Sep) a S in S out Rn b Forest Pasture 600 Fluxes of radiation (W m-2) 400 200 0 550 L in 500 450 400 350 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 L out c d Local time (hours) Figure 2. The result. is that the long wave radiation budget is larger (more negative) in the pasture than in the forest area.6. inverted triangles) and net radiation (Rn. during (a) wet season and (b) dry season. the net radiation is much higher in the forest than in the pasture. sites. but with similar values during night time. being significantly higher during daytime over pasture. 1996). 38 . triangles) terrestrial radiation. Since it is mainly dependent on the surface temperature. The combined effect of higher reflectivity (albedo) and higher daytime long wave emission in the pasture cause a large difference between the net radiation over the two surfaces. mainly driven by a larger daytime loss. with open symbols. diamonds). but for incident (Lin. Top panels: average daily patterns of incident solar radiation (Sin. this reflects the effect of higher diurnal temperature variation observed in the pasture (Culf et al. although there is an indication that it is slightly higher in the pasture in the wet season. reflected solar radiation (Sout. during (c) wet and (d) dry season periods. on the other hand.
the absolute differences between pasture (P) and forest (F) measurements and the percentage.6 2.0 202. 2. the most pronounced changes are in the components related to the transfer from the atmosphere to the surface (incident radiation).20 0. Ln is the net longwave radiation. represent the relative effect of changing the surface from forest to pasture vegetation cover.6b.7 +4. this causes an average reduction of 13. defined as the difference between Lin and Lout.Table 2.13 0.7 Land use change effect Table 2.6a and 2. over the two sites.6 Sout 26. the incident long wave radiation. The seasonal differences in incident solar radiation may be caused by larger cloud cover during the wet season. is notably lower in the dry season than in the wet season 39 .9 Ln -36.3 Albedo 0. On the other hand. The absolute differences between pasture and forest measurements are represented by P-F.0 +0.6b). It can be seen that the most important change occurs in the reflected short wave radiation. and the percentage.5 Lout 448. The upward components (solar radiation reflected and terrestrial radiation emitted by the surface) have more or less similar values in the two season composites (fig.0 451.0 -1.3 -38.6 3. calculated by these differences divided by the measurements at the forest. over forest and pasture sites. To quantify the effect on the radiation components of changing the vegetation cover from rainforest to a cattle ranch.07 +57. Average values of radiation components in W m-2. That is. it is interesting to notice that the main differences are observed in the downward components. but probably lessened by the effect of burning activities that are very frequent during the dry season and cause strong smoke pollution over the whole region. Combined with an increase of 4.5 +55. calculated by these differences divided by the measurements at the forest.5 Lin 411. Comparing the different characteristics between the two seasons.2 -1.7 +0. are also presented.8 Rn 143. which increases by about 55 % when changing from forest to pasture.3 % in the net radiation.2 -19.2 presents the average values of all the radiation components in W m-2.2 124.8 -3.1 40.6 14.6 413. As we can see comparing fig.6a and 2.7 % in long wave radiation loss.2. 2. the high cloudiness during the wet season reduces the average incident solar radiation more than the smoke in the dry season.0 -13. largely affected by the humidity in the atmosphere. Sin Forest Pasture P–F (P-F)/F (%) 206.
Therefore the short wave balance. higher in the wet season (fig.(fig.6c and 2. At the pasture the monthly variation is not well correlated with soil moisture.12 0.18 0. probably because the overall albedo measured is a combination of the reflectance from both the grass and from the bare soil. as discussed by Culf et al. both in the pasture and in the forest. during 1999-2002. summarised as follows.14 0.22 0. Wright et al.7. 2. with the highest values occurring at the same time as the driest soil moisture conditions. Figure 2.6b. (1996) showed that the albedo there is related to the leaf area index. 2. The decrease that might be expected during the wet season appears to be offset by the higher albedo of new grass growth. Monthly averages of albedo over forest (closed circles) and pasture (open circles) during 1999-2002.16 0. (1995).6c and 2.6d). the net all wave radiation is higher in the dry season. Our data show similar results. Combining with data from other pasture areas. (1996) presented detailed discussions. and.7 shows monthly average values of albedo for data collected at Rebio Jaru and FNS sites. As noticed comparing fig. (1995) and Culf et al.24 0.6d). 2. the pasture albedo does not have a regular seasonal cycle. at both sites. higher in the dry season (fig.6a and 2. the long wave loss is slightly higher (more negative) in the dry season.10 1 2 3 4 5 6 7 8 9 10 11 12 forest pasture Month Figure 2. 2. Because of this. using data collected from 1991 to 1993. 0. even with the surface emitting more or less the same amount of terrestrial radiation in both seasons.20 Albedo 0. Concerning the seasonal variations and the differences in albedo between the sites Culf et al. 40 .6b) competes with the long wave balance. therefore more influenced by the short wave component. The seasonal variation at the forest site is evident. during ABRACOS project.6a and 2.
together with the curves of net radiation. the variability of net radiation is mainly driven by the short wave balance. 2. the soil heat flux (G) is also included.To better visualise the seasonal variations on the net radiation budget.4.8 presents monthly averages of the net shortwave balance (Sn = Sin – Sout). 2. fig. In this context our results are not in disagreement.4. In the case of the pasture. concluded that the long-wave balance was a more important determinant of seasonal variations of the net radiation. using a slightly different analysis for a forest area in central Amazonia. Sensible and latent heat fluxes In a similar way to the wet and dry season composites for the radiation components. For the energy balance in the forest area. That way. Malhi et al. as it was previously shown that there was little variation in the outgoing short wave. than the albedo. It is noted that. instead of one curve with the average values. limited by the two methods of energy balance closure forcing: (i) replacing the λE measurements by the residual of the energy balance. 2. although the long-wave components play a role. or (ii) adjusting both H and λE fluxes to maintain the measured Bowen ratio. For both the sensible and latent heat fluxes. (2002). as measured by 4 flux plates buried close to the surface (1 cm). a range of curves are presented. the average daily patterns of sensible and latent heat fluxes over forest and pasture were calculated and are presented in fig. they highlighted that the surface reflection (albedo) is not as important to the seasonal variability of net radiation as the long wave balance. net long wave balance (Ln = Lin – Lout). However they analysed the variations by normalising the components by the incident solar radiation. This “storage term” was estimated by the parameterisation according to changes in air temperature and humidity changes inside the canopy proposed by Moore and Fisch (1986). Analysing the impact of the dry season at both sites 41 .9. the change in energy storage in the canopy air space and biomass is significant and these fluxes were added. and net all-wave radiation (Rn).
but relative changes in the energy partition are hardly seen in this figure.9b and 2. with deep roots. (2002) showed that. p. over forest (closed symbols) and pasture (open symbols). (c) and (d)) it is clearly seen that very little change is observed in the forest while larger changes occur in the pasture. In the pasture. on the other hand. the sensible heat flux is largely increased. the fluxes are slightly higher during the dry season. with the bottom ones. can maintain a large uptake of soil water even after a long dry period. Von Randow et al. longwave radiation (Ln = Lin – Lout. Monthly averages of radiation balances of shortwave radiation (Sn = Sin – Sout. In the forest.200 Net all wave and net solar radiation (W m ) -2 80 60 40 20 Net shortwave (Sn) Net all-wave (Rn) Net long wave (Ln) 160 140 120 100 80 60 1 2 3 4 5 6 7 8 9 10 11 12 Month 0 -20 -40 -60 Figure 2. in the presence of strong convection activity and precipitating clouds. 2. as seen by the soil water storage records (fig.5). while the evapotranspiration rates are reduced. due to a small increase in the net radiation. These differences are explained by the fact that the forest trees. inverted triangles). These "disturbed" boundary layers have qualitatively different characteristics than are observed in undisturbed boundary layers. squares) and net all-wave radiation (Rn. it is also interesting to note that the range of flux uncertainty in the pasture is higher during the wet season. This may explain the higher uncertainty in the flux calculations during the wet season. the boundary layer in the tropics presents complex interscale links. during the wet season. 2. As discussed by Garstang and Fitzjarrald (1999. 42 Net long wave radiation (W m-2) 180 .8. (comparing the top panels.9d. 285-287). circles). the boundary layer over the region is indeed largely influenced by such processes. (a) and (b). and the occurrence of strong updrafts and outflows make a large contribution to the exchange processes in the surface-atmosphere interface. Comparing fig.
5 83.1 H 38.1 10.9 . sensible and latent heat fluxes (H and λE respectively) and soil heat fluxes (+ heat storage in canopy at forest).2.5 H 31.1 128.6 -7. for: (a) wet season at forest.Figure 2.77 0. Table 2.7 .8 + 28.41.64 -0.6 63.4.13 -16. Wet season Rn Forest Pasture P-F (P-F)/F (%) 136.3 Land use change effect 43 .9 + 220.127.116.11 45.74 0. As in Table 2. Average daily patterns of net radiation (Rn).5 13.56 -0.5 λE/Rn 0.1).9 -44.2 λE/Rn 0. (b) wet season at pasture. but for average values of sensible and latent heat fluxes calculated during wet and dry season periods.3 49.0 -21.23. (c) dry season at forest and (d) dry season at pasture.5.1 λE 108.0 -33.9 Rn 146.18 -24.2 Dry Season λE 104.9.5 . The evaporative fraction (λE/Rn) is also shown. Dashed areas represent the ranges of heat fluxes calculated using two different procedures (see section 2.5 .9 113.
The top curves (triangles facing up) are the values calculated from the heat fluxes measured directly by the eddy covariance system. In the pasture.10. 44 . at (a) forest and (b) pasture.1 1 2 measured λE as residual average 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Month Figure 2. To better visualise these differences on energy partition over the two vegetation covers and beween the seasons.5 0.7 0. the averages and the differences between these averages of the sensible and latent heat fluxes are presented in Table 2. are the averages of the fluxes adjusted for energy balance closure maintaining the measured Bowen ratio.10 presents the evolution of the monthly variation of the Bowen ratio at both sites.3. fig. 2. presented in W m-2.4.4 0.From the separate wet and dry season composites. Monthly averages of Bowen ratio (β = H/λE) as measured from the eddy covariance system (triangles). and the average between the two cited methods (circles). and the middle curves (circles) represent the average between the two. As expected. large differences between the two types of surface are noticed. the bottom curves (triangles facing down) are the Bowen ratios obtained after calculating the latent heat fluxes as the residual of the energy balance.8 Forest measured λE as residual average (a) Pasture (b) Bowen ratio (H/λE) 0.10a). The Bowen ratio varies little over the year in the forest.3 to 0. This difference is increased to 24 % during the dry season. In the 0.6 0.9 0. The values. calculated after estimating λE by the residual of the energy balance (inverted triangles). The differences between the two sites obtained by adjusting λE as the residual are of similar amounts.3 0. although a slight increase can be seen at the end of the dry season (fig. with values ranging from 0. 2. In the wet season the evaporative fraction (λE/Rn) in the pasture is 17 % lower than in the forest. the sensible heat fluxes are 28 – 45 % higher while the evapotranspiration rates are 20 – 41 % lower.2 0.
1 m.s-1.4.6 in the wet season to 0.s-1. 1998. 2.2 ) where Fc is the flux of CO2 measured by the eddy covariance and the second term is the storage of CO2 (Mc is the molar weight of carbon and dC/dt is the change in CO2 concentration between several heights). Grace et al.pasture. To estimate this flux after this period the following procedure was employed: using the 14 months of data of concentration profiles available we first calculated the average storage flux at each time of day separating the data into three classes depending on the friction velocity (u*) averaged on the night before (or current. the storage term is calculated by approximating the derivatives as finite differences between two successive measurements and the integrals by weighted sums of the variables at the 6 levels. for nighttime data) night. for nighttime data) of the measurement. according to the amount of turbulence observed during the preceding (or current. These stratified storage values were then used as a lookup table to substitute the storage when no data were available. (ii) 0. These classes are when average night-time u* ranges between (i) 0 and 0. 2.1 and 0.3 – 0. Net Ecosystem Exchange of CO2 The Net Ecosystem Exchange (NEE) of carbon dioxide between the forest and the atmosphere can be estimated by combining the flux measurements from the eddy covariance with profile measurements of CO2 concentration below the flux sensors.6 – 0. again highlighting the effect of water stress during the dry season. a large seasonality is observed (fig. with the Bowen ratio changing from 0.s-1 and (iii) higher than 0. The quality of this empirical CO2 45 . 1995) NEE = Fc + ∫ M c dC dz dt ( 2. After June 2000 the profile system broke down and no direct measurements of the flux due to change in storage were then available.5.2 m. In practice.8.2 m.10b). The NEE is calculated as the sum of the fluxes measured at the top of the tower and the change in storage of CO2 in the layer below (Lee.
4. The results are shown in fig. Net ecosystem exchange estimated using the empirical CO2 storage model (see Section 2. Miller et al. the results were considered satisfactory for the purpose of this paper. 2004. such as an inability to measure CO2 advection during calm nights. 2002. in W m-2) 20 0 -20 -40 -40 -20 0 Measured NEE (W m ) -2 20 40 Figure 2. This is due to the fact the NEE is a relatively small difference between two large quantities: the uptake of CO2 due to photosynthesis during daytime..5) plotted against measurements (calculated with measured CO2 concentration profiles).11. referred to as the Gross Primary Production (GPP) and the release by the ecosystem respiration (Reco). Kruijt et al. This way. several researchers plot nighttime NEE measurements against u* and filter out the data when a 46 . may create large errors in the annual budget. for 10 days in 1999. Although the model is very simple. The 1:1 line is also shown. 2004). In an attempt to account for this possible night time loss. 2. No other corrections were applied to the NEE estimates.11.40 NEE (calculated from modeled storage. despite several recent papers indicating that the interpretation of nocturnal measurements is the largest single source of uncertainty in the absolute accuracy of daily or annual totals of carbon exchange in the tropical forests (Araujo et al. storage model was tested by withholding data during 10 days that presented a range of different weather conditions and comparing the NEE calculated using the empirical model against the measurements.. a small day to night bias on NEE measurements.
using the data collected during the first year of measurements (see fig. 2004). Net ecosystem exchange of CO2 (NEE) averaged according to classes of Photosynthetically Active Radiation (PAR). for wet and dry season periods. whereas in the dry season fluxes are strongly reduced after the PAR becomes higher than 1000 μmol m-2 s-1.reduction of respiration is seen during low u* conditions (Miller et al. 2004).12 shows the NEE values averaged according to classes of incident Photosynthetically Active Radiation (PAR). and the light response curve is not linear. Kruijt et al. 9d of Kruijt et al. no further corrections were applied to the data. 2004). A small reduction in the initial slope and in the maximum assimilation is observed in the dry season curve. This reflects the fact that the carbon 10 10 Forest NEE (μmol CO2 m-2 s-1) 0 (a) Pasture NEE (μmol CO2 m s ) -1 -2 (b) Dry season Wet season 0 -10 -10 -20 -20 Dry season Wet season -30 0 500 1000 1500 -2 -1 -30 0 500 1000 1500 -2 -1 2000 2500 2000 2500 PAR (μmol m s ) PAR (μmol m s ) Figure 2. for data collected during dry and wet seasons at the forest (a) and pasture (b). 2002. Figure 2. In the wet season.. the typical near-linear light response of the C4 grasses is evident. Goulden et al. saturating at a PAR of about 1000 μmol m-2 s-1. however. 2004 showed. 47 .12a) and at FNS (fig.. and since our aim here is to concentrate on the relative differences in CO2 exchange among the sites (and the seasonal patterns). with average saturation values of –10 μmol m-2 s-1. while for the dry season it reaches about –20 μmol m-2 s-1.12.12b). that no clear reduction was seen at the Rebio Jaru site. For the pasture site. The shape of these curves is very similar to that reported for other forest areas in Amazonia (Malhi et al. Carswell et al. for both periods. 2. The NEE curve for the wet season period saturates at about –23 μmol m-2 s-1. For this reason. 2. measured at Rebio Jaru (fig. the differences between the wet season and the dry season light responses are much bigger. Figure 2. 1998.12a shows that the ecosystem light responses at the forest show the typical behaviour of initial strong decrease (increase of uptake) with PAR.
There is also a noticeable reduction in the fluxes at both sites during the dry season. Daily patterns of NEE for both sites. Understanding what controls the seasonal differences across sites is an essential component of 48 .13. with negative fluxes during the day (photosynthesis activity higher than respiration) and positive fluxes during the night (respiration activity only). These large differences on the seasonal variability observed at the two sites are mainly explained by the ability of the forest to avoid severe drought stress by extracting the water from deep layers in the soil (fig. The latter pattern appears to be driven by a strong decrease in respiration during the dry season. Figure 2.13 presents the average daily patterns of CO2 fluxes over the two sites. during wet (a) and dry (b) seasons. Although there is a small reduction in the respiration. 2..5). without a comparable reduction in photosynthetic activity. while near Santarém (Goulden et al. a bigger effect on the daytime fluxes causes a reduction in the NEE (less uptake) in the dry season. during wet and dry season periods. At two sites. 2004) a higher uptake was measured during the dry season. It is interesting to compare this pattern at Rebio Jaru with other sites in the Amazon Basin. 2003). Vourlitis et al..15 10 Wet season Dry season NEE (μmol m s ) -1 5 0 -5 -10 -15 -20 -25 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 24 -2 forest pasture Local time (hours) Figure 2. (2004) also found for a transitional forest (ecotonal between rainforest and savanna) that the most negative NEE occurred during the rainy season. The fluxes present the typical patterns of vegetated areas. near Manaus (Araujo et al. little seasonal variation was observed. assimilation by the pasture vegetation is strongly affected by a reduction in soil moisture content. 2002) and Caxiuana (Carswell et al.
if that were the case. Those authors used measurements collected during 11 days in the dry season of 1992 and 44 days in the wet season of 1993 to fit a process-based model. 2. 2004. However. but the uptake rates implied from 1999-2002 data are still very high and considered implausible by many ecologists.13 that the daytime fluxes are very similar at both the forest and pasture sites during the afternoon. McWilliam et al. (1995). the afternoon decline in photosynthesis is probably caused by forest stomatal closure responding to high vapour pressure deficit. averaging 6 to 9 μmol m-2 s-1. 2004). additionally.5 mol C m-2 year-1 (that would represent an average of about 2.8 kg C ha-1 day-1). but a circadian rhythm may also play a role. It is also interesting to notice in fig. 8. 1996.LBA. It is interesting to compare our measurements with previous measurements at the Rebio Jaru site shown by Grace et al. for a forest site in eastern Amazonia. the authors also considered the possibility of an internal circadian rhythm controlling stomatal closure of some species. we would expect the pasture site to present a similar response. (2004) showed. (1996) did not find a correlation between the stomatal conductance and light intensity (although they argue that this may be due to poor sample size). innate phenological controls may play an important role in the regulation of seasonal carbon uptake (Keller et al. This decline in afternoon fluxes was also reported for other forest sites in Amazonia (Malhi et al.). during a "friagem" event. Goulden et al. Measuring the stomatal conductance of several tree species in Rebio Jaru during ABRACOS. The values of daily totals measured during 1999-2002 shown in Table 2. on average. 1998. and indeed. that this afternoon decline is not correlated to the soil water content. They then used the model to estimate that the forest absorbed. 2004). Since there 49 . Both datasets indicate little evidence of nighttime losses during calm nights (Grace et al. but lower (higher uptake) in the forest in the morning.. Kruijt et al. Therefore. we notice that the respiration rates are much higher in the forest. while at the forest the fluxes are clearly different in the morning and afternoon hours.. The average differences between the NEE measured at the forest and pasture are presented in Table 2. but they found a good correlation with the vapour pressure deficit.4 suggest rates 4 to 6 times bigger. Access to deep soil water may vary across sites and. In the pasture the daytime evolution is more or less symmetric around the negative peak at noon. while at the pasture they reach only 3 to 5 μmol m-2 s-1 throughout the night. Comparing nighttime fluxes. Goulden et al.4. air temperature or leaf water potential.
therefore indicate a decrease in the productivity.3 Pasture -11.0 -13.T.2 +28. peak photosynthesis activity.4 8.2 -13. As the reduction in the nocturnal respiration is higher than the reduction in the daytime uptake.5 7. therefore these results should be viewed with caution.Table 2. it is still not clear why the forest at Rebio Jaru seems to absorb that much carbon from the atmosphere.9 * -44. Wet Season μmol m-2 s-1 Daytime Nighttime Daytime Peak Dry Season μmol m-2 s-1 Daily total Daytime (kg C ha-1 day-1) Nighttime Daytime Peak Daily total (kg C ha-1 day-1) Forest -14.0 -4.1 -3. with averages 44 % lower. During the wet season. even a very small slope in the terrain may cause a significant drainage loss. these differences increase: 28 % less photosynthesis and 57 % less respiration are observed in the pasture compared to the forest. The nighttime respiration is also reduced compared to the forest.5 -8. the daytime averages.7 * -67.3 -5.2 4. 50 . since the growth of the vegetation is constantly renewed. especially during the dry season.8 -7.2 -21.6 5. calculated with data collected between 08:00 and 18:00 L.3 * -57.1 4. This high uptake in the pasture site is not surprising.3 -10. and.2 -3. Average values of Net Ecosystem Exchange measured over the two sites during daytime (08:00 – 18:00 h).6 3. are 22 % higher (less negative) in the pasture. under these situations. one other important factor that may add a large uncertainty on these numbers is the effect of drainage of CO2 during stable conditions that may not be accounted for. Comparing the fluxes at the forest and pasture. nighttime (19:00 – 05:00 h). are no other physiological measurements available at the site to support the eddy covariance measurements.5 * Note that this increase is related to negative values (uptake by photosynthesis).4 +24.5 -17. large differences are noticed. During the dry season.4 -18. a very stable layer is frequently observed at night.4. and daily totals. compared to the forest. In the pasture site.8 Land use change effect P-F 3. while the cattle remove the biomass.5 +23.1 -22..6 (P-F)/F (%) +21.3 * -19.5 3. As discussed before. The higher daytime fluxes actually indicate less photosynthesis activity at the pasture.1 -17. the combined effect is a 19 – 67 % higher daily uptake of CO2 in the pasture.
the soil water storage changes more rapidly in this layer . large differences are observed in precipitation. Changes in soil moisture storage profiles give indications of the uptake of water by the vegetation and the drainage at the two sites. mainly caused by drainage. we adjusted the turbulent fluxes in two ways: calculating the latent heat flux as the residual of the energy balance or adjusting both the sensible and latent heat flux to maintain the Bowen ratio as measured by the eddy covariance system. The reasons for the apparent underestimation of the turbulent fluxes are still unclear and may be related to two factors: (i) slow wind direction changes on undulating terrain in the region. compared to the measurements during wet season (December to March) at both sites. on the other hand.the seasonal change was about 290 mm in the forest and 110 mm in the pasture. The pasture vegetation withdraws water only from the upper layers of the soil with the water stored in the layer from 2 to 3.4 m deep showing only little variation. Comparing the two sites.5. adding a significant low frequency component that we are not able to capture using short time scale rotations. Comparisons are made between the measurements over the two sites and between wet and dry season periods. but still not always achieved. To concentrate on the comparative measurements over the two contrasting types of vegetation cover. These large variations in the forest are partly caused 51 . the specific humidity and precipitation amounts are greatly reduced during the dry season periods (June to September). specific humidity and specific humidity deficit. (ii) horizontal flux divergences that are simply not possible to estimate from measurements made on a single tower. In the forest. At the pasture the energy balance closure is better. While the temperatures present little variation between the seasons. The energy balance closure at the forest site is poor: the sum of the turbulent fluxes reaches only about 74 % of the available energy. Conclusions In this work we present the data of radiation flux components and turbulent fluxes of energy and CO2 collected almost continuously from February 1999 to September 2002 in two different sites in south western Amazonia: one in a forest reserve (Rebio Jaru) and one in a pasture cattle ranch (FNS).2.
Seasonal changes at both sites are observed mainly in the incident radiation (short and long wave). For the pasture site. Differences in daytime NEE between the two sites are then larger in the dry season: daytime productivity is about 28 % lower in the pasture in the dry season. In the dry season. A small reduction in the initial slope and in the maximum assimilation was observed in the dry season curve.3 % in the net radiation in the pasture. driving large seasonal variations in the net radiation. The nighttime 52 . Analysing the seasonal variations we observed that the Bowen ratio is relatively constant over the forest. In the wet season the evaporative fraction (λE/Rn) at the pasture is 17 % lower than at the forest. with average saturation NEE of -10 μmol m-2 s-1. In the wet season. with the Bowen ratio changing from 0.3 – 0. the typical near-linear light response of the C4 grasses is observed. this causes an average reduction of 13. The NEE at the forest show the typical behaviour of initial strong decrease (increase of CO2 uptake) with increasing PAR. Combined with an increase of 4. whereas in the dry season fluxes are strongly reduced after the PAR becomes higher than 1000 μmol m-2 s-1. Large differences between the two types of surface are also noticed in the energy partition between sensible and latent heat fluxes. Although the long-wave components play a role. although a slight increase was observed in the end of the dry season. The most important changes occur in the reflected short wave radiation. on the other hand. which increases about 55 % when changing from forest to pasture. saturating at -20 to 23 µmol m-2 s-1 at a PAR of about 1000 μmol m-2 s-1.4. but also give an indication of the ability of forest vegetation to uptake water from deep layers in the soil. The radiation flux components are markedly different between the two sites.by lateral drainage. Light response curves of the net ecosystem exchange of CO2 were analysed with data from the two sites for wet and dry seasons. the differences between the wet season and the dry season light responses are much bigger. while the evapotranspiration rates are 20 % lower in the pasture. the variability of net radiation is shown to be mainly driven by the short wave balance.7 % on long wave radiation loss.3 – 0. compared to the forest.6 in the wet season to 0. while this difference is about 20 % in the wet season. a large variation was observed.6 – 0. compared to the forest. varying between 0.8 in the dry. the differences are lower in the sensible heat (fluxes are 28 % higher in the pasture). while the changes in evapotranspiration are large (rates are 41 % lower in the pasture). At the pasture site. In the wet season the sensible heat fluxes are 45 % higher. This difference increases to 24 % during the dry season.
respectively. As the reduction in the nocturnal respiration is higher than the reduction in the daytime uptake. the combined effect is a 19 – 67 % higher daily uptake of CO2 in the pasture. with averages 44 % and 57 % lower in the wet and dry seasons. This high uptake in the pasture site is not surprising. compared to the forest. since the growth of the vegetation is constantly renewed.time respiration in the pasture is also reduced compared to the forest. 53 . as the cattle remove the biomass.
____________________ This chapter is published as C. A. latent heat and CO2 flux contributions occurred on 'turbulent' length scales. 107(0). Based on the characteristics of the scale dependence of the scalar fluxes. to have an insight to explain the origin of the variability of the scalar fields close to Amazonian forest. 2002. Mesoscale eddy motions. Arlino and B.D. von Randow.S. what suggests that different boundary layer moisture regimes occur during dry and wet season. P. So.D. Diurnal Conditions. The results have shown that a two-categories classification is the more appropriate to describe the kind of observed fluctuations: 'turbulent' and 'mesoscale' contributions.A. Sá. Daubechies-8 orthogonal wavelet is used to scale project turbulent signals in order to provide scale variance and covariance estimations. The largest amount of the sensible heat. Analysis of scale correlation coefficient (rTvq) between virtual temperature (Tv) and humidity (q) signals show that the scale pattern of Tv and q variability are not similar. G. 8062. 55 .Chapter 3 Scale variability of atmospheric surface layer fluxes of energy and carbon over a tropical rain forest in South-West Amazonia. Kruijt. however. Manzi. Prasad. Scale humidity skewness calculations are negative during dry season and positive during wet season. I.S. This permits to study some of the statistical characteristics of the scalar turbulent fields in each one of these classes and thus. doi:10. Journal of Geophysical Research. Abstract The aim of this study is to investigate the low-frequency characteristics of diurnal turbulent scalar spectra and cospectra near the Amazonian rain forest during wet and dry season.1029/2001JD000379. Scale variability of atmospheric surface layer fluxes of energy and carbon over a tropical rain forest in South-West Amazonia. This is because the available turbulent data are often nonstationary and there is no clear spectral gap to separate data in 'mean' and 'turbulent' parts. can contribute up to 30 % to the total covariances under weak wind conditions. L. some classification criteria of this scale dependence are investigated. and rTvq < 1 for all analyzed scales. the total scalar covariance of each 4 hours data run was partitioned in categories of scale covariance contributions.
2000).1).1. Mahrt. although their importance are evident. as discussed by (Shuttleworth et al. Vickers and Mahrt. On this work we will discuss the second aspect of the problem. 1999). such as w (vertical wind velocity) sampled at one hour scale do not show a clear spectral gap (see figure 3. Introduction The discussions about the accuracy on turbulent flux estimations using the eddycorrelation method are many times forgotten in scientific works. on their classical study about atmospheric turbulence structure. 1998). This imprecision on flux estimations appears when the power spectra of turbulent variables. Wyngaard. 1964. p. among others. It is evident that turbulent signals under such conditions led to difficulties on defining correct averaging periods and fluctuations (Hildebrand. in the context of disturbed surface layers (McNaughton and Laubach. 1998). 1997. Mahrt. or under transient conditions (Mahrt. which can be aggravated when measurements are made under certain peculiar conditions. 1991. This problem was discussed by several authors: Lumley and Panofsky (1964. 2000) or disturbed wet tropical boundary layers (Garstang and Fitzjarrald. 2000).3. 56 . identified scenarios where the autocorrelation function does not tend to zero with time increasing (or length) in such a way that the determination of integral scales sometimes becomes a particularly difficult task. 1984. McNaughton and Laubach. 1992. It is a complex problem and contains several uncertainties associated with the nature of turbulence itself (Lumley and Panofsky. by means of investigating questions related to the measurement of turbulent fluxes under nonstationary conditions or over non-homogeneous surfaces and the dependence of flux calculations on the choice of cutoff frequency (or averaging period) used on these calculations. and the second comes from the way in which eddycorrelation method is used. Mahrt. 1998. 45). McNaughton and Laubach. Two main sources of imprecision on turbulent fluxes estimations could be stressed: the first one is result of all the errors associated with the measurements themselves. such as on towers or aircrafts.. 1991a. over complex surfaces.
Two major objections could be raised against the current statement concerning the validity of the stationarity and horizontal homogeneity hypotheses on atmospheric boundary layer (ABL) turbulent fields. the existence of complex distributions of sources or sinks of scalars at the surface (Brutsaert. the effects originated from the top of the ABL (Mahrt. 2000). Fourier power spectra of vertical wind velocity component (w. turbulent variables statistically calculated are strongly dependent on sampling period and often their variances do not show a stable mean value along the data sampling. such as mesoscale ones.1 0. Consequently. whose existence is connected to perturbations introduced by low-frequency motions (McNaughton and Laubach. 1991).1 1 10 log(f) in Hz Figure 3. The second complication is associated with the factors regarding the variability imposed to the flow by external forcing such as the roughness of the terrain (McNaughton and Laubacch.01 vertical velocity (w) temperature (T) 0. measured from 10:00 to 12:00 h (local time). over Rebio Jaru forest. 1998).1. solid line) and virtual temperature (Tv. on day 98.1 f Sαα(f) 0. 2000).001 0. The first one results from the fact that the power spectra of turbulent fluctuations extend to larger scales of motion. 57 .001 0.01 0. dotted line) at arbitrary units. etc.
Trying to deal with the determination of scale dependent fluxes, Sun et al (1996), based on the characteristics of this dependence, proposed to partition the total flux into turbulent, large-eddy, and mesoscale fluxes due to motions on scales smaller than 1 km, between 1 and 5 km, and bigger than 5 km, respectively. In order to explain these characteristics of flux-scale dependence, McNaughton and Laubach (2000) have proposed a different three class scheme based on the earlier ideas of Kader and Yaglom (1990) and their attempt to find a generalization of the classical Monin-Obukhov Similarity Theory (MOST). After McNaughton and Laubach (2000), three different scaling regimes could act on the surface layer turbulent exchanges: an inner-layer scaling (ILS), an outer-layer scaling (OLS) and a combined scaling (CS). Each one of these regimes would explain the characteristics of turbulent power spectra and cospectra on specific regions and their classification depends on height and frequency. They have associated OLS components with "inactive" turbulence and ILS components with "active" turbulence in the sense proposed by Townsend's hypothesis about turbulence atmospheric structure (Townsend, 1976). Still after McNaughton and Laubach, if the power spectra of active and inactive components of turbulence were separated by a spectral gap, there would be no interactions between them. But, on the other hand, if there is no obvious spectral gaps, there would be a matching region with -1 and no more -5/3 power law (for u, v, and scalar variables, but not for w-wind velocity component) corresponding to the CS region. Still with respect to flux scale dependence issue, Williams et al. (1996) in their study about flux airborne measures in the intertropical convergence zone ABL have presented detectable basic structural differences between the eddies containing the high and low wavenumber fluxes. After them, the turbulent characteristics associated with the higher wavenumbers show more coherent and repeatable behavior than the ones associated with low wavenumbers. This because the low-frequency component to the total flux is not expected to be controlled by only the atmospheric boundary layer similarity parameters, but also by phenomena which are generated outside of the boundary-layer environment. Each one of these classes should have a specific parameterization to represent the subgrid fluxes on models of simulations of ABL, what is particularly difficult for large eddies and mesoscale motions situations. Other class-separation criteria have been proposed, still. Howell and Mahrt (1994) also studied the scale dependence of surface fluxes using the Haar wavelet transform to decompose the turbulent signals in 4 classes, including the three cited before, and the 'fine scale' class, corresponding to very small scale motions, with nearly isotropic characteristics. 58
The classes established by Sun et al. (1996) and by Williams et al. (1996) refer to oceanic tropical boundary layer that in some specific situations, in the vicinity of large cloud cluster systems, do not show a clear gap in the power spectra (Williams et al, 1996) and present many affinities with the ABL over Amazon rain forest on wet-season (Garstang and Fitzjarrald, 1999). According to Garstang and Fitzjarrald, as the tropical marine boundary layer, the tropical wet season ABL could be characterized as a disturbed state of the boundary layer, which presents peculiar inter-scale links properties. It has qualitatively different characteristics than that observed on undisturbed boundary layers and exhibits peculiar tropospheric phenomena, such as outflows, among others, which play important roles on vertical transports and on defining vertical exchange processes time-scales. Under these conditions, is hard to characterize a superior border for the ABL, whose thermodynamic characteristics are strictly associated to the nature of the convection on the humid troposphere. It is important to mention that, although not hardly studied, the exchanges of heat and moisture on the top of ABL also appear to influence the variability of turbulent scalar variables measured at surface (Mahrt, 1991; Wiliams et al., 1996; Garstang and Fitzjarrald, 1999). Mahrt (1991) showed that, depending on the ABL moisture regime, different characteristic scales of potential virtual temperature and humidity are present on some close to surface turbulent processes. Even disregarding the top of ABL influence, at least four physically distinct factors may contribute to the variability of measurements at surface: (i) local circulation induced by horizontal heterogeneity of the terrain (Mahrt et al., 1994, Mahrt et al., 1998), which is expected to exist in Rebio Jaru site (this study) due the existence of "fish-bone pattern" strips of alternating forested/deforested areas surrounding the reserve and also due to the wavy pattern of the vegetated crown top; (ii) contribution from updrafts and downdrafts to surface fluxes, associated with the very strong convective activity present in Amazonia (Garstang and Fitzjarrald, 1999); (iii) contributions due to the existence of coherent structures on turbulent flow, what presents specific characteristics over vegetated covers, particularly with respect to scalar fluxes (Gao and Li, 1993; Collineau and Brunet, 1993; Brunet and Irvine, 2000); (iv) the effects of slow wind direction and wind strength variation on scalar transport processes near the ground (McNaughton and Laubach, 2000). Such last variations are identified with the concept of "inactive" turbulence (Townsend, 1976).
These factors can drive important contributions to the total fluxes by low frequency eddy motions, as reported by Sakai (2000). The author found for a summer-time signal, obtained above a midlatitude deciduous forest in Canada, that large eddies presenting periods ranging from 4 to 30 minutes might contribute with about 17% to total surface fluxes of heat, water vapor and CO2. However, it is likely that these flux fractions would not be taken into account if they were calculated by the current popular averaging-periods procedures. This is because these usual sampling time intervals are too short to resolve the larger eddies present in the flow (Mahrt, 1998). Sakai (2000) has also found that short-averaging periods might underestimate daytime CO2 fluxes at standard towers by 10-40 %, depending on wind speed conditions. In this work we analyze turbulent data measured at Amazon rain forest during LBA 1999 wet-season campaign and during 2000 dry season period. The measuring heights, 62 m and 67 m, are likely to be in a surface transition sub-layer. During daytime conditions, lowfrequency contributions to eddy-correlation turbulent fluxes were often present. We apply wavelet transform to project the data on scales and calculate scale covariances in a similar way to the ones used by Howell and Mahrt (1994) and Katul and Parlange (1994). To accomplish such calculations, we use the Daubechies-8 wavelet transform (Daubechies, 1992) to decompose turbulent signals of vertical wind velocity (w), virtual temperature (Tv), and humidity (q) and CO2 concentration (c). In spite of some restrictions concerning the feasibility of its applications (Treviño and Andreas, 1996), wavelet analysis is a powerful tool to analyze turbulent signals (Farge, 1992; Katul and Parlange, 1994). Using the scale-projected signals, we then identify the more adequate way to partition the fluxes in the Amazonian atmospheric boundary layer.
3.2. Site description and deployed instruments
During the 1999 Wet Season in Amazonia, several activities of the LBA Project (Large Scale Biosphere-Atmosphere Experiment in Amazonia) took place at the biological reserve of Jaru (Rebio Jaru), located about 100 km North of Ji-Paraná, Rondonia, Brazil 60
(2001) give details of vegetation at this site. 2001). Andreae et al. during WETAMC campaign. within the collaboration of two subprojects: LBA/WETAMC (first LBA major wet season Atmospheric Mesoscale Campaign. 8th) of 1999.L. No humidity or CO2 concentration measurements were available during this period of dataset B. and is part of long term measurements of surface fluxes supported by the LBA/EUSTACH project. 61o 56. Data quality control was done using the QC pack software of Vickers and Mahrt (1997). The sonic anemometers measure the three wind velocity components (u.. 31st) of 2000. where the Ji-Paraná River forms the western boundary of the reserve. An extra mast of 7 m was built later at the top of this tower.706' S. Andreae et al.027' W.7 m. The first one (dataset A) is composed by measurements made between days 93 (Apr. Rebio Jaru is a terra firme forest reserve owned by the Brazilian Environmental Protection Agency (IBAMA). measuring at a sampling rate of 16 Hz.. a new 60 m tall micrometeorological tower was built at this site (10o 4. Gill Instruments). The Rebio Jaru reserve canopy has a mean height of 35m.S. together with an infrared gas analyzer (LI-6262. However. These sensors were installed approximately 20 cm apart from each other. and between days 234 (Aug. At the end of 1998. however.). in the other directions. LICOR Inc. w) and virtual air temperature (Tv). both recording data at a sampling rate of 10. with a 3-D sonic anemometer (Solent A1012R. following a several kilometers wide very peculiar fish-bone pattern.4 Hz. the fetch condition is mainly of an undisturbed forest for tens of kilometers. it is much lesser (about 800 m). This data were obtained at the height of 62. 1st) of 1999. Records with bad data points. some of the higher tree branches have heights up to 45m. 2001) and LBA/EUSTACH (European Studies on Trace gases and Atmospheric Chemistry.). which alternates patches of forest and of degraded land are observed. In the predominant wind direction (sector from the north west clockwise to south-south east). v. The second dataset (B) is composed by turbulent data collected from day 31 (Jan. 2001). instrument 61 . On the other side of the river the rain forest has been progressively cleared during the last 25 years.. Two datasets were used on this work. 3rd) and 98 (Apr. Campbell Scientific Inc. 21st) and 244 (Aug. at the height of 145 m A. Silva Dias et al.) installed at the height of 67 m above the forest floor. with a different 3-D sonic anemometer (CSAT3.(Silva Dias et al. and the IRGA measures the concentration of water vapor (q) and CO2 (c) in the air. 31st) to 60 (Mar.
dropouts, poor resolution and abrupt changes are flagged and then examined visually. Bad data records were not used in the analysis.
3.3. Scale variability analysis theoretical elements
3.3.1. Determination of power spectra and cospectra low frequency ends of the turbulent variables
The Fourier energy spectrum has been one of the most familiar techniques for analysis of signals. Indeed, many of the traditional methods work in the Fourier space for most of the time. The Fourier energy spectrum E(k) of the real function f(x) is defined by E(k) = | f*(k) |2 for k ≥ 0 ( 3.1 )
where f*(k) signifies Fourier transform, given by
f *(k )= ∫
f ( x ) e −ikx dx
( 3.2 )
The so-called “spectral gap” usually appears on power spectra of turbulent variables measured at middle and high latitude sites over uniform terrain and provides the necessary information to a good choice of the sampling segment size (time of duration of measurements to flux determination). This also supplies information on the temporal scale where the variables should be decomposed into mean and fluctuation parts. Although is broadly accepted that the spectral gap exists, at least on a statistical sense, on temporal scales close to 1 hour (Lumley and Panofsky, 1964; Stull, 1988), there are situations when is hard to observe a clear gap, on intervals up to hundred of minutes (Sun et al., 1996; Williams et al., 1996; Mahrt, 1998; Mc Naughton and Laubach, 2000). Most of the power spectra observed over Rebio Jaru with data measured during morning and afternoon times show variance spread over a wide range of frequencies, without 62
any obvious spectral gaps, even when the sampling sizes were increased to several hundreds of minutes. Figure 3.1 displays the power spectral density of vertical wind velocity component and virtual temperature signals measured from 10:00 to 12:00 h (local time) on day 98. This difficulty in determining a clear cutoff frequency were observed both in wet and dry season periods and motivated the authors to use the methodology based on wavelet transforms, a mathematical tool which allows spectral analyses of nonstationary data, to analyze the scale variability of surface fluxes over this site.
3.3.2. Wavelet transforms
The wavelet transform (WT) is a powerful mathematical analysis tool, which permits an evolutionary spectral study of turbulent atmospheric signals (Daubechies, 1992; Farge, 1992). WT is similar to, but an extension of Fourier analysis. WT is computationally similar in principle to Fast Fourier Transform (FFT). The FFT uses cosines, sines and exponentials to represent a signal, and is most useful for representing stationary functions. Since many 1-D and 2-D signals display nonlinear, chaotic, intermittent or fractal behavior, Fourier analysis is less suitable for analyzing such signals. Wavelets offer a more adequate method to analyze complex signals as they decompose such signals into contributions of different scales as well as different locations. The following material is also discussed by Katul and Parlange (1994) and the main points are presented here for completeness. WT is classified under two broad categories: (1) continuous WT, and (2) discrete WT. Daubechies (1992, p. 7) further classifies the discrete WT as (1) redundant discrete systems (also known as frames) and (2) orthonormal wavelet expansions. For analysis of turbulence measurements, discrete orthonormal WT is preferable since it is suitable to provide nonredundant decomposition information and it permits to obtain an inverse WT. The discrete WT is the representation of a given signal f(t) ∈ L2(R) using a set of functions ψa,b(t), which are the scaled (by factor a) and shifted (by b) versions of a single function ψ(t) ∈ L2(R), called
the mother wavelet. The function ψ(t) has to satisfy the admissibility condition ( ∫ ψ( t ) dt = 0 ) to be a wavelet.
As shown by Daubechies (1992, p.10), using a logarithmic uniform spacing for the scale discretization with increasingly coarser spatial resolution at larger scales, a complete orthogonal wavelet basis can be constructed with
m ⎛ y − jb0 a0 ψ ([ m ]) ( y ) = a0m/2 ψ⎜ j m ⎜ a0 ⎝
⎞ ⎟, ⎟ ⎠
( 3.3 )
where m and j are variable scale and position indices, respectively, a0 is the base of the dilation, b0 is the translation length in units of a0m, and (m) is used as a scale index (not to be confused with power m). The simplest and most efficient case for practical computations is the dyadic arrangement (a0=2 and b0=1). All scales along octaves 2m and translations along 2m j contribute to the construction of f(xi) = f(j) using
m = ∞ i = +∞ m =1 i = −∞
[ i ] g ( m ) [ i − 2m j ] ,
( 3.4 )
where g(m)(i) is a discrete version of the continuous wavelet ψ(t) at scale m, and the wavelet coefficients W(m)(i) are obtained from the signal, by the following convolution W ( m ) [i ] =
i = +∞ i = −∞
[i − 2 m j ] f ( j ) ,
( 3.5 )
and they satisfy the conservation of energy condition
j = −∞
f ( j) 2 =
m = +∞ i = +∞ m =1 i = −∞
∑ ∑ (W
[i ]) 2 .
( 3.6 )
In general, the number of observations is finite and the summations in the above equations do not extend to infinity. If N = 2M is the number of observations (i.e., N is an integer power of 2), the scale index m then varies from 1 to M= log2(N) and the position index at scale m varies from 1 to N × 2-m. Note that this definition implies that as the scale increases, the spatial resolution becomes much coarser (e.g., at m=1, we have N/2 coefficients, at m=2 we have N/4 coefficients, at m=M we have 1 coefficient).
the total energy TE contained in scale Rm = (2m dy) is given by TE = N −1 i =2M − M i =1 ∑ (W ( m) f [i ]) 2 ( 3.3. is deduced from (6) using var( f ) = N −1 m= M i = N m =1 i =1 ∑ ∑(W (m) f [ i ]) 2 ( 3. Assuming that the observations are sampled every dy meters (or in the case of observations sampled at a fixed point.Just as the Fourier transform can be computed using the Fast Fourier Transform (FFT).7 ) where N is the number of observations (multiples of 2).3. m is the scale index. The wavelet coefficients are computed using a pair of dyadic orthogonal filters called Quadrature Mirror Filters (QMF). The detail signals at stage m and location i are the wavelet coefficients W(m)(i) (see Katul and Parlange (1994) for more details). with the flow passing through the sensors. 3. the discrete wavelet transform can be computed using the fast filtering scheme of Mallat (1989). 65 .8 ) This energy can be directly interpreted as the contribution from scale Rm to the total variance of the signal. The QMF filter associated with the scaling function yields an approximate or smoothed version of the original signal at successive resolutions. while the output of the filter associated with the wavelet gives the details of the signal. which are related to the mother wavelet and the scaling functions. in terms of its wavelet coefficients. Wavelet statistics Some statistical tools that utilize the wavelet coefficients can be deduced for characterizing the contribution of different scales to the total variances and covariances of turbulent signals. one can analyze in terms of time scales or use Taylor’s hypothesis). and i is the position index. The variance of a signal f(t). The outputs of the approximation filter are cascaded to give the different scales. M is log2(N).
1997). f 2 ) = N −1 m= M i = N m =1 i =1 ∑ ∑W (m) f1 [i] Wf 2 (m) [i] ( 3. Each record of vertical wind velocity component (w). that correspond to approximately 3. the turbulence time series of dataset A (B).4. using the Daubechies-8 WT. This WT is a discrete orthogonal wavelet and the cospectra based on this kind of detail separation can be interpreted as fluxes decomposed into values computed from moving averages (Howell and Mahrt. and smooth wavelets. the covariance of two signals f1(t) and f2(t) can be expressed as Cov( f 1 . Results and discussion In order to project transient signals on selected scales. 66 . in case of dataset B).Similarly. were divided on overlapping records of 131072 (262144) data points. Since our interest is mainly on the flux computation.9 ) and scale contribution to total covariance can be analyzed using (m) Cov ( f1 .10 ) 3.5) hours. such as Battle-Lemarie. humidity (q) and CO2 concentration (c) were then decomposed into 16 scales (15 scales. f 2 ) = N −1 i =2M − m i =1 ∑W (m) f1 [i] Wf 2 (m) [i] ( 3. Various wavelets were examined and there was very little differences in the scale variances and covariances. the exact shape of the wavelet is not very important. recovered from the datasets following a time step of one hour. and because under diurnal conditions. result similar to the one obtained by Katul and Parlange (1994) for turbulent data measured at atmospheric surface layer under several stability conditions.5 (4. like Haar. These record lengths were chosen to provide better analyses of low frequency motions. We have therefore chosen the Daubechies-8 wavelet as a compromise between very short abruptly changing wavelets. turbulent signals seldom presented spectra or cospectra low-frequency end at scales larger than 2 to 3 hours. virtual air temperature (Tv).
in almost all 4-hour records the intensity of the turbulent signal Iu follows the relation Iu < 0. 3. 23 records on wet season and 37 records on dry season were available for our analyses. in dataset A and 82 daytime records in dataset B.Next. (v) scale similarity regarding scalar fluxes and scale Bowen ratio analysis.1. Therefore. According to our data analysis. we can convert from time increments to space increments using y = <U> t ( 3.33 and so WC test for TH validity holds. one starting step should be addressed on Taylor’s hypothesis (TH) range of validity.4. results and discussion are presented split in the following parts: (i) lowerfrequency time series analyses and Taylor’s hypothesis range of validity.11 ) To assess TH range of validity we used Wyngaard and Clifford (1977) test (here referred to as WC) after which the turbulent intensity Iu (=σu / <U>. for most of our data. Taylor's frozen turbulence hypothesis can be used to convert from time scales to length scales: the instruments installed on a fixed point permit to obtain time records of the turbulent variables as the flow blows past the sensors with an a mean wind speed <U>. (vi) scale heat flux and mean horizontal wind speed relationship. (iii) scale covariance analyses providing useful information about momentum and scalar turbulent fluxes. (iv) separation of variances and fluxes on three classes.33. After excluding the records where WC test fails. where σu is the standard deviation of the turbulent longitudinal wind velocity component and <U> is the mean wind velocity which impacts the measuring instrument) should not exceed 0. (ii) scale variance of wind velocity w-component and scalars. Lower frequency time series analyses and Taylor’s hypothesis range of validity To perform our scale variability analysis. These records were excluded from the analysis. 67 . The only situations in which WC test failed are when the wind velocity averaged in the 4-hour period was relatively small.
w-variances during dry season present a more pronounced peak on scales ranging from 100 to 300 m than during wet season.4. The second period analyzed encloses days 234 to 244. 68 . The main differences observed between the two seasons are: (i) During dry season there are no significant differences among the shape of scalar variance curves along the full range of scales analyzed. the scale variances of w.2b.2. The first period. In the former.T. q and c. the q-variance still presents pronounced energy amount on scales ranging from 10 to 1000 m. we compare figures 3. While Tv-variance and c-variance present similar pattern as in the dry season. In the last. in the dry season of 2000. Variances of Tv. In order to understand the seasonal variability of these spectral energy distributions. were calculated for dataset A. Scale variances To perform an analysis on the scale contributions from different physical processes to total fluxes. q and c show a nearly exponential pattern growing from very small values at small scales. which were estimated using TH. we present the scale variances of w and of scalar variables during the wet season.3. the energy contained in each scale (TE(m)) is normalized by the total variance of the signals. showing a nearly linear growth. Nevertheless. comparatively to the dry season. we present the same variables. we observe that the scale variation patterns are similar during the two seasons. The scales are presented as eddy length scales. In such conditions. (iii) Regarding the differences between the scale variance of w in dry and wet seasons. is representative of late wet season. using (7).). the forest environment is likely to present higher soil moisture content and a more intensive convective activity predominates during this season. to their largest values at mesoscale. (ii) During the wet season period there are clear differences between scale variability of q-variance and of the other scalars variances. and then averaged during diurnal periods (9:00 to 17:00 h. but for the dry season. compared to larger scales.2a and 3. Tv. enclosing days 93 to 98 of 1999. L. with the same maximum-energy range scale. For comparison of scale variability.
virtual temperature (Tv).15 0. This in spite of positive temperature and vertical velocity skewness associated with warm moisture updrafts in the same region. with important boundary-layer instability. (b) Same as figure 3.05 0.05 0. we have calculated the 69 .20 0. p. These differences could be attributed to different dominant eddy structures in the dry and wet seasons.05 1 10 100 1000 10000 Length scale (m) Length scale (m) Figure 3.15 var(T) (xi) / vartot(xi) 0.(a) 0. but for averages from day 234 to 244 (dry season period).05 1 10 100 1000 10000 -0. boundary layer top-down eddy motions could transport dry air from the entrainment layer down to the surface layer leading to negative moisture skewness values there. 178). As possible mechanisms responsible by the seasonal var(q) differences we could take in consideration the remarks of Mahrt (1991) about boundary layer regime classes. As was demonstrated by him. over Rebio Jaru forest.20 (b) Dry season var(w) var(T) var(q) var(c) Wet season var(w) 0. The investigation of the physical mechanisms which generate different eddy structures in the ABL lead many authors to propose physical criteria to characterize distinct atmospheric boundary layer regimes.10 (m) (m) var 0. (a) Scale variances of vertical wind velocity (w). averaged during diurnal periods (9:00 to 17:00 h). The error bars represent the 95 % confidence level. for data collected from day 93 to 98 (wet season period). relatively weak surface evaporation and drier air aloft. where moisture skewness is positive near the surface. To investigate the existence of such mechanism in Amazon forest ABL.2a. An opposite situation occurs associated with greater surface evaporation regimes.2.10 var(c) (xi) / vartot(xi) var var(q) 0.00 0. and humidity (q) and CO2 concentrations (c).00 -0. A kind of classification of them was proposed by LeMone (1976) and discussed by authors such as Mahrt (1991) and Garstang and Fitzjarrald (1999.
Sq < 0 for all scales except two. These results probably explain our differences between Tv and q fields by means of the humidity transport mechanisms proposed by Mahrt (1991). The skewness factor of the scalar x at m-scale can be computed by ( S xm ) = (W (W (m) [i] 2 (m) [i] ) ) 3 3/ 2 . two of the scales in which Sq > 0 for dry season and Sq < 0 for wet season are the smallest ones and the results in this range are probably affected by very local phenomena related to roughness sublayer physical aspects. show clearly that. in dry season.scale skewness values of q (Sq) for wet and dry season periods.3. Sq > 0 for all scales except three and that. in wet season.2.3. This higher dispersion in wet season period is likely related to the more transient character of the flow during this season. 1 wet season dry season Skewness (q) 0 -1 0. consequent of the entrainment-drying boundary layer or of boundary-layer eddies which transport warmer. 70 .12 ) Our results. drier air towards the surface. However. It is also possible to observe that Sq values dispersion are more pronounced in wet season comparatively to dry season. presented in figure 3.1 1 10 100 1000 10000 Length scale (m) Figure 3. a mechanism earlier pointed out by Nicholls and Lemone (1980). ( 3. Scale skewness factor of humidity signals (Sq) for the same periods as in figure 3.
Details are the same as in figure 3. Figure 3. q and c) from the wavelet coefficients were carried out to assess the partial contribution of specific scales to the total fluxes of energy and carbon. averaged at the same periods as in figure 3.s ) -2 -1 80 (a) 1 0 (b) 40 -1 -2 -3 -4 1 10 100 1000 10000 wet season dry season 20 0 1 10 100 1000 10000 Length scale (m) Length scale (m) Figure 3. that should involve nearly isotropic motions. the anisotropy predominates in lowest scales and the contributions for turbulent fluxes in this interval are not null has been analyzed by Katul et al. Both sensible and latent heat fluxes. under certain conditions. and figure 3. On the smallest scales. reach a maximum value at scales close to 400 m (corresponding to time scales of approximately 3 minutes).4. Scale covariance contribution to fluxes of (a) sensible (H) and latent heat (λE) fluxes. as well as CO2 flux. averaged for the same period as in figure 3. the fluxes increase significantly. and to (b) CO2 fluxes. (1997).m-2) 60 H (wet season) λE (wet season) H (dry season) λE (dry season) Scale CO2 fluxes (μmol.m .4a shows this scale dependence for sensible (H) and latent (λE) heat fluxes. who investigated situations where large-scale anisotropy disturbs the inertial subrange isotropy.2. We will discuss some aspects of the high and low-frequency features of our results. although different from zero. and then decrease to smaller Scale heat fluxes (W.4b. 71 . corresponding to the fully developed turbulence inertial subrange.2. The fact that. what expresses a strong turbulent transport contribution to the total turbulent transport coming from this interval. Scale fluxes Scale covariance calculations of w and scalars (Tv. for CO2 flux.3. This could be an explanation concerning our results. From 10 m towards larger scales.2. the fluxes are very small.3.4.
4. Coincidentally. we suggest that cloud gap effects represent the more important source of w-TKE in the Amazon forest (wet) convective boundary layer and they could explain. over an undisturbed forest area in Central Amazonia. The physical origins of these maximum-energy eddies in such scale interval are likely related to convective processes which are typical of the Amazonian tropical boundary layer. two seasonal differences are observed: (i) one is related to the available energy partition between sensible and latent heat flux (figure 3. is clearly observed in figure 3. (2001) have recently presented evidences of cloud modulation of solar irradiance in a Amazonian pasture (located not far from our experimental site) associated with cloud gap patterns whose long time fluctuations are of the order of 3 min. at least in part. related to different CO2 diurnal uptake from the atmosphere by forest vegetation in wet and dry seasons. During dry season the H mean value increases and λE mean value decreases compared to mean values obtained during the wet season. Although the largest amount of the energy and mass fluxes occur on turbulent scales lower or of the order of the w-spectral peak.4a). This shows up by the estimated large sampling errors that are observed in this figure. This kind of water availability constraint is likely to diminish our diurnal CO2 flux measurements too. in Amazonian forest. compared to wet season. the soil water stress observed during dry season lead to smaller CO2 assimilation by vegetation. and therefore. Gu et al.values at larger scales. A similar seasonal behavior of CO2 diurnal fluxes was also observed by Malhi et al. the same time-scale order as we have obtained in our results. (ii) the second. However. Based on them. during wet and dry seasons. the diurnal CO2 uptake by photosynthesis activity is apparently constrained by water availability. 72 . According to their measurements. (1998).4b. larger scale eddy motions could generate important contributions to the total fluxes. This is an expected result since the dominant hydrology conditions during dry season lead to lower evapotranspiration and higher sensible heat fluxes from the vegetation to the atmosphere in this period. at largest scales region. According to our results. They are likely to be responsible for convective turbulence regimes during diurnal unstable conditions over Amazonia. these low-frequency contributions to surface fluxes also show large variation among the various investigated data records. the physical origin of our 3 min energy-peaks. The general pattern of heat and CO2 fluxes scale dependence is very similar considering wet and dry season periods. as we can see on figure 3.
4.13 ) which is a ratio between the standard deviation of the calculated m-scale flux for scalar x and the same variable mean value. However. there is a clear increase of NSDF with length scale.5b we show estimations of NSDF for sensible heat. there is no important variation of NSDF along the scales. In figures 3. Despite the fact that these shapes are the same for the two studied periods. In the first. the threshold length scale separating the two NSDF variability regimes in wet season is lesser than the one obtained in the dry season.4.3. (1996). Variances and fluxes separated in classes There is no consensus in the literature with respect to categories in which the variances and fluxes might be classified regarding their scale dependence. associated with the larger scales. To obtain such scale dispersions we calculate a normalized standard deviation of the scale flux (NSDF) by means of the expression: NSDF( m ) = var 1 / 2 (( w' x' )( m ) ) mean(( w' x' )( m ) ) ( 3. In this section we will investigate the feasibility of application of the above criteria. this is not true for (wc) curve. Among the various classification schemes mentioned in our introduction. We propose to assess the more adequate classification by both analyzing our earlier section results and investigating the scale dispersion of the flux calculations.5a and 3. whose two-regimes threshold occurs at length scales smaller than the ones for (wT) and (wq). latent heat and CO2 fluxes at wet and dry seasons. In the second. We will perform this based on the available measured variables to choose an appropriate classification scheme concerning scale dependence fluxes in the Amazonian ABL. associated with smaller scales range. we will investigate two classification suggestions: the three classes proposed by Sun et al. (1996) and the two main classes proposed by Williams et al. respectively. 73 . All the curves in both dry and wet seasons depict a clearly two general categories pattern with respect to scale variation of NSDF. the curves for (wT) and (wq) present a very similar behavior. except for the largest scales analyzed. We also observe that in all available situations.
Other aspects of the flux scale dependence issue will be discussed after presentation of our next section results. and these percentages are even slightly higher during dry season 29 % and 30 %. wq and wc scale covariances. 74 . However. we present Table 3. (1996) three-classes scheme. q and c averaged during the two periods of dataset A. (1996) all suggest only one clear distinction isolating turbulent fluctuations from larger-scale variations. After Sun et al. the existence of three distinct classes is suggested by the dependence of momentum flux calculations on both the cutoff length scale and flux averaging scale. for sensible and latent heat fluxes.1. in the wet season. (b) dry season. In this table we present the mean by-class and total values of heat fluxes. this threshold would separate the low-frequency components in two classes: ‘large eddies’ and ‘mesoscale’. These results led us to consider that a two-classes scheme for eddies containing fluxes is more appropriate to explain scale scalar flux variability in Amazonia than the Sun et al. a number of independent considerations from authors such as Williams et al. based only on scale scalar flux behavior. As their measurements of momentum flux are nearly independent of flux averaging scale until scales close to 5 km. Our results confirm the existence of two main classes of scalar fluxes. To provide quantitative information about the amount of flux contained in these two turbulence-pattern classes. Tv. About 28 % of sensible heat and 27 % of latent heat is transported by motions in scales higher than 800 m. CO2 flux and variances of w.(a) 4 Wet season (b) 4 Dry season 3 (w' xi') (m) NSDF NSDF (m) 2 1 0 1 10 100 1000 10000 (w' xi') w'T' w'q' w'c' 3 w'T' w'q' w'c' 2 1 0 1 10 100 1000 10000 Length scale (m) Length scale (m) Figure 3.2: (a) wet season. (1996). Normalized m-scale standard deviations of fluxes (NSDF) for wT. Bowen ratio. calculated during the same periods as in figure 3.5.
m-2.11 0.16 18.m-2) CO2 flux (μmol. For CO2 fluxes.78 83. After them.28 0. By means of a theoretical treatment.12 0.91 Turbulent 101. the results are similar: 30 % and 27 %. Such a non-homogeneous boundary condition could generate mesoscale circulations that would explain our low-frequency scalar fluxes.07 2.36 0.90 -9. Table 3.23 Total 99.04 -11. during wet and dry season periods.60 -4.15 14.32 0.50 Dry season Mesoscale 41.10 -2 λE (W. which would require averaging over a much larger dataset including a wider range of conditions.81 -7.90 0.67 0. but provide a simple summary of the scale dependence of fluxes over Rebio Jaru forest.46 0. These calculations do not provide a true climatology of fluxes.32 0.40 0.78 190.04 0.68 .24 0.29 21. the surface fluxes of temperature and humidity do indeed vary in step with low-frequency variations in the wind under certain non-homogeneous surface conditions. It is important also to keep in mind that our experimental site is in a rain forest strip.33 0.respectively. too.1.30 0.mol-2) 2 -2 2 -2 0.35 0.12 3. respectively. as described by Garstang and Fitzjarrald (1999).53 0.09 -2.06 0. occur on large eddy motions.51 0. Partition of the total surface flux contribution from each one of the two scale ranges for wet and dry season.22 16.98 β (=H/λE) var(w) (m . which is surrounded by deforested areas in a very peculiar "fish-bone" pattern.42 0.98 75 H (W. An explanation of our results can be based on the conclusions presented by McNaughton and Laubach (2000) in their investigation about the consequences of the unsteadiness of the wind field on the scalar fields.48 Total 143. Our purpose here is to warn about the numerous effects that can influence the exchange processes on this disturbed ABL.52 0.09 -15. Wet season Turbulent Mesoscale 27.61 0.s-1) 83.50 221.94 304.78 0.56 273. they showed that this would affect the values of the eddy-diffusivities for temperature and humidity in different ways.m-2) 71. It is interesting to note that this kind of low-frequency wind velocity variability and that dissimilarity between Tv and q fluctuations were observed in Rebio-Jaru turbulent data.s ) var(T) (K ) var(q) (g kg ) var(c) (μmol2.50 0.
0 -0. plotted against its related length scale.4. 1989) and provide useful information about external forcings acting on ABL borders. In this section we will investigate two related subjects: the scale correlation coefficient between q and Tv and the scale relationship between sensible and latent heat fluxes (Bowen ratio concept extended to a scale assessment). 76 .6 we present the scale correlation coefficient between Tv and q.4 1 10 100 1000 10000 wet season dry season Length scale (m) Figure 3.6.2. rTvq = 0 at the length scales of the order of 1 m and increases until 1. for dry and wet seasons. the structure of moisture fluctuations in the boundary layer is different of that for heat turns the comparison of the moisture and heat statistics an useful tool to obtain some more physical insight with reference to flux scale-dependence problem. 1991). Scale correlation coefficient between virtual temperature (Tv) and humidity (q).4 0. In figure 3. 1984.8 0.2 0.0 0. Temperature-humidity correlation and Bowen ratio scale dependence Studies about similarity or dissimilarity in the potential temperature and humidity fields in the ABL are well known (Moeng and Wyngaard.6 Correlation T-q 0. Observing these results. The fact that under certain conditions. averaged for the same periods as in figure 3. rTvq.5. it is clear that rTvq<1 for all analyzed scales. This is particularly interesting with respect to entrainment zone influences on boundary layer top and even on atmospheric surface layers (Mahrt.3.2 -0. The error bars represent the 95% confidence level. Hill.
85. In this context. and therefore the rTvq should diminish only slowly with the length scale. This fast decrement of rTvq curve for the smallest length scales close to 1 m is not expected. 1981.05 to 0.2. the mean hourly β calculated from values from 07:00 to 16:00 h varied from 0. In such scales. However. whose structure device center and inlet tube entrance separation is of the order of 20 cm). From the scale of 10 m up to approximately 100 m. We could formulate tentative explanations for the above results: (i) For the scales of the order of tens of meters. one first possible remark is that the atmospheric surface layer Monin-Obukhov Similarity Theory (MOST) does not hold for all investigated scale-ranges. 1990). it would be useful for us to take in mind McNaughton and Laubach (2000) propositions to explain the breakdown of MOST observed in our results. rather than the much faster "normal co-spectra" decrease (Andreas. Fitzjarrald et al. 1987).4. (1988). Bowen ratio (β = H/λE) is an important micrometeorological parameter that expresses how the surface available energy is shared in sensible and latent heat fluxes. concerning different boundary layer moisture regimes and their consequences for dissimilarity between Tv and q fluctuations as we have already discussed in section 3. Such results are not in opposition 77 . what could justify the drastic fall observed in correlation coefficient values. since at these scales the q-Tv co-spectrum should behave as a – 5/3 power law. we indeed expect the failure of MOST (Raupach and Thom. this our result might be attributed to the distance between the Tv and q measuring probes (recall that Tv is measured by sonic anemometer and q is measured by IRGA. However. both rTvq curves fall down until cross the zero-axis at a length scale of the order of 3 km.7 in the dry-season curve and of approximately 0. Earlier investigations have already determined the overall characteristics of Bowen ratio variability in Amazon forest. From this point. (ii) Another argument is based on the fact that there are topographically induced eddying as well as convectively induced eddying over the Rebio Jaru site. Tv and q turbulent signals must present an important phase-difference. probably the more adequate explanation is that proposed by Mahrt (1991).6 in the wet season curve. such as studies by Sá et al. such boundary conditions could introduce low-frequency wind speed fluctuations that generate dissimilarities between the eddy diffusivities for temperature and humidity. rTvq presents a nearly constant value of approximately 0. Starting from these results and based on Hill (1989) and De Bruin et al. as the measurements were performed in a transition roughness sublayer. For the length scales greater than 100 m. they become negative.reaching the 10 m length scale. After them. So it is not surprising to obtain 0 < rTvq < 1 in that region. (1999) findings. As these authors have shown.
for the same periods as in figure 3. However.2 wet season dry season 0. with our mean values shown in Table 3.5 during dry season.0 0.1. Additionally. 78 . the scale β remains almost constant. the fact that Tv and q measurements were performed 20 cm apart from each other can also introduce some errors there. (ii) at the largest scales.3 during wet season and 0. as expected.7. no systematic study has been carried out to assess the scale variability of this ratio in Amazon forest environment. The fact that β presents discrepant values only at the edges of the investigated scale range may be attributed to: (i) at the smallest length scales.2. as the fluxes are very small.8 Scale Ratio H/λE 0. we do not expect that rTvq is statistically a robust value and the physical processes which determine the heat exchanges there do not concerns inner boundary layer processes. the ratio between them can be largely influenced by errors. the ratio is larger in the former than in the last period.1. Ratio between scale sensible and latent heat fluxes. Except for length scales lesser than 10 m or larger than a few kilometers.6 0. for each one of the investigated classes. as discussed by Williams et al.7 the shape of the scale variability of H/λE mean values is quite similar for dry and wet seasons and. around 0.4 0.0 1 10 100 1000 10000 Length scale (m) Figure 3. (1996). After figure 3. but external peculiar forcings.
For the low-frequency contributions. this seems to confirm our suggestion that these contributions to the total heat flux are associated with convective cloud gap patterns associated with updrafts.6.25 0. The larger scale Rm show a clear decrease pattern with increasing of <U>.00 -0. Ratio of sensible heat fluxes of two eddy-pattern classes to total sensible heat flux (Rm).25 Turbulent Mesoscale 1. 79 . calculated from day 31 to 60 over Rebio Jaru forest. as a function of horizontal wind speed. since they probably drive the main TKE generation mechanism in such region.75 0. The ratios were separated by wind speed classes and then averaged for each class. calculated from dataset B (wet season period only). Scale heat flux and mean horizontal wind speed relationship In figure 3. the greater the relative importance of the horizontal temperature gradient induced mesoscale motions to the overall wind field configuration .8 we present Rm (the ratio of scale wTv covariance contributions of a specific eddy-pattern class to total sensible heat flux) as a function of the mean horizontal wind speed.50 0.25 1 2 3 4 Horizontal wind speed (m/s) Figure 3. Starting from the supposition that these horizontal T-gradient are 1. The standard errors are also shown. For the turbulent contributions. this would be attributed to the fact that the lower the mean speed.4.00 Ratio Rm (Hclass/Htotal) 0. It is interesting to observe that the turbulent Rm becomes more important as <U> enhances. <U>.8.3.
Based on this information. were projected into 15 scales using the Daubechies-8 wavelet transform (WT). the velocity fluctuations may more closely approach pure updraft and downdraft motions and the special flow distortion effects under these situations would enhance low frequency motions. q and c show the higher values at mesoscale. involving processes not expected to be controlled only by ABL parameters. ranging from the inertial subrange domain up to scales on the order of 800 m (or 6. in south west Amazonia. and normalized scale stardanrd deviation of scalar fluxes were calculated. 3. Variances of Tv. particularly concerning CO2 fluctuations. (1998) have pointed out the importance of distribution of wind directions in Amazonia. virtual temperature (Tv).associated with fish-bone deforestated strips and taking in consideration that wind direction often changes above Amazonian forest. (ii) the existence of stationary eddies.5 minutes). a peculiar feature of the TKE generation in equatorial regions (Garstang and Fitzjarrald. two main flux scaledependence classes have been identified: (1) Turbulent scales– main scales of vertical turbulent transport of mass and energy in the atmospheric surface layer. Summary and conclusions High frequency measurements (10. or 16 Hz) of vertical wind velocity (w).5. Malhi et al. The variance of w shows most of turbulent kinetic energy occurring on turbulent scales reaching a maximum on scales close to 300 m and decreasing to very low values on larger scales. we might expect important transient cellular motions under such conditions. which could be attached to surface heterogeneity elements or could be slowly moving with weak winds. indicating a likely high influence of mesoscale motions and convective systems that act on amazonian ABL on 80 . humidity (q) and CO2 concentration (c) of air.4 Hz. obtained over Rebio Jaru tropical rain forest reservation. (2) Large scale eddies – motions involving eddies of scales on the order of or larger than the height of the ABL. 1999). The relative contributions of each scale to the total variances and covariances were then assessed. Some remarks of Mahrt (1998) might support other possible arguments to explain this results: (i) under weak large-scale flow and significant surface heating.
these lowfrequency contributions to surface fluxes show large variation among the several investigated data records. latent heat and CO2 fluxes occur on turbulent length scales below or of the order of the w-spectral peak scale. 81 . however. The results show that the turbulent fluctuations pattern of Tv and q are not similar. This suggests that during dry season top-down eddy motions could transport dry air from the entrainment layer down to surface layer leading to negative moisture skewness values.these data. between virtual temperature (Tv) and humidity (q). The largest amount of the sensible heat. and can be either positive or negative irrespectively of mean ABL gradient conditions. About 30 % of scalar fluxes were found to be transported by motions on scales larger than 800 m. In addition. Scale skewness calculation for humidity data were predominantly negative in dry season and positive in wet season. Larger scale eddy motions could. rTvq. on both wet and dry season studied periods. and rTvq < 1 for all analyzed scales. We also performed scale calculations of the correlation coefficients. generate important contributions to the total fluxes. what does not occur during wet season.
Kruijt. 192207. von Randow. Holtslag. It is found that the MOS relationships and the fluxvariance method provide reasonable results only when the w-T correlation (rwt) is above 0. We test the estimation of heat fluxes by the flux-variance method. to scale the standard deviations and parameterize the modulation of lowfrequency motions. and in these conditions the surface layer cannot be characterized by the ‘textbook’ descriptions of the surface layer observed over uniform terrain. indicate that during roughly 20 % of the time the unstable surface layer above the forest deviates from the ‘classical’ description. we test the use of the ‘dissipation velocity’ uε = (kzε)1/3. Low-frequency modulation of surface layer over Amazonian rain forest and its implication for similarity relationships. In this sense. ____________________ This chapter is published as C. proposed by McNaughton (2006). The systematic variation with stability is taken out by the use of the new parameters. a greater influence of low-frequency processes tends to cause rwt and ruw to decrease.5. Agricultural and Forest Meteorology. north of Manaus.Chapter 4 Low-frequency modulation of the atmospheric surface layer over amazonian rain forest and its implication for similarity relationships Abstract The application of Monin-Obukhov similarity theory (MOS) is based on empirical relationships derived over uniform surfaces in flat terrain. Brazil. This study investigates the influence of low-frequency motions in the structure of the atmospheric surface layer over Amazonian forest and its implication for the application of MOS theory. 83 . which is based on MOS theory. and the scaled variables become independent of the MOS stability parameter ζ.A. 141. which represent the modulation of the outerlayer motions on the surface layer. The results highlight the complexities of the surface layer above vegetation such as Amazonian forest. 2006. for measurements in unstable conditions in the K34 forest site in central Amazonia. A. It is not clear to what extent these relationships hold for complex surfaces such as tropical forest or hilly terrain. This result is consistent with the self-organizing nature of the turbulent structure in the modulated surface layer. Estimations of the parameter v*/u*. B.M. As an alternative to the conventional MOS scaling. Examining the scale dependence of rwt and of u-w correlation (ruw) revealed that w variations tend to be not well correlated with fluctuations in u or T at low frequencies.
1993).g. McNaughton and Brunet. These are low-frequency motions that often span the whole boundary layer (outer layer). most of the empirical relationships derived with the objective of applying MOS theory have been determined for uniform surfaces with flat terrain. The sources of the low-frequency contributions to the total exchange and the interactions between the large boundary-layer eddies and the surface layer are not yet well understood. used in the flux-variance method (Tillman.. Townsend. However.. frequently proposed as the cause for the lack of energy balance closure at many sites (Twine et al. 2003. they are likely to explain part of the scatter around the expected trend lines of similarity relationships. the flux-variance method has been used as an indicator of the validity of the MOS in some sites (e. von Randow et al. therefore. parallel to the ground. however. and so are ‘inactive’ and do not interact with the small-scale eddies (“Townsend hypothesis”. 2001. such as the height of the boundary layer (zi) and the boundary-layer convective velocity scale (w*). however.. De Bruin et al.. The variability in the vertical wind component and the fluxes should be. 1991. Johansson et al. Finnigan et al. which is based on MOS theory. In the surface layer. The flux-variance method. Moreover. 2002). It is not clear to what extent these relationships hold for complex surfaces such as tropical forests or hilly terrain. They are. such as the relationships between the standard deviations of scalar signals and the stability parameter ζ. it is now known that the small-scale turbulence processes that obey MOS scaling (depending only on ζ) coexist with larger-scale motions that are related to parameters from outside this local surface layer. 2000. De Bruin et al.. 2000.1. normalized with appropriate (local) surface layer parameters and the height z. where L is the MoninObukhov length).4. are universal functions of ζ (=z/L. Near the surface it has been assumed that these low-frequency motions are mainly horizontal. related only to local parameters. This theory predicts that dependent variables. 1961). estimates the surface fluxes 84 . 2004). Introduction Many of the measurement techniques and models of turbulent transport in the atmospheric surface layer rely on Monin-Obukhov Similarity (MOS) theory. mounting evidence that interactions do occur and the outer-layer processes ‘modulate’ the turbulence structure in the surface layer (McNaughton and Laubach. In fact. There is. 1972)..
1993). These cause extra energy to be transported down into the surface layer. especially for stable conditions. McNaughton (2006) analyzed the terms of the kinetic energy budget equation and argued that the dissipation rate of TKE could be used to parameterize a fluctuating friction velocity v*. and u' is another fluctuating component representing the inner-scale motions of the surface layer eddy structures and their breakdown products (note that simple Reynolds decomposition ~ would combine both u and u' within one fluctuating part only). This method also relies on a similarity relationship to relate CT2 and the heat flux. that for evaluation of the regional budget of energy and mass fluxes. 1991. Lloyd et al. It has been shown that the method works well in deriving the daytime heat and momentum fluxes over uniform terrain (Weaver 1990. According to this view. but do not affect its local structure (McNaughton ~ 2006). however.using measurements of the variances (or the standard deviations) of the scalars. thus modulating the processes. that the low-frequency contributions to the variances will add uncertainties to the estimations.. outer-scale convective motions that cause the local alignment and power of the shear turbulence across the surface layer to vary. the structure parameter of temperature (CT2) can also be used to estimate the heat fluxes (“structure parameter” method). 2002). that represents the additional energy transported down from the variable motions of the outer layer to the surface layer. von Randow et al. Vickers and Mahrt (2003) suggested a multi-resolution decomposition (wavelets) to define a timescale that includes only the “turbulent motions” and show improvement in the similarity relationships. The latter are associated with the large. In a similar way to the flux-variance method. An appropriate velocity scale to be 85 . 2001. regardless of their origins.. De Bruin et al. The author proposed that the surface layer is driven from above by an outer layer that has both mean and variable motions. u is a fluctuating component representing the variable part of the outer-scale forcing of the surface layer. It is likely. It should be emphasized. however. cannot be excluded (Sakai et al. McNaughton (2004) proposed a new model to explain how the turbulent processes in the outer part of the boundary layer modulate the processes near the ground. the low-frequency contributions to the total exchange. so enhancing dissipation. where u is a mean component. Based on this model.. the velocity field can be divided into three parts: u = u + u + u ' ~ (similarly for v and w components). To what extent these methods (and the MOS theory in general) work in complex terrain such as Amazonian forest is still unknown..
1.2.used in the surface layer is then a ‘dissipation velocity’. 4. Using the dissipation velocity scale to parameterize the outer-layer modulation of the surface layer. The standard deviation of a scalar x. g is the acceleration of gravity and w' Tv ' is the flux of virtual potential temperature θv). Theoretical background 4. uε = (kzε)1/3 = (u*3 + v*3 )1/3 . the zero-plane displacement can be over 20 m.1 ) Here ζ = z/L.2. According to MOS the standard deviations of the horizontal (u) and vertical (w) wind speed components scaled by the friction velocity (u*) follow a universal function of ζ: σu u* = f u (ζ ) . k is the von Karman constant. σw u* = f w (ζ ) ( 4. Monin-Obukhov scaling and the flux-variance method The flux-variance method is based on the relationship between the variances (or the standard deviation) of the scalars and the Monin-Obukhov parameter (ζ). like those found in Amazonia.u*3 T / k g w' Tv ' . that is. as in MOS theory. T is the air temperature. In tall forests. we base our analyses on the new McNaughton theory. and evaluate how the low-frequency motions influence the turbulence structure near the surface over Amazonian forest. it is important to consider the height z as z-d. such as the air temperature (T) and the specific humidity (q). Note that. with the variable ζ. should be scaled by its respective scaling variable x* (= w' x ' /u*): σx x* 86 = f x (ζ ) ( 4.2 ) . and not u* only. the height above the zero-plane displacement d. with z the height above a zero-plane displacement and L the Obukhov length (L = .
CT2 can also be used to estimate the heat fluxes based on a Monin-Obukhov similarity relationship (Wyngaard et al. It should be noted that in the case of the horizontal wind velocity u. using an initial estimate for the value of ζ. 1991).9 and cT2 = 28. 1971) 2 CT z 2 / 3 = fTT (ζ ) 2 T* 2 σw u*2 = 1. De Bruin et al. Since ζ is still dependent on u* and on the surface heat flux. 1971): f x (ζ ) = ± c x1 (1 − c x 2ζ ) ±1 / 3 ( 4. and the – sign to temperature and water vapor.Several authors have applied the flux-variance method using MOS scaling successfully over uniform terrain (e. consistent with field observations and LES simulations of the boundary layer.g. 1984. or estimating u* from the average horizontal wind speed and the standard flux-profile relation (Panosfky and Dutton.5ζ ) 2 / 3 (Eq.4 with data collected in the surface layer over a relatively homogeneous area in the plain of La Crau (South of France). The attractive idea is to measure the standard deviations of the scalars with relatively simple instruments and then use the universal functions from (1) and (2) to derive u* and x*. Monin and Yaglom. (1993) have successfully used cT1 = 2. In a similar way to the flux-variance method. As a practical approach for unstable cases. using a wind speed scale based on the dissipation rate ε and a temperature scale based on the structure parameter of temperature (CT2). is Holtslag and Moeng.3 ) In which cx1 and cx2 are constants and the + sign applies to w. 1984).3 should be solved iteratively. Lloyd et al. Equations 4. 15b of ( 4. The form usually proposed for the functions fx(ζ) is (Panofsky and Dutton. De Bruin et al.44(1 − 1.. Holtslag and Moeng (1991) have found that a better function to describe the variance of w.4 ) 87 . 1991). For the standard deviation of air temperature. (1993) proposed an approximation solution interpolating the solutions for neutral conditions and the free convection limit (both independent of ζ) that eliminates the need for iteration.1-4. Hartogensis and De Bruin (2005) also proposed a method that does not require a numerical iteration. and so MOS is not followed (Panofsky and Dutton. 1984).. there is evidence that the variability is influenced by large boundary-layer eddies.
McNaughton (2006) relates the dissipation rate of TKE. As an additional test for the application of MOS and indirect methods to estimate the fluxes in Amazonian forest. Scaling on ‘dissipation velocity’ MOS was developed when turbulence was considered as consisting of small eddies that respond only to local conditions in the flow and small scale processes. and it should be included here because v* is variable and it is its average effect that balances the dissipation rate in the TKE budget equation. the new McNaughton theory (2004. with cTT1 = 4. so larger-scale influences were not included. It should also be pointed out that while McNaughton’s discussion of the TKE budget was for the regular part of surface layer. To simplify the notation. with both the mean motion and the variable convective motions of the outer layer creating shear across in the surface layer. in our case we have 88 . The outer convective motions modulate the turbulence processes in the surface layer. Based on this model. 2006) views the turbulence near the surface as driven from above. In contrast.De Bruin et al. hereafter we redefine v* as v*3 ( ) 1/ 3 . but the energy divergence is cumulative.9 and cTT2 = 9 yielded good results. In spite of this general observation. The bar over v*3 represents an average over runs. where the roughness sublayer can extend its effects up to 2 or 3 times the canopy height.5 ) According to these analyses. causing variable downwards fluxes of momentum (scaled by the fluctuating friction velocity v*) and associated kinetic energy.2. we compare these proposed relationships with the data collected at our site. (1993) found that fTT = cTT1 (1 – cTT2 ζ)-2/3 .2. u* and v* by ε= 1 3 u* + v*3 kz ( ) ( 4. the measurements discussed in this paper are made over a tall tropical forest. uε = (kzε)1/3 (= (u*3 + v*3 )1/3 . 4. increasing the rate at which that energy is dissipated. The variable components of momentum flux sum to zero. the appropriate velocity scale to be used in the surface layer is the ‘dissipation velocity’.
however this requires the assumption that the transport mechanisms for momentum and scalars are the same (Reynolds analogy) and that the turbulent Prandtl number equals unity. based on McNaughton’s discussions. where εo is the dissipation rate of the outer layer). ( 4. By analogy with the velocity scale.estimated that the effects of the roughness sublayer are small at the height we are measuring (see Section 3.7 ) σT Tε = FT (ζ ') . the dissipation rate of temperature variance εT could be used. a relevant length scale is the height of the surface layer (zs) which is determined by the interaction of the large outer scale eddies with the surface layer eddies (zs ~ uε3 / kεo.6 ) It might then be expected that the scaled standard deviations follow relationships of the form σu uε = Fu (ζ ') . but that requires an independent temperature scale that represents the equivalent effects of low-frequency modulation in the temperature field. There is 89 . An alternative flux-variance approach for heat fluxes based on these views would also be useful. In fact. thus according to this new view the processes would be described as functions of z/zs. temperature (Tε) and humidity (qε) as Tε ∝ − w' T ' uε and qε ∝ − w' q' uε ( 4. the ‘dissipation velocity’ uε can be used to scale the wind components. σq qε = Fq (ζ ') where ζ' is a dimensionless height (of this new framework) equivalent to the conventional MOS parameter ζ.3 below). σw uε = Fw (ζ ') . This would replace the Monin-Obukhov length scale L. A first guess would be to extend the structural model of McNaughton (2006) to the temperature processes. We will remain with the MOS traditional parameter ζ in our figures because this gives direct comparisons with the Monin-Obukhov scaling. As with Monin-Obukhov scaling.
9 ) where S2 is an empirical constant.2.55 which is also in agreement with the commonly accepted value of the von Karman constant k = 0.4 (Kaimal and Finnigan. Note that κ and r are related as κ = 2π / r. The temperature scale Tε should be proportional to the heat flux and inversely proportional to uε. 4. Relations between spectra. We use α = 0. ε and CT2 Kolmogorov (1941) derived a well-known description of the velocity spectrum in the inertial subrange as: Su(κ) = α ε2/3 κ-5/3 ( 4.02 α (Albertson et al. pg.8 ) where Su(κ) is the spectral energy density at wavenumber κ and α is an empirical constant. keeping in mind that the uncertainties due to the assumption of Reynolds analogy might limit its application. estimated as being between 0. and partly on the mixing action of TKE presented in Eq.evidence that this is not the case (Kays. where Δu(r) is the difference between wind velocity at two points separated by a distance r and the angle brackets denote an average operator): Duu (r ) = S 2ε 2 / 3 r 2 / 3 ( 4. The description of the inertial range of the temperature spectrum depends partly on the T-dissipation (εT). 1994. 63-64).10 ) 90 . A counterpart relation involves the second-order structure function (Duu(r) ≡ <(Δu(r))2>.3.5 and 0.8) by ε: ST(κ) = βT εT ε-1/3 κ-5/3 ( 4. 1997). 1994. and Taylor’s hypothesis of frozen turbulence is generally used for expressing temporal velocity data taken at a single point as spatial data. it is useful to attempt the estimation of the heat fluxes using an alternative temperature scale. Tennekes. but with an unknown scale factor. Nevertheless. 1970). (4. and the empirical constants are related under a constant skewness assumption as S2 = 4.6..
we can see that CT2 can be regarded as a scaling factor for the T-spectrum in the inertial subrange ST(κ) = 0. at a height of ~20 m above the canopy (53 m above the ground). 91 . Site description and data processing The data used in this paper were collected during 2000 and 2001 as part of the LBA project (“Large Scale Biosphere-Atmosphere Experiment in Amazonia) at the “K34” experimental field site.11 ) From Equations 4. It is a pristine rain forest reserve.g. These are closed-path systems. The eddy covariance system has a similar design to that described by Moncrieff et al. Since mid-1999 the main observation tower has been instrumented with.1 s response time) closed path.10 and 4.12 ) 4. ε and εT relate to each other through the structure parameter of temperature.where βT is the so-called Obukhov-Corrsin constant. The landscape consists of plateaus dissected by often waterlogged valleys. infrared gas analyzer (IRGA.4 Hz. Gill Instruments. in central Amazonia.3. More detailed information about the site and the instrumentation on the K34 tower are provided by Araujo et al.25 CT2 κ-5/3 ( 4. UK) and a fast-response (0. Monin and Yaglom.11. The height of the vegetation is around 30 m. composed of a three axis sonic anemometer (Solent 1012R2. located about 50 km north of Manaus. an eddy covariance system measuring the fluxes of energy and carbon dioxide. LICOR. USA). INPA). among other equipment. 1975) CT2 = 4 βT εT ε-1/3 ( 4. LI-6262. CT2 (e. Wind velocity components and temperature measured by the sonic and the H2O and CO2 mixing ratio analogue signals output by the IRGA were recorded at a rate of 10.(2002). protected by the Brazilian National Institute for Amazonian Research (“Instituto Nacional de Pesquisas da Amazônia”. (1997). a scaling constant of the 1-dimensional scalar spectrum.
to include cross-wind and humidity corrections of the sonic temperature signal (Schotanus. The first criterion to select the data is based on the difference between two commonly chosen detrending methods to split the mean and turbulent quantities in the flux calculations. 1983) and to include corrections for instrument responses and damping of fluctuations through the IRGA tube. therefore. with ζ < 0 and the ‘uncertainty factor’ for both sensible and latent heat below 40 W m-2 (see Appendix). 4. Generally for this site and instrumental setup. using block averages to separate the mean and fluctuating parts of the turbulent signals. For a study of the influence of the low frequencies in the surface layer similarity relationships. records of 4 hours were gathered. which can be adapted to a number of different hardware configurations and program options. Data screening For the analyses of the flux-variance relationships only data for unstable conditions during daytime are considered. The program was set to apply two axis rotations to align the coordinate frame with the mean streamlines and to force the mean vertical component (w) to zero (McMillen. the corrections are relatively small and do not represent a large uncertainty factor in the final values (Kruijt et al. to compensate for the time delay in the IRGA signals.. 1988). 92 . (1997) and Aubinet et al. selects only daytime measurements between 10:00 and 17:00 hours. (2000). Before the calculations. Alteddy is in-house developed software written in FORTRAN. The first data screening.3. apart from the coordinate rotations. following the methodology described by Moncrieff et al. These values were calculated for hourly intervals. 2004). We briefly describe the calculation of this ‘uncertainty factor’ in the Appendix.1.The calculations of the standard deviations of the signals and the fluxes were performed using the Alteddy software (see below). the data were screened to reduce the uncertainties introduced by bad instrument performance and strong non-stationary conditions.
1.33 (Taylor’s hypothesis validity range..4 Hz. Subsequently the spectra were smoothed by averaging the data within logarithmically spaced bins of wavenumber.3.5 we study how the fluctuations on relatively large scales (low frequencies) influence this correlation coefficient. which is expected to be -0. Estimation of ε and CT2 Based on the relations between ε and the structure function or velocity spectrum.001 to 100. 2002) (ii) The inertial subrange behavior with an expected power decay of κ-5/3 was identified in the data.5 (for -2<ζ<0). 1994).A second data selection criterion used in the analyses of the following sections is based on the correlation coefficients between w and T (hereafter referred to as rwT). In this work we calculate ε using both the structure functions and the spectra of u. (1997). To examine how the conditions in our site diverge from this ‘theoretical’ surface layer. In Section 4. Descriptions of some of these methods are given by Hartogensis (PhD thesis. and how it affects the correlation coefficient between u and w (hereafter referred to as ruw). These correspond to frequencies usually ranging from 0. 10 classes per decade. 93 . we evaluate the flux-variance relationships selecting the data using two main criteria: 0. 4. several methods can be used to estimate ε and CT2.1 to 1. and between CT2 and the T-spectrum.04 to 0.5.3 < rwT < 0. The structure function Duu was also calculated for records where turbulent intensity (Iu = σu \/ u ) do not exceed 0. According to Kaimal and Finnigan (1994. rwT should be approximately 0. Bins were separated to provide classes of wavenumber ranging from 0.2. von Randow et al. pg 19-20) for the surface layer over uniform terrain. and CT2 using T-spectra. 2006) and Albertson et al. following the methodology below: (i) Wind velocity and temperature spectra were calculated for every 1-hour record selected after the data screening described in Section 3.35 in the textbook surface layer (Kaimal and Finnigan. approximately ranging from κ=0.5 and rwT > 0.
The structure functions scale smoothly with r. For CT2.3.9).12 it is expected that the slope of these regressions is approximately -5/3. The vertical velocity skewness (Skw = w' 3 / σw3 ). and the regressions log(Duu) × log(r) yield ε estimates with uncertainty usually below 3 %. Skw changes sign. which can represent the vertical component of the turbulent transport of wvariance (∼ w' w' 2 ). largely due to the enhanced importance of turbulent transports of eddy fluxes and variances in the RSL near the canopy. Fitzjarrald et al. we propagated the standard errors of the intercepts and found that the uncertainty in ε estimates from the power spectra can be up to 25 %. 1981. 94 . the intercept of the regressions allow ε and CT2 to be calculated. the propagation of errors gives 10 – 20 % uncertainty. giving much more reliable estimates. 4. since the calculations are very sensitive to errors in the regression intercepts. A similar approach is performed with a linear regression log(Duu) × log(r) (see Equation 4.(iii) Using the averaged values of a spectrum in the identified inertial subrange..8 and 4. Although the estimations from the power spectra and from the structure functions agree. For a first order error estimate.3 Roughness sublayer height There are observed differences between the plain turbulent surface layer and the roughness sublayer (RSL). At some level above the canopy. (iv) To ensure quality of the calculations only data when the calculated slopes were within ± 10% of the expected value were accepted. it is important to examine the accuracy of these estimations. From Equations 4. 1981). is typically observed to be negative in the RSL and positive in plain surface layers (Raupach and Thom. 1990). a linear regression between log(Su) and log(κ) and between log(ST) and log(κ) was determined. and it has been suggested that this level should be regarded as the height of the RSL (Raupach and Thom.
These results indicate that MOS holds only under a few strict conditions. against ζ.5 h – 1.1 a plot for the temperature structure parameter CT2.1 includes only data with high rwT correlation (rwt > 0. At our site. at the height of 1. even after the selection of high rwT. For the scalars. (2000) measured the vertical velocity skewness within and above the forest canopy at a site 11 km away from our site. fT(ζ) and fTT(ζ) proposed by De Bruin et al. (1993) and fq(ζ) adjusted using fq(ζ) = fT(ζ) q*/T* as discussed by Hill (1989). we first plot in figure 4. up to a level of 1. It is expected that 95 .4. To test the validity of the functions. 4. 4. to Skw ∼ 0 at 1. we measured Skw ∼ 0. and found values ranging from Skw ∼ -1 close to the canopy top. and at the height we are measuring it is likely that the RSL effects are small. q* for latent heat flux). The data presented in figure 4. The solid lines are the empirical functions: fw(ζ) proposed by Holtslag and Moeng (1991). we include in figure 4. the scatter in the figures become even larger. From these results we conclude that the height of the RSL is around 1. however.1 the standard deviations of w.3) to determine u* and x* (T* for sensible heat flux. calculated from the eddy flux measurements on the K34 tower.5). As an additional test for the application of MOS to estimate the fluxes in Amazonian forest.1.1 during daytime.Kruijt et al. T* and q*. some scatter still remains.3 < rwt < 0. If we include the data with lower correlations (0.8 h at this forest.5 times the canopy height.5). where the forest is similar to the one studied by Kruijt et al (2000).4. It can be seen that for σw the empirical function fits the data very well. T and q scaled by u*.8 h. Results 4. The plot of the functions proposed in the literature for σu / u* (not shown) resulted in poor agreement and the measurements hardly seem to behave as one function of ζ only.5 h (where h is the height of the canopy). Flux-variance method – MOS scaling To estimate the surface heat fluxes using the flux-variance method we use the functions fx(ζ) (Eq.
Selfcorrelation. and (d) the structure parameter of temperature CT2.the scatter in these relationships propagate into the estimation of fluxes by the flux-variance method. Standard deviations of (a) vertical velocity w. We recalculated σw/u*.1 we performed an exercise suggested by Hicks (1981). also referred to in the literature as spurious correlation. σT/T* and ζ using randomly generated values for u* and T*. (2002) reply to this by stating that the effect of errors in T* should also be included. randomizing “critical parts” of the original data set. When using the MOS scaling in these analyses it should be noted that some degree of self-correlation is present since u* and T* are present on both x. Baas et al. Andreas and Hicks (2002) show how errors in u* might give misleading conclusions about different scatter in plots of momentum and temperature gradients against ζ. 1981. Hicks. Johansson et al. To test potentially spurious correlation in the plots of figure 4.1. Figure 4. makes the process of relating MOS scaled variables to the stability parameter ζ highly susceptible to errors (Hicks. (b) air temperature T and (c) humidity q. measured over Amazonian 96 . 2006).. 1978.and y-axes of the plots.
40 for figures 1a and 1b.45 and 0. the residual value for σw/u* vs. 4. with points that appear almost unrelated to any function. As u* and T* are present also in fx(ζ).13) for the measured values and the curves (proposed in the literature) in figure 4.1a and 4. scaled by the conventional scaling parameters. respectively. then a first estimate of u* and T* is determined. ζ. by De Bruin et al. One difficulty of applying this kind of method in complex terrain like in Amazonian forest is that often the iteration processes do not converge to stable values.24 and reaches 7691 for σT/T* vs. however with significantly greater scatter and lower values for near neutral conditions. and the same σT curve scaled for humidity (multiplied by q*/T*. The plots of σw/u* vs.forest at the K34 site. The residual variance for the original dataset is 0. 4. In those cases the traditional MOS based methods cannot be applied. with the normalized standard deviation increasing with increasing instability. except by using 97 .1b. The plot of σT/T* vs. For the partially randomized data. and so forth. A measure of the scatter in the figures might be estimated by 2 Sx = ∑( d i =1 n i min )2 ( 4. σT and σq and the functions fx(ζ) (Eq. ζ with the random data gives much more variability. as a function of -ζ. σx/x*] to the line [proposed function fx(ζ)] in the figure. The solid lines represent the empirical curves proposed by Holtslag and Moeng (1991) (σw). To estimate a measure of the scatter we calculated the ‘residual variance’ (Equation 4.1a. ζ is 5. This iteration process should be executed until the estimated values converge to a stable result. which will then be used to estimate new values of u* and T*. Those estimates are used to calculate a new value for ζ.3) to determine u*. the equations should be solved iteratively: usually an initial value of ζ is given. For the estimation of heat fluxes using the traditional flux-variance method we then use the measurements of σw. ζ with the randomly generated data (not shown here) do slightly resemble the plot of figure 4. This shows that the functions for fx(ζ) proposed in the literature and plotted in fig. 1989). T* and q*. (1993) (σT and CT2).1 represent a reasonably good fit to the measured data and are not greatly affected by spurious correlation. Hill.13 ) n−2 where dmin represents the shortest distance from a point [ζ.
2 open symbols represent all the data with rwt > 0.5 greatly improves the estimation of the heat fluxes.5 is represented by closed symbols. which largely reduces the applicability of methods based on local similarity relationships.2 0.1129x .4 w T (eddy covariance) Figure 4.5 (closed symbols). The data used in these analyses were collected in the period May to December 2001.2. Selecting the data for rwt > 0.3 0.9124 0. Comparisons between the heat fluxes estimated by the flux-variance method against the fluxes measured by the eddy covariance system for data with 0. this does limit the range of applicability of the method in complex sites.3.2 0. approximation methods such as the ones proposed by De Bruin (1993) or Hartogensis and De Bruin (2005) which eliminate the need for the iteration.3 0.2 we present the results of the flux-variance method. It is clear that a much larger scatter results from data with relatively low rwt.1 0.3 < rwt < 0. 4. In fig.5 (open symbols) and data with rwt > 0. as can be seen by the regression line plotted in figure 4.0.2.y = 1.5.4 u* T* (flux-variance) 0. However. the more restricted criterion of only data with rwt > 0. The regression line is calculated using only data with rwt > 0. In figure 4. where conditions outside this range may be often 98 .1 0 0 0. The fluxes measured from the eddy covariance are plotted on the x-axis and the estimated flux using the fluxvariance relationships on the y-axis.0047 2 R = 0.
According to (4. Though we estimate errors in ε to be small (< 3%). especially for σq and CT2. with comparable 99 . All the functions Fx(ζ) seem to be constant.4. the uncertainties of the flux measurements propagate into the calculated scaling parameters. RiG might represent a good account of atmospheric stability avoiding the pitfalls associated with spurious correlation in plots against ζ (Hicks. we simply use a multiplicative factor of 1. and new scales for temperature (Tε) and humidity (qε).3 we plot σw. It is very interesting to note that almost all of the systematic variation with stability (ζ) is taken out by changing the velocity scale.observed. 4. as a function of -ζ.2. we test the use of the ‘dissipation velocity’ uε = (kzε)1/3 to scale the standard deviation of w.6) the temperature and humidity scales should be proportional to the heat fluxes. To investigate whether the interpretation from the plots in figure 4. an analysis of the scale dependence of rwt will be presented to give better insight of the influence of the low-frequency motions and the implications for this type of MOS-based estimation. but we believe that part of it is due to the inherent uncertainties of the measurement of these variables themselves.3 may also be affected by spurious correlations we also analyzed the dependence of the normalized variables upon the gradient Richardson number (RiG. we simply define Tε ≡ − w' T ' uε and qε ≡ − w' q' uε then evaluate how these parameters normalize the standard deviations.5. Standard deviations scaled by dissipation velocity As an alternative to the conventional MOS scaling. 1994. σT. σq and CT2 normalized with the alternative scaling parameters. With the main objective of evaluating whether these parameters successfully scale the variables.6. that is. 14). In figure 4. Further in Section 4. Kaimal and Finnigan. and inversely proportional to uε. but the multiplicative factor is still unknown. 1981).3. defined using Equation 4. It is noted that the scatter on these relationships is large. pg. All the plots of the normalized variables against RiG (not shown here) give similar behavior to the plots in figure 4.
3. We conclude that the effects of spurious correlation can also be considered to be small in these analyses. as a function of -ζ.60 r2 = 0. measured over Amazonian forest at the K34 site. Since the standard deviations are more easily measured than ε. (b) air temperature T and (c) humidity q. a simple parameterization of the dissipation velocity and scalar scales can be proposed using a linear relationship with the standard deviations. The linear regressions of type y = a x obtained with the data collected during 2000 at K34 site give the following expressions: uε = 0.15 ) 100 . Standard deviations of (a) vertical velocity w. this provides a convenient practical parameterization for the velocity scale. normalized by the scaling parameters based on ε (see text).14 ) ( 4. scatter.Figure 4.59 ( 4. Having found no dependence of the scaled standard deviations on stability.03 σT r2 = 0.59 σw Tε = 1.
the scatter around the similarity relationships is increased in these cases.1). and it can be seen in figure 4. might be useful to describe the structure of turbulence in the surface layer over this Amazonian forest site. such as the limitation of the performance of these methods to conditions of relatively high w-T correlation. As a tool to evaluate how the scale variability and the low-frequency motions influence the correlation coefficients we use multiresolution decomposition to project the turbulent signals into temporal scales and assess the contribution of motions of different scales to the “total” standard deviations and covariances.5 (for -2 < ζ < 0) and ruw ~ -0. 4. The advantage of the wavelet decomposition of the turbulent signals is that it “locally” decomposes the variances and covariances indicating the scale of the dominant (local) fluctuations in the signals.The regression for the humidity signals gives a very poor coefficient of determination (r2<0. In this section we show that these deviations are characterized especially by the influence of motions of (slow passing) large eddies of scales that may span the whole boundary layer. The correlation coefficient between two variables is determined by the ratio between their covariance and the product of their respective standard deviations. apart from some scatter. Multiresolution is a type of wavelet transform that performs a simple orthogonal decomposition of the records into averages on different time scales (Katul and Parlange.35 (-1 < ζ < 1) (Kaimal and Finnigan. Howell and Mahrt. 1997). while the traditional Fourier decomposition 101 . Our measurements above Amazonian forest frequently deviate from these values. as shown in the previous sections. There are still some issues that need to be addressed. flat terrain give rwt ~ 0.3c that the scatter for humidity signals is large. The results illustrate that the scaling parameters based on the dissipation velocity uε. We examine this issue in the next section. and. 1994. Low frequencies and correlation coefficients Measurements in the surface layer over uniform.5. 1994).
4 shows that most of the variability of w occurs on motions at scales from 10 to 300 m (corresponding to time scales on the order of 1 min). The composite records follow a time step of 1 hour. After the application of the Haar wavelet the curve of partial contribution of each scale band to the variances of w. (2002). 2002). As the objective of this study is to evaluate the contributions of motions on scales of up to 1 hour. For example. and so forth. u and T and to the covariances of w-T and u-w was determined for each record. It is interesting to note that on average the contributions 102 . whilst for σu2 and σT2 the contributions increase with the scale. We apply the Haar transform to the turbulent signals measured at the K34 site in the period from days 286 to 302 of year 2000. to indicate the time scale of the processes. Approximate time labels are also included above the x-axes.4 and 4. The x-axes in these figures show the spatial scales.33 (Taylor’s hypothesis validity range. where turbulent intensity (Iu = σu / u ) did not exceed 0. Katul and Parlange (1994) assessed the influence of different wavelet basis functions on the spectral and statistical behavior of turbulent signals and concluded that. the next from 11:00 to 15:00 h.. (2002). the wavelet decomposition is more suitable for non-stationary signals. (2002) also observed similar results for different wavelets. similar to the method applied by von Randow et al. the overall trend for all wavelets is essentially the same. for the type of analyses used here. there is no unique basis function for the orthonormal wavelet transforms. von Randow et al. von Randow et al. to some extent. one record includes the data collected from 10:00 to 14:00 h. Still.5. For that reason we chose to use the simple Haar basis function. This methodology was chosen to smooth the random variations within one scale band and. following a similar methodology to that used by Katul and Parlange (1994) and von Randow et al. Unlike Fourier transforms. so the slopes of inertial subranges are preserved. Figure 4. it is advisable to perform the wavelet analyses on data records at least 4 times longer.provides the principal periodicities. give more weight to the lower frequencies in the analyses. These curves were further averaged and plotted in figures 4. the wavelet preserves the exponent of any power-law segments. Therefore. which are estimated from the temporal scale values using the average wind velocities and the assumption of Taylor’s Hypothesis. Composite records of 4-hour length were thus prepared with 87 runs being selected during daytime unstable periods.
(2002) and Malhi et al.5. It is observed that the peak of contributions to the covariances occurs at scales of 100 – 500 m. It is likely that these (slow) large-scale fluctuations are caused by large eddies that cover the whole boundary layer. Similar assessment using Fourier analysis has been done by other authors (e. 4. We calculate the correlation coefficient r(m) at scale index m by rwt(m) = covwt(m) / σw(m) σT(m) ( 4.for σu2 increase almost linearly (on a log scale) with the spatial scale. It can be seen that the average line and dispersion are increased at scales larger than 1000 m due to few particular periods that present high values at those scales.5.16 ) 103 . 4. According to McBean and Miyake (1972). which presents the σT2 spectral curves for a number of 4-hour records. (2004) obtained similar results at a site in south west Amazonia. We thus included the panel seen in the top left of the figure. g. large roll vortices and local circulations induced by topography or surface heterogeneity. the spatial scales higher than 1000 m dominate. The contributions of different scales to the covariances wT and uw (cospectra) are presented in fig. especially for σT2. although the average curve indicates close to zero contribution on scales of few kilometers (time scales from 30 min to 1 hour). such as. These low-frequency processes can include deep convection. As in fig.. and.4. McBean and Miyake. These results emphasize that for the larger scales. Combining the scale covariances and standard deviations (calculated from the variances). which corresponds to time scales on the order of 1 to 5 min. von Randow et al. individual contributions at these scales can be very important. the spectral correlation coefficients can be considered to be a measure of the transfer efficiency as a function of scale. being of either sign (von Randow et al. 1972). The covariance uw is negative (as momentum is transferred downwards from the atmosphere to the surface) but is presented in this figure multiplied by -1. 2002). as large as – or even larger than –contributions from ‘turbulent’ scales. a panel showing a set of individual curves of scale contributions to the covariances wT is included in figure 4. A disadvantage of presenting only one average line is that it may not clearly show aspects of individual cases. the individual records most influencing the dispersion. we can estimate a hypothetical “scale correlation coefficient” (extending the concept of correlation coefficient to a scale-dependence assessment). for example.
u and T.5. Average scale contributions to the covariances covwT and covuw (absolute values). measured using the wavelet projected signals over Amazonian forest at the K34 site. The error bars are the standard errors. Average scale contributions to the variances of w. Error bars indicate the standard errors. measured using the wavelet projected signals over Amazonian forest at the K34 site . Length and time scales are indicated on the x-axis. Figure 4. 104 . The small panel in the figure shows w-T cospectral curves for a number of records to better illustrate the scatter. The small panel in the figure shows the spectral curves of σT2 for a number of records to better illustrate the scatter.Figure 4. Length and time scales are indicated on the x-axis.4.
6 we present the results of the scale dependence of rwt and ruw. so is ruw. measured from Equation 4.5).4 and 4. Relative scale contributions to the correlation coefficients rwt and ruw at the K34 site. like the spectra and cospectra. As covuw is negative. This behavior is an indication that the influence of the low-frequency motions in the calculated correlation coefficients is erratic. Figure 4. we can see that on scales higher than 1000 m there is a very large variability.6. the area below the curves represents the actual values of the total variance and covariance.6. It is very useful. The correlation ruw(m) is calculated in a similar way. though. with values ranging from negative to positive values. possibly enhancing or reducing the values in a complex way. However. but it is presented in the figure multiplied by -1. 105 . Note that in case of the scale variances and covariances calculated from the wavelets (figures 4. In figure 4. Looking at the individual curves of rwt(m) in the small panel on the top left of figure 4.where (m) is used as a scale index (not to be confused with power m).16. that this analysis provides a qualitative assessment of how the variability of w and T (or u) in a particular scale band correlate with each other (similar to a coherence spectrum). this is not the case for the scale correlation coefficients.
Note that in this case.1 0.01 0.1 15 min 1 min 1 sec 0.6 0. Hence the measurements in low-wind regions. To better investigate this hypothesis we plot in figure 4. Furthermore.7 a number of lines showing the calculated rwt depending on the record length used. the relative importance of the low-frequency motions increases in low wind conditions (von Randow et al.3 0.1 1 10 1h 100 record length (minutes) Figure 4.Although the effects of the low-frequency motions might sometimes influence the correlations to enhance the calculated values.4 rwt 0. such as many tropical sites.7 0. when it tends to decrease as longer of (low-frequency) large eddies compared to the smaller ones. the effect of including the low frequencies will be a reduction of rwt.2 0. the surface layer indeed presents characteristics slightly different to the ‘theoretical’ surface layer observed over uniform terrain (which is shown by reduced correlation w-T). timescales are included.7 that. 106 . we expect that in many cases. 2002).7 represent general conditions of moderate to strong instability. Some near neutral conditions that give very low rwt values were excluded from the plot for clarity. for moderately and strongly unstable conditions measured at the K34 site. These results indicate that when there is a relatively large influence 0. Timescale dependence of the correlation coefficient rwt.5 0. and not only on a particular scale band around that timescale as is the case of the previous figures. The cases presented in figure 4. indeed. We can see from figure 4. in general there is an increase of rwt with increasing the record length until around 10 minutes. are particularly prone to these complications..7.001 30 min 0.0 -0. the values will represent the correlation coefficient calculated including all the time scales up until the record timescale.
The results from previous sections demonstrate that MOS theory performs reasonably well only for data where the correlation rwt is higher than 0. Further. McNaughton proposes that the results usually attributed to the local action of stability are in fact due to the action of larger eddies in the convective boundary layer.5 value are characterized by the presence of motions of (low-frequency) large eddies.5. A new view of how the outer-layer convection modulates the processes near the ground was proposed by McNaughton (2004.6. with the friction velocity u*. McNaughton and Brunet. 1994). it gives similar predictions to McNaughton (2004) theory. it is not difficult to think of a number of situations where MOS will not work. For example where flow over heterogeneous areas produces CBL turbulence whose power is unrelated to the local surface heat flux. modulating the surface layer and causing deviations from the expected theory (e. the turbulence in the surface layer is related only to local parameters (represented by the parameter ζ). it is interesting to investigate the relation between the parameter v* (“fluctuating” friction velocity).35 in the surface layer over flat uniform terrain (Kaimal and Finnigan. The relative success of MOS theory may be explained by the fact that in some conditions. 2006). It can be observed that when v* is of comparable magnitude to u* or higher (v*/u* ≥ 1) the correlation ruw is clearly lower than the usually observed value of (-)0. 2002).g. that challenges the MOS model at its conception. Discussion In the general framework of MOS theory. In figure 4. however. introduced by McNaughton (2006) in the new structural model of the surface layer.4. It can be estimated from measurements of ε and u* (Equation 4. there are indications that the deviations of rwt from the textbook 0. 2006). 1990. As shown. Many results from recent studies present. However.5). evidence that (low-frequency) large eddies can largely influence the turbulent processes. 107 .8 a relationship between the correlation coefficient ruw and the ratio v*/u* is shown. Note that ruw is negative and we are presenting it as positive values here. Recall that v* is a parameter that represents the additional energy transported down from the variable motions of the outer layer to the surface layer (McNaughton. Högström. especially highly convective conditions.
In figure 4. the relationship presented in fig.0 0. Zhuang (1995) presented evidence that while heat is quite efficiently transported within updrafts of convective plumes. the extension of the same ideas to the mechanisms of scalar exchange should still be carefully examined. Relation between the correlation coefficient ruw (absolute values) and the ratio v*/u*. Therefore.5 1.5 v * / u* Figure 4. We believe this result is explained by the difference between the transport mechanisms of momentum and temperature.8. Momentum is mostly transported by a mechanism of pressure redistribution of turbulent kinetic energy. 4. this is not the case for momentum.5 2. 2006) provides a foundation for a new similarity model of momentum exchange in the surface layer. However.0 0. we cannot directly assume that the dissipation of temperature variance would be affected by the outer-layer modulation in the same way as ε.4 ruw 0.5 0.0.9 we present the histogram of the ratio v*/u* for a dataset gathered at the K34 site from May to December 2001 (unstable conditions only).0 3.2 0.0 1.1 0. A similar plot of rwt versus v*/u* was made (not shown here) but a decrease in rwt with increasing v*/u* was not clearly visible. Nevertheless. measured over Amazonian forest at the K34 site.3 0.8 support the interpretation of the results discussed previously: the low-frequency motions indeed modulate the structure of the surface layer and induce deviations from the ‘classic’ surface layer over Amazonian forest.0 2.5 3.6 0. The general scheme proposed by McNaughton (2004. It is observed that v* is higher than 108 .
5 2.0 3. which.4. Histogram of ratio v*/u*. Analyzing the kinetic energy budget based on these views.9. This is consistent with the fundamental proposition of McNaughton’s structure model that the local turbulence is insensitive to local stability. with clear implications to any estimate from indirect methods such as the flux-variance . One interesting result appears when we plot the standard deviations and structure parameter scaled by the ‘dissipation velocity’ (and respective scalar scales) against ζ: almost all of the systematic variation of the variables with stability is taken out by the use of the new scales. so buoyancy has no direct local effect on turbulence structure (McNaughton. The explanation given by the author is that this insensitivity is a result of the self-organizing nature of the turbulence structure: the energy produced locally by buoyancy is redistributed through the whole flow in such a way as to maintain a large-scale organization of the whole.5 1. of occurrences 100 80 60 40 20 0 0. It is also interesting to further discuss the results of Section 4.or models of flux exchange that rely on this theory. During these cases we expect MOS theory to be invalid.2. although increasing the energy of the turbulence – by changing the velocity scale – maintain the general structure unaltered. McNaughton (2006) shows that the tendency of buoyancy to enhance the vertical motions (and modify the structure of eddies) should be opposed by pressure reaction forces. or of comparable magnitude to u* for roughly 20 % of the time.5 v */ u * Figure 4. 2004).140 120 N.5 3. 109 .0 2.0 1.0 0.
. x* for a scalar x). the variability of humidity appears to be highly influenced by the particular moisture regime (Mahrt. Nonetheless 110 . 2002).the coefficient of determination was very low (r2 < 0. The functions fx(ζ) (Equations 4. but the scatter around these functions is relatively large. we also compared the temperature structure parameter CT2 with a function fTT (ζ) (Eq. affecting the variance and skewness of humidity.1 and 4. They emphasized that this result might also be due to entrainment of dry air from the top of the boundary layer reaching down to the surface.4)).1) and as seen in figure 4. derived in the surface layer over uniform terrain. the scatter was too large. As an additional test for the application of MOS based relationships to estimate the fluxes. 2006). frequently characterized by strong convective motions and a large heterogeneity of the surface. in a relatively homogeneous (dry) area in the plain of La Crau (South of France). These authors presented empirical evidence of the influence of entrainment of dry air from the top of the boundary layer reaching down to the surface layer.Another result shows that.3c.7. De Bruin et al (1993) also found that MOS scaling did not work for water vapor measurements and measured correlation coefficients between water vapor and temperature significantly smaller than 1. von Randow et al. for the humidity signals it was not possible to find a relationship with a humidity scale qε . based on using the dissipation rate as a scaling parameter (McNaughton. The functions are in agreement with the data. This result is related to the results obtained by von Randow et al. In the case of the Amazonian boundary layer. 1991.2) that according to the MOS describe the standard deviations of the wind components and scalars scaled by their respective scaling parameters (u* for wind components. (2002) regarding the variability of humidity in south west Amazonia. based on MOS scaling parameters. 4. Conclusions In this work we explore the application of the flux-variance method for unstable conditions at a forest site in central Amazonia (“Manaus K34” site). (4. 2004. and an alternative scaling approach. were compared with the data collected over our site.
the vertical wind variations tend to be not correlated with variations in the horizontal wind speed and in the temperature. where these conditions may often be observed. and the scaled variables become independent of ζ. The results indicated that. and new scales for temperature (Tε) and humidity (qε). The structural model proposed by McNaughton (2004) provides a step forward in this. for which the MOS theory was developed. It is therefore not surprising to see that MOS based relationships give a large scatter. when the influence of these low-frequency motions is large. we test the use of the ‘dissipation velocity’ uε = (kzε)1/3. As a tool to evaluate how the scale variability and especially the low frequencies influence the correlation coefficients we used wavelet analyses to project the turbulent signals into frequency (and spatial) scale classes and assess the contribution of motions of different scales to the “total” variances and covariances. 111 . in an attempt to explain how the low-frequency motions modulate the turbulence in the surface layer.the flux-variance method gives reasonable estimates of the sensible heat fluxes when we select only data with relatively large w-T correlation coefficient (rwt > 0. Especially with its frequent strong convective systems. 2004). This indicates that. proposed by McNaughton (2006). As an alternative to the conventional MOS scaling. the surface layer cannot be characterized by the ‘theoretical’ surface layer observed over uniform terrain. This limits the range of applicability of these methods in complex sites. the parameterization of the ratio v*/u* indicates that during roughly 20 % of the time the unstable boundary layer deviates from the ‘classical’ description. to scale the standard deviation of w. the influence of low-frequency processes in the surface layer tends to cause the correlation coefficients rwt and ruw to decrease. This result is consistent with the self-organizing nature of the turbulent structure (McNaughton. More detailed understanding of how the turbulent processes in the boundary layer can affect the turbulent transport near the ground is still highly desirable to improve the methods of measurements and modeling in the surface layer.5). on scales larger than a few hundred meters. The systematic variation with stability is taken out by the use of the new parameters. In general our results highlight the complexities of the surface layer above vegetation like Amazonian forest. The flux estimates are much more scattered when we include data with lower w-T correlations. In this sense.
but give different values in non-stationary conditions (Culf. with the running mean influenced by the lower temperatures in the preceding hour. For the K34 site. errors in the measurements. it is likely that the conditions for that record are not appropriate for the flux calculations (this could be. Records that give larger factors are excluded from the analyses. 2000). for example. If the difference between the fluxes calculated using the RMF and the BA is large. etc). also referred to as “block average” (BA). 112 . For a practical criterion to quality check our eddy covariance calculations operationally. sensor drifts.1 Hz data for vertical velocity (w) and air temperature (T).Appendix A There are different ways to separate the ‘average’ and ‘fluctuating’ components of turbulent signals in eddy covariance calculations. In the top and mid panels the running mean of time constant of 800 s and the hourly block average are plotted alongside 0. For illustrative purposes we present in figure A1 one example of the effect of the two different calculations. Two common approaches used are the autoregressive “running mean filter” (RMF) and a simple removal of the mean. The RMF shows a clear ‘memory effect’ for the temperature signal in this example. The two methods agree well in conditions of stationarity. The product w’T’ calculated using the two approaches is plotted cumulatively in the lowest panel. conditions of large non-stationarity. we define an ‘uncertainty factor’ simply by the difference between the RMF and BA calculations. This value is somewhat arbitrary but was found to appropriately filter out relatively large non-stationary and problematic records. we chose a maximum acceptable value of 40 W m-2 for the ‘uncertainty factor’ for sensible heat fluxes. for one selected hour during daytime.
0 1.0 -0.0 -1. for a period of one hour.5 0 10 (c) Cumulative w'T' 20 30 40 50 60 RMF Cumulative w'T' (m s-1 K) 800 600 BA RMF 400 200 0 0 10 20 30 40 50 60 Time (minutes) Figure A1. Data for vertical velocity (w.5 300.0 BA 300. air temperature (T. 113 . middle panel) and cumulative w’T’ (bottom panel).5 0.5 -1.0 0 302.0 299.5 302.2.5 -2. illustrating the calculations using an autoregressive running mean filter (RMF) of 800 s time constant and the block average (BA) of the whole record.0 (a) Vertical wind velocity Running mean filter (RMF) w (m/s) 0. top panel).5 1.0 301.5 10 (b) Air temperature 20 30 40 50 60 Block average (BA) T (K) 301.
115 . 2007. we found that its closure improves when data with increasingly higher rwT are used. the results suggest that rwT can be used as an indicator of the importance of low-frequency motions in the surface layer. Holtslag. Submitted to Agricultural and Forest Meteorology. M. Evaluating the energy balance for different ranges of rwT. Kruijt. B. Exploring eddy covariance and scintillometer measurements in an Amazonian rain forest. and these fluxes are compared with an eddy covariance system (EC) to analyze conditions of low-frequency modulation in the surface layer. von Randow. less so for 30-min-averages. The differences between measured T* (and related heat flux) by EC and estimated from LAS measurements and Monin Obhukov Similarity (MOS) relationships are observed to increase with both increasing non-stationarity conditions and decreasing correlation between vertical wind and temperatue (rwT). Generally. A. M. ____________________ This chapter is submitted for publication as C.Chapter 5 Exploring eddy covariance and scintillometer measurements in an Amazonian rain forest Abstract A Large Aperture Scintillometer (LAS) is used to estimate the surface sensible heat fluxes in an Amazonian rain forest site. B. The results show that the flux estimations from the EC are often lower than from the LAS. The results are attributed to the spatial averaging effect of the LAS. de Oliveira. we observe that the largest differences between the LAS and the EC fluxes are found for 10-min-averages. L. Using different averaging times on EC calculations. while 1-hour averages give the smallest differences. A.
the longer the averaging time chosen. typical averaging time values used are 30 or 60 minutes. The time scale should be long enough to sample a sufficiently large number of eddies passing by the sensors to provide stable statistics of the exchange.. Finnigan et al. In the literature. however. the bigger the uncertainties in the flux calculations will be. Kruijt et al. 2002. 2004). In certain conditions. or correlated trends in the signals that should be disregarded in the flux calculations. However. it is not always clear to what extent these frequencies represent actual transport that should be accounted for to evaluate the total exchange in budget studies. just as in several other complex terrain sites in the world.. using the eddy covariance (EC) technique. especially due to the heterogeneous aspect of the topography in the region (terra firme plateau areas dissected by valleys with very wet soils). there are indications that the low-frequency variations contribute up to 30 % of the total covariance in Amazon forests (Von Randow et al. The limitations and uncertainties in flux measurements over complex surfaces have been discussed in recent papers (e. located about 60 km north of Manaus. 1973). due to non-stationarity of the signals. it is therefore not easy to choose an averaging time scale to define the ‘mean’ and ‘fluctuation’ parts in eddy covariance calculations. 2003).g. these time scales might be too short to properly sample slow moving large turbulent structures. In some cases.1. However. Brazil. however. the flux measurements are still subject to large uncertainties. the energy and carbon fluxes between the surface and the atmosphere have been measured in a rain forest area in the Central Amazon (the LBA K34 site). In the presence of slowly passing eddies. Unfortunately. The atmospheric boundary layer over Amazonian forest frequently presents slowlymoving large eddies caused by strong convective motions and/or local circulations induced by the heterogeneity of the surface.5. it is still difficult to quantify their effects on the turbulent exchange processes in the surface layer. Another 116 . These motions occur at time scales longer than the usual turbulent time scales and are here referred to as low-frequency motions.. Mahrt. Finnigan et al. 2003. Introduction For several years now.. 1998. One example of these conditions is elongation of eddies in the downwind direction and formation of roll vortices (LeMone.
which is mainly dependent on fluctuations in air temperature and humidity.. 2004). The scintillations recorded by the instrument can be related to the structure parameter of the refractive index of air (Cn2). Unfortunately. In this work.. 1992.g. analyzing the effects of low-frequency motions on sensible heat fluxes. one disadvantage: to estimate heat fluxes from the measured Cn2. 2006). such as the K34 site in central Amazon (von Randow et al. 1995) using the Monin-Obukhov Similarity (MOS) theory.. and the averaging time scale can be reduced. De Bruin et al. a similarity relationship between the two quantities is necessary. The scintillometry method has. Hill et al. 2002). 1989. We compare the LAS measurements with the data collected by an eddy covariance system in the same location. Nevertheless. the MOS theory is not always appropriate in more complex sites. Measurements using different types of scintillometers have been successfully made over relatively uniform sites (e. A scintillometer is an instrument that can measure the amount of scintillations in the air by emitting a beam of light over a horizontal path of the order of a few meters to 10 km (Hill. a combination of a scintillometer and an eddy covariance system might provide improvements to measure surface fluxes modulated by low-frequency variations. we present measurements of a large aperture scintillometer (LAS) recently installed at the K34 site. so the uncertainty of scintillometer estimations might be large. The main advantage of the scintillometer is that it provides a measurement that represents a spatial average of the turbulent eddies along the whole path.. 1992. For this reason. An alternative method to estimate the surface heat fluxes that may have advantages over eddy covariance in these conditions is the scintillometry method. Andreas. 117 . however.example is inhomogeneity in surface heat flux or topography that may cause convective cells to lock into fixed positions on the landscape (Malhi et al. De Bruin. it is not necessary to use a long time scale to sample a large number of eddies.
5.2. Theoretical background
5.2.1 Fluxes using scintillometry
The estimation of fluxes with the scintillometer is based on the measurement of the fluctuations of intensity of an electromagnetic radiation beam, also known as scintillations. Detailed descriptions of the measurement principle and of different types of scintillometers are found in many papers (e.g., Wesely, 1976a; Andreas, 1989; Hill, 1992; Hill, 1997). A Large Aperture Scintillometer (LAS) consists of a transmitter and a receiver, installed at a certain height zLAS. The relationship between the measured variance of the logarithmic intensity fluctuations σlnI2 and the structure parameter of the refractive index of air Cn2 is obtained from the equation of propagation of a spherical wave through a medium with random refractive index fluctuations. For a LAS with equal transmitting and receiving apertures, Wang et al. (1978) derives
2 2 C n = 1.12σ ln I D 7 / 3 L−3
( 5.1 )
where D is the aperture diameter of the LAS and L the distance between the transmitter and the receiver. The scintillations are primarily the result of fluctuations in temperature and air humidity. Strictly speaking, the measured Cn2 value is related to the structure parameters of temperature CT2, of humidity Cq2, and of the covariant term CTq. For electromagnetic waves in the visible and near-infrared region, however, humidity related scintillations are much smaller than temperature related scintillations. Wesely (1976b) argues that, for a LAS operating at a near-infrared wavelength, we can estimate the structure parameter of temperature CT2 from the measured Cn2 using
⎛ ⎞ T2 ⎜ ⎟ ≈ Cn ⎜ −6 − 0.78 ⋅ 10 P ⎟ ⎝ ⎠
⎛ 0.03 ⎞ ⎟ ⎜1 + ⎜ β ⎟ ⎠ ⎝
( 5.2 )
where P is the atmospheric pressure, T is the absolute air temperature and β is the Bowenratio, the ratio between the sensible and latent heat fluxes. The Bowen-ratio term in this expression provides a correction for humidity related scintillations. Once CT2 is known, the sensible heat flux (H) can be derived from a universal function fT (z-d/LMO) that is based on Monin-Obukhov Similarity theory (MOS).
CT ( z LAS − d ) 2 / 3 T*
⎛z −d ⎞ ⎟ = f T (ζ ) ( LMO < 0) = f T ⎜ LAS ⎜ L ⎟ MO ⎝ ⎠
( 5.3 )
where d is a zero-displacement height, zLAS the effective height of the scintillometer beam above the surface, along the path (Hartogensis et al., 2003), LMO is the Obukhov length, and T* is a temperature scale defined as
T* = −H ρ c p u*
( 5.4 )
In the latter equation, ρ is air density, cp is the specific heat of air and u* is the friction velocity. In this work we adopted fT (ζ) = 4.9 (1 – 6.1ζ)-2/3 , (ζ < 0), after Andreas (1989). Note that the LAS provides only CT2 and a measure of u* is necessary. We use wind speed measurements and the following flux profile relationship to estimate u*:
⎛ z kU = ln⎜ ⎜z u* ⎝ 0 ⎞ ⎛ z ⎟ −ψ m ⎜ ⎟ ⎜L ⎠ ⎝ MO ⎞ ⎛ z ⎟ +ψ m ⎜ ⎟ ⎜L ⎠ ⎝ MO ⎞ ⎟, ⎟ ⎠
( 5.5 )
where ψm is the integrated stability function for momentum ⎧ ⎪ − 5ζ ⎪ ⎪ =⎨ ⎪2ln[(1 + x ) / 2] + ln (1 + x 2 ) / 2 − 2arctan( x ) + π / 2 ⎪ ⎪with x = (1 − 16ζ )1 / 4 ⎩
for ζ > 0 , for ζ < 0 ( 5.6 )
known as Businger-Dyer relation. Equations 5.3-5.6 should then be solved iteratively to derive momentum and heat fluxes.
5.2.2. Footprint estimation
For a comparison of the fluxes estimated by the LAS with the EC, it is important to evaluate the footprint of the two systems. Some scatter may be introduced in the comparisons if there is a mismatch of the footprints. The footprint function f relates the measured flux, F(x, zm) to the spatial distribution of upwind surface flux S0(x) (also termed source strength), i.e.,
F ( x , z m ) = ∫ S 0 ( x ) f ( x , z m )dx ,
( 5.7 )
where x is the distance upwind and zm is the measurement height. In case of the LAS, the footprint concept is extended to combine with the bell-shaped weighting function that describes the contribution along the scintillometer path (Meijninger et al., 2002). For a simple footprint analysis we use the formulation of Hsieh et al. (2000). Based on a combination of Lagrangian stochastic dispersion model results the authors found from dimensional analysis that x / |L| can be expressed as x −1 D( z u / | L |) P , = 2 | L | k ln( F / S 0 ) ( 5.8 )
where k = 0.4 is the von Karman constant , D and P are similarity constants, and zu is a length scale defined as zu = zm ( ln (zm / z0) – 1 + z0/zm ), ( 5.9 )
where z0 is the roughness length. The ratio F/S0 represents the relative cumulative contribution to the flux at upwind distance x – a 90% contribution is represented by F/S0 = 0.9. Using the results of a Lagrangian model to calculate the footprint for a range of zm, z0 and L values, Hsieh et al. (2000) found D = 0.28 and P = 0.59 for unstable conditions. An expression for the footprint function is
f ( x, z m ) =
1 1− P 1− P ⎞ ⎛ −1 P P exp⎜ 2 Dzu LMO ⎟ . Dzu LMO 2 k x ⎝k x ⎠
( 5.10 )
Figure 5. (1997). Gill Instruments. LICOR.48’’ W. were recorded at a sampling rate of 10. and H2O and CO2 mixing ratios. The landscape consists of plateaus. The height of the vegetation is around 30 m. is equipped with an eddy covariance system (hereafter referred to as EC) measuring the fluxes of momentum. in operation since mid 1999. indicated by EC in figure 5. energy and carbon dioxide at a height of approximately 36 m above the displacement height (53 m above the ground). The calculations of the eddy fluxes were performed for three different averaging time intervals. in the central part of Amazonia. measured by the sonic anemometer.5.3. guarded by the Brazilian National Institute for Amazon Research. Wind velocity components and temperature. The program was configured to apply two-axis rotations to align the coordinate frame with the mean streamlines and force the mean vertical component (w) to zero (McMillen.4 Hz. UK) and a fast-response closed path infrared gas analyser (IRGA) (LI-6262. composed of a three axis sonic anemometer (Solent 1012R2. USA).1. (2002). indicating the position of two towers used in the experiment. Experimental description 5.3. using Alteddy. The main experimental tower. 1983). The 121 . 130 m asl). dissected by valleys that are frequently flooded after rainfall. Detailed information about the site and the instrumentation at the K34 tower are provided by Araujo et al. Brazil. The EC system is similar in design to the system described by Moncrieff et al. Alteddy is in-house developed software written in FORTRAN.1 shows an overview of the topography in the experimental area.1 Measurements and site description The data used in this paper were collected in 2005 from May 21 to June 14 and from September 23 to October 3. The site is located in a pristine rain forest reserve. 10 min. at the field experimental site “K34” (2o36’32.67’’ S. located about 50 km north of Manaus. to compensate for the time delay in the IRGA signals. which can be adapted to a number of different hardware configurations and program options. 60o12’33. 1988). It is a closed-path system. measured by the IRGA. to include cross-wind and humidity corrections of the sonic temperature signal (Schotanus. 30 min and 1 hour.
N LAS on Elevati 13500 EC 200 150 100 50 9000 1 km 9500 10000 13000 12500 12000 11500 11000 10500 10500 UTM . Generally for this site and instrumental setup. The second tower was erected at about 1. connected to a datalogger (CR23X.7 m. in central Amazon. following the methodology described by Moncrieff et al. As the two towers are separated by a valley. Using the 122 UTM - Y Dat a (UT M*) . 2004). * UTM coordinates: add 800000 to X and 9700000 to Y to get actual UTM values. The vertical bars show the position of the towers (where the letters EC indicate the eddy covariance tower) and the dotted line shows the LAS path. A LAS (LAS 150. supported by the two towers. however. Campbell Scientific. the effective height of the LAS beam is greater. Overview of the experimental site K34.1 km north of the main tower. Kipp & Zonen. USA).1.. the corrections are relatively small and do not represent large uncertainty factors in the final values (Kruijt et al. The Netherlands) was then installed at the site.2 m above the ground. Delft. The transmitter was installed at the second tower at a height of 43. The receiver of the LAS was installed at the main tower at a height of 51. (2000). calculations also include corrections for instrument responses and damping of fluctuations through the IRGA tube. apart from the coordinate rotations. (1997) and Aubinet et al.X 11000 Data (U TM*) 11500 10000 12000 Figure 5.
Figure 5. Nieveen et al.1 m above the displacement height. Correcting for Tower Vibrations Ideally both the transmitter and receiver of the LAS should be mounted in robust. being tall and thin. Spectra were then examined and corrected for the effects of tower vibrations and signal absorption (see next section).. but the strong peak in the middle immediately draws the viewer’s attention. Our towers. This is important because the instrument cannot distinguish fluctuations in the signal caused by instrument vibrations from scintillations. contaminating the measurements of the LAS. The frequency of the beam used by the LAS is close to a resonance frequency of water vapour. Clifford (1971) derived the theoretical spectrum form for a spherical wave propagating through a turbulent atmosphere..2 shows a typical 10-min recorded spectrum measured by the LAS at the K34 site. along with the curve for the real component of the theoretical spectrum described by Nieveen et al. values of Cn2. We propose a method to correct for these vibrations. CT2 and H were derived and compared with the values provided by the EC system.method described by Hartogensis et al. so part of the beam’s energy is absorbed by atmospheric humidity.3. The real component represents the fluctuations in the received signals by refraction (equation 4 of Nieveen et al. 1998).2. Measurements of beam intensity fluctuations and of its spectrum were recorded at the LAS receiver for 10 min intervals. by measuring the Fourier spectra of the intensity fluctuations at the receiver. and comparing them with an expected theoretical shape of the spectrum. (1998) fitted to our data. (1998) further described it as a combination of real and imaginary parts. unfortunately are sometimes subject to vibrations. A characteristic levelling off of the spectrum at frequencies ranging from 1 to 10 Hz is observed. After that. (2003). We attribute this peak to the effect of vibrations on 123 . 1998) and the imaginary component represents the effects of absorption (equation 5 of Nieveen et al. stable constructions to avoid vibrations in the set. 5. we estimate an effective height for our LAS of 54.
we applied the following methodology to our LAS measurements: first we recorded the spectrum of intensity fluctuations using the FFT built-in function of the datalogger. 1000 100 S (mV Hz ) 2 -1 10 1 0. The latter feature is particularly important for the application in tropical forests. Spectrum of intensity signals measured by the LAS for day 146. Furthermore. second.1. an enhancement is observed at low frequencies. we calculate a “vibration-free” variance of intensity fluctuations (σI2) from the adjusted spectrum. we calculate a corrected value for Cn2 using equation 5. Note that this approach provides a correction for both artificial vibrations and absorptions. third. we identify in the region between 1 and 10 Hz the level of the frequency-independent part of the spectrum. along with the theoretical spectrum curve fitted to the data (continuous line).the tower. where humidity is high. 124 .2. 14:00 h (dotted line).1 1 10 100 f (Hz) Figure 5. (1998). Similar results were obtained by Nieveen et al. This is attributed to absorption: at the lower frequencies the absorption mechanism enhances the fluctuations of the received signal. and adjust the theoretical curve to match this level. and finally.1 0. using a previously determined relationship between σI2 and σlnI2. Based on these observations.
the LAS is mainly sensitive to temperature fluctuations and not directly sensitive to mechanical turbulence. the application of the LAS in stable conditions is mostly a test of the flux-profile relationships. It is expected that the uncertainty in the LAS estimations are large in stable conditions. Therefore. especially due to the poor performance of similarity theory. 125 . it is observed that in some periods the fluxes measured by the LAS are higher than by the EC.3. where the LAS information only has indirect influence through the stability corrections. In stable conditions the only turbulence generating mechanism is mechanical turbulence. 2005.5. 400 LAS Eddy cov 300 H(Wm ) -2 200 100 0 -100 143 144 145 146 147 148 149 150 Day of year Figure 5. Example time series of half-hourly sensible heat fluxes at K34 site measured by the LAS and the eddy covariance system.4. using fluxprofile relationships. Although the general behavior of the two systems is the same.3 shows a typical time series of half hourly sensible heat fluxes. Moreover. In the following analyses we focus on daytime conditions only. u* must be estimated from a measurement of wind velocity and an estimate of z0. measured by the EC system and by the LAS in the end of May. Recall that the LAS provides only an estimate of T* (from CT2). Results Figure 5.
and that the difference slightly increases with increasing footprint distance. This figure shows boxplots of the measured fluxes according to classes of x (distance of 90 % contribution). according to wind direction. Radius represents the distance upwind (m) at which the relative cumulative contribution to fluxes is 90 %. The radius represents the distance upwind up to which the relative contribution to the flux measurements is 90%. according to wind direction. However. The diamonds and the traced line show the beam of the LAS. in general. we expect that. Observing figure 5. but it is clear that the flux estimations from the EC are often lower than the LAS. Footprint of the eddy covariance system for unstable conditions. figure 5.5 shows that the flux estimates of the two systems often differ. N 3000 2000 1000 W 0 E S Figure 5. figure 5.4.4 shows a polar plot of the footprint of the EC system for unstable conditions. with lines connecting the mean values.Considering the footprint of the systems. 126 .4. the footprint of the two systems overlap. for unstable conditions. It is clearly noted that easterly winds predominate and that the footprint function indicates that the source area is mostly within 2 km of the tower in unstable conditions. The variability is large.
The horizontal lines indicate the means. The estimation of fluxes from the LAS uses the reference function fT(ζ) (eq. where x is the distance of 90 % cumulative contribution to the fluxes.5. The estimation of fluxes from the LAS is based on this relation: from measurements of CT2.1 < rwt < 0.10). (2006). as a function of the MOS stability function fT(ζ).400 HEC HLAS 300 Heat fluxes (W. Figure 5. CT2 here is derived from the magnitude of T-spectra’s inertial subrange. low 127 . To identify a potential cause for the differences between the two systems. scaled by the appropriate scaling variables (CT2 z2/3 / T*). according to classes of x. following the methodology described by von Randow et al. To analyze in more detail in which conditions the flux estimations from the two systems differ.6 the points where the correlation coefficient between vertical wind velocity and air temperature (rwT) is 0. we highlight in figure 5.6 shows CT2 derived by the EC.6. Points where T* measured from by EC is lower than predicted from the reference line (therefore lower than the estimation of the LAS) appear above this line in the figure. The deviations of the EC measurements from this line therefore cause differences in the flux estimates of the scintillometer system. we studied the relationship between the structure parameter of temperature and the temperature scale T*.3.5. T* and subsequently H are derived using the stability function fT(ζ). as shown in figure 5.m ) -2 200 100 0 0 500 1000 1500 2000 distance of 90 % contribution (m) Figure 5. Boxplots of sensible heat fluxes measured by the eddy covariance (EC) and by the LAS.
When considering half-hourly or hourly records of turbulent signals. The line shows the stability function from Andreas (1989).6.9 (1-6. Measurements with rwt < 0. Eddy covariance measurements of the structure parameter of temperature.6 then indicate that the presence of slow moving (low-frequency) eddies might explain the differences between the EC and LAS estimations. as a function of -ζ.5). Highlighted points are measurements where 0. 2006) we suggest that low rwt conditions in this site are generally related to the modulation of surface layer turbulence by low-frequency motions.1 1 10 100 -z/LMO Figure 5. It is noted that most of the differences are related low rwT.1 were excluded from the analysis.1ζ) -2/3 0. Mahrt (1998) proposes a measure of nonstationarity based on the fact that for stationary conditions.1 0. the standard error of the record mean (due to variability within the record) predicts the variability between the 128 .1 < rwt < 0. scaled by measurement height and temperature scale of turbulence. The results in figure 5.01 0. low-frequency variations usually represent nonstationarity on these records.3. In a previous study (von Randow et al.01 0. values compared to the value expected for the surface layer over uniform terrain (about 0..100 10 CT2 z2/3 / T*2 1 fT(ζ) = 4. and mostly in near-neutral and weakly unstable conditions.
To calculate this measure. the time series is divided into I records and each record divided into J subrecord segments. it is noted that ratios based on 10-min average EC values tend to be bigger and more variable within a particular NR class. a between-record standard deviation is computed as σ btw = 2 1 I ∑ F ( i )− < F > I − 1 i =1 [ ] ( 5.13 ) where < F > is the segment flux averaged over all of the segments and records. for the three different averaging periods. 30 minutes and 1 hour. J − 1 j =1 [ ] ( 5. the nonstationarity ratio is defined as NR ≡ σ btw RE ( 5. a “within-record” standard deviation of the flux for the ith record and related random error are respectively computed as σ wi ( i ) = 2 1 J ∑ F ( i. ( 5.12 ) where F(i. in the following analysis we examine how the ratio of the LAS and EC fluxes change depending on nonstationarity. 129 .14 ) To further analyze the aspects of the timescales of the processes.record means – a deviation from this condition can then form a measure of nonstationarity. The LAS derived values. j ) − F ( i ) . which are later combined to 30 min and 1 hour records in the comparisons. first. Later. Comparing the averaging times. and F (i ) is the average of these segment fluxes for the ith record. Figure 5.11 ) RE = σ wi J . Finally. Then. are only calculated using 10 minutes CT2 averages. however. using three different averaging periods to define the mean and fluctuation parts in the EC calculations: 10 minutes. For all of them. HLAS / HEC is increasingly bigger with increasing NR.7 shows box plots of the ratio HLAS / HEC according to classes of NR.j) is the flux for the jth of the ith record.
The horizontal lines connect the means of each class.4 0.3 0.0 0.5 0.7 w . 6 10 min avg.6 10 min avg.1 0.2 0. 5 4 HLAS / HEC 3 2 1 0 0. 130 . but for classes of correlation coefficient between vertical wind velocity and air temperature (rwT). 1998) Figure 5. for 10- min. Boxplots of HLAS / HEC. 30 min avg.8.7. 30 min avg.T correlation coefficient Figure 5.6 0. 1 h avg. according to classes of nonstationarity ratio NR.7. 30-min and 1-hour averaging periods. Same as figure 5. 1 h avg. 5 4 HLAS / HEC 3 2 1 0 0 1 2 3 4 Non-stationarity ratio (Mahrt.
For highly unstable conditions. with the values obtained from the T-spectra measured by the EC. for classes of w-T correlation coefficient. CT2 measured at a lower level z1 will be larger than (z2/z1)2/3 times the value measured at a higher level z2 (Hartogensis. Furthermore.0 0. and 1-hour averages give the smallest. 131 .5 0. 2006). Boxplots of ratio of CT2 measured by the LAS and the EC. We note also that for all but the smallest rwT class.3 0.T correlation coefficient Figure 5. that there is a height difference between the EC and the LAS. depending on the shape of the fT functions. scaled to take height dependence into account. but has much higher values when rwT is small.2 0. Note. Hartogensis (2006) shows that the height dependence of CT2 is affected by stability: for neutral conditions. The ratio ranges around 1 for high values of rwT. CT2.7 w . it is also interesting to compare the basic variable measured from the LAS.Figure 5. but can be smaller or larger than this factor for non-neutral conditions. Because the estimation of fluxes from the LAS involves a number of calculations and use of MOS. smaller differences are observed for 30-min-averages. so care must be taken: the temperature variances usually decrease with increasing height. however.4 0. the ratio of CT2 measured at two different heights z1 and z2 is (z2/z1)2/3. A clearer relationship is observed in this case. 5 10 min records 30 min records 4 CT )EC 3 CT )LAS/ (z (z 2/3 2/3 2 2 2 1 0 0.8 shows box plots of the ratio HLAS / HEC according to classes of rwT.1 0. and so does CT2.6 0. the biggest differences between the LAS and the EC fluxes are obtained for 10-minaverages.9.
scaled for neutral height dependence. the measuring principle of the LAS would provide an advantage over the EC: because the LAS measures a variable that represents a spatial average along the path.. In that sense. 2006).. It has been shown that these low-frequency motions might significantly contribute to the surface fluxes (Sakai et al.5. according to classes of rwT. Our approach is slightly different. possibly including also transient periods. The challenge of inferring the low-frequency exchange using an EC system in a single-tower is due to the fact that it is necessary to rely on Taylor’s frozen turbulence hypothesis to relate the measured temporal fluctuations to the transporting eddy structures. (2001) studied the cospectra of turbulent variables selecting only data measured around midday. when the boundary layer is well developed and stationary.. Finnigan et al. 2003) and therefore should be accounted for when evaluating the surface-atmosphere exchange. Note that using only the neutral scaling in this figure we might slightly underestimate the CT2 height dependence for strongly unstable conditions. the LAS measures higher heat fluxes than the EC. 2001. when rwT is small. These conditions are usually related to the presence of slow moving and large structures in the surface layer (von Randow et al. increasing the averaging time interval in EC calculations also increases the uncertainty due to random errors and nonstationarity. 5.5 – 5. With a single tower EC system.Figure 5. especially characterized by low correlation between the vertical wind and air temperature signals. in some conditions. Discussion The results presented in figures 5. in general it is observed that the LAS appears to ‘capture’ more structures than the EC. Sakai et al.9 show that. 132 . Nevertheless. To evaluate the influence of low-frequency variations avoiding transition periods. it samples a larger number of eddies than the EC in a shorter time interval. it is not possible to differentiate whether the low-frequency temporal variations are due to slow moving large structures or due to changes in the flow conditions (non-stationarity). However.9 shows box plots of the ratios of CT2 values obtained from the LAS and the EC.
Energy balance closure for different ranges of rwT. 1998). As for the LAS the parameters are averaged over space as well as over time. Moreover.86 0.1 < rwT < 0. and λE is the latent heat flux. Nevertheless it is reasonable to expect that the energy balance closure would be better with the higher fluxes measured by the LAS. where Rn is net radiation. calculated using the parameterisation proposed by Moore and Fisch (1986). furthermore. From our results. We evaluate the energy balance closure for different ranges of rwT. the slope (a) of the fitted line Rn-S = a (H+λE).3 < rwT < 0. S is the sum of the soil heat flux and the heat storage in the canopy air space and biomass. we found that HLAS / HEC is increasingly bigger with increasing non-stationarity ratio (NR). a direct test of the energy balance closure using our LAS measurements is not possible.3 133 . which is inferred by the ratio between within-record and between-record statistics (Mahrt. because the LAS only provides sensible heat fluxes. Table 5. and. The coefficient of determination and number of points used in the regressions are also shown. in Table 1. it is expected that the EC calculations are less prone to lowfrequency modulations when rwT is large.1. when we would expect bigger changes of the flow conditions. the results qualitatively suggest that the effects we are observing are likely due to slow moving large eddies.5 Energy balance slope 0.83 0. As expected from the discussion above. when we evaluated the ratio HLAS / HEC for different times of the day (not shown).83 0.87 # of points 425 696 803 0.5 0. we found that the ratio is not particularly higher for transition periods in early morning or late afternoon.77 r2 0. during daytime unstable conditions. the energy balance closure improves when data with higher rwT are used.In the previous sections. This table shows. an appropriate measure of the spatially averaged net radiation would be necessary.90 0. Unfortunately. Data included in these analyses are half-hourly measurements. for different ranges of rwT. rwT range rwT > 0. and to a lesser extent due to transient conditions.
Smaller differences are observed for 30-min-averages. The results are attributed to the spatial averaging effect of LAS measurements: a larger number of eddies is sampled by the LAS than by the EC in a similar time interval. Looking into the relationship between the structure parameter of temperature (CT2) and the temperature scale T*. 134 . the EC often measured lower fluxes than the LAS. under unstable conditions. we found that the scaled CT2 differs from the expected values predicted from Monin-Obukhov Similarity (MOS) when the correlation between vertical wind velocity and air temperature signals (rwT) is low. significantly modulating the turbulence in the surface layer. It is also noted that this happens mostly in near-neutral and weakly unstable conditions. for example. with the objective of studying the conditions of low-frequency modulation in the surface layer. a more adequate measure of the spatial fluxes is necessary using. and 1-hour averages give the smallest. we obtain that the biggest differences between the LAS and the EC fluxes are found for 10-min-averages. A single tower EC can then fail to capture their spatial variability. or other methods such as aircraft flux measurements or Boundary Layer budgets. 5. Conclusions In this study we compared measurements from a Large Aperture Scintillometer (LAS) and from an Eddy Covariance (EC) system over an Amazonian rain forest canopy. a large aperture scintillometer. Further analyzing aspects of the timescales of the processes using different averaging times on EC calculations.6. In these cases. The difference is observed to increase with both increasing non-stationarity and decreasing correlation between vertical velocity and air temperature signals (rwT). Generally comparing the two systems by analyzing the ratio of estimated heat fluxes HLAS / HEC.Our results suggest that rwT can be an indicator of the importance of low-frequency motions in the surface layer.
135 .Generally. we found that the closure improves when data with increasingly higher rwT are used. our results suggest that rwT can be used as an indicator of the importance of low-frequency motions in the surface layer. Evaluating the energy balance for different ranges of rwT.
These large-scale (low-frequency) motions modulate the smaller-scale turbulent structures over the forested surface at time scales that are relatively long compared to the time scales observed in the surface layer over low vegetation and flat. the area has a relatively strong seasonal variability in rainfall.Chapter 6 Summary and Perspectives 6. The atmospheric boundary layer over Amazonia frequently contains slowly-moving large eddies induced by strong convective motions or local circulations related to the heterogeneity of the surface. in south-west Amazonia. with distinct wet and dry season periods. The experimental sites studied are located in south-western and central parts of the Brazilian Amazonia. in conditions of low-frequency motions. The Rebio Jaru forest. and is often affected by local circulations induced by the heterogeneity of the terrain. To better understand the influence of these low-frequency variations. Moreover. combining spatial averaging and time averaging of turbulence statistics.1 Summary The objective of this thesis is to investigate the role of low-frequency variations on turbulent exchange processes over Amazonian forests. rather than time averaging alone as is the case with the eddy correlation (EC) method. homogeneous surfaces. in comparison with the eddy covariance. we studied how turbulence statistics depend on time scales and spatial scales. Moreover. The application of a scintillometer over a tropical forest is an innovative aspect of this study. Large Aperture Scintillometry. Comparative data collected at a pasture site (FNS) 85 km away make the two datasets valuable to study the different functioning of the 137 . The influence of low-frequency motions on similarity and correlations between turbulent signals and the implications for the application of Monin-Obukhov Similarity is also investigated. we explore a relatively new measurement technique. is situated near substantially deforested areas.
Seasonal changes in energy partitioning are small in the forest compared to large reductions in evaporation observed in the dry season at the pasture. the K34 site is covered with pristine forest. This is because the forest vegetation can withdraw water from deeper layers in the soil under water stress. collected almost continuously during 4 years at Rebio Jaru forest and FNS pasture. the combined measurements over forest and pasture presented in Chapter 2 provide a valuable dataset to calibrate models of surfaceatmosphere interaction and to better understand the behaviour of surface fluxes over the two types of vegetation under conditions of water stress. its implication for similarity relationships (Chapter 4). The difficulty of measuring turbulent fluxes over the forest and the importance of low-frequency contributions is illustrated by the poor energy balance closure: the sum of the heat fluxes reaches only 84 % of the available energy. This undulating topography may also influence the atmospheric flow and the position of convective structures. we observe that both photosynthesis and respiration are relatively lower in the pasture compared to the forest. sensible heat.two vegetation types and the scale variability of the fluxes. In central Amazonia. but consists of a mosaic of well-drained plateaus dissected by valleys that are often waterlogged after rain. the effect is a relatively stronger sink of CO2 in the pasture. even when the averaging time is increased to a few hours. evapotranspiration and CO2. (ii) horizontal flux divergences that cannot be estimated from measurements made on a single tower significantly contribute to the local exchange of energy and mass flux. In this site we investigated the influence of low-frequency motions in the structure and scales of turbulence in the surface layer. Despite the uncertainty in absolute accuracy. and the application of the Large Aperture Scintillometer (Chapter 5). during wet and dry seasons (Chapters 2 and 3). Our analyses suggest that the lack of energy balance closure is caused by one or both the following reasons: (i) changes in wind direction and vertical wind associated with slowly moving large structures that cause transports on timescales longer than 1-2 hours. Comparing the CO2 exchange at the two sites. Although this is 138 . Chapter 2 presents the fluxes of radiation. As the reduction in the respiration is larger than the reduction in daytime uptake (photosynthesis).
especially in the dry season. we tested the estimation of heat fluxes by the flux-variance method and empirical relationships based on MOS theory. for modelling experiments to be driven by realistic scenarios of vegetation cover and land use change (i. considering the effect of cattle grazing constantly renewing the growth of the vegetation in the pasture. which are particularly interesting with respect to the entrainment zone influence on the boundary layer and even on the surface layer.not surprising. for example. both in wet and dry season periods. The largest contributions to sensible heat. however. For this study. can also be observed as modulation of the smaller eddies near the ground. The wavelet analyses also provided an assessment of the scale dependence of humidity skewness and of the correlation coefficients between virtual temperature and humidity. well marked in humidity signals (Chapter 3). Based on the behaviour of the decomposed signals according to different temporal scales. latent heat and CO2 fluxes occur in turbulent scales (length scales up to 1000 m and time scales up to 15 min). In Chapter 4 we studied in more detail the influence of low-frequency variations on the structure of turbulence in the surface layer and the implications for the application of MoninObukhov Similarity (MOS) theory. We found that 139 . using wavelet transforms to decompose the measured signals assessing the contributions of different temporal scale classes. conversion of forests to pasture and agricultural areas). This exercise permitted the study of similarity patterns between these two variables. can contribute with up to 30 % to the total covariances under weak wind conditions. In Chapter 3 the characteristics of scale dependence of variances and covariances (fluxes) of turbulent variables were studied in more detail. emphasizing the need. The results showed that the covariances are highly variable on larger (meso) scales and can be positive or negative regardless of stability conditions or the mean gradients in the boundary layer. The influence of large eddies in the surface layer.e. A “normalized standard deviation of the scale flux” (NSDF) was defined and proven to be a good indicator to separate the mesoscale regime from the turbulent signals: a clear jump in NSDF values is observed where motion scales increase to above 1000 m (figure 3. the results highlight important differences between the two types of surface. Low-frequency motions (larger eddies and mesoscale motions). we proposed to partition the fluxes in two categories: ‘turbulent’ and ‘mesoscale’ fluxes.5). The results show that large eddies carry drier air from the entrainment zone down to the surface layer.
for 20 % of the unstable records analyzed. 2006) proposes a new model of turbulence structure. In a new view of the formation of eddies in the surface layer. for example. ruw is lower than the usually observed value of (-)0. it is very difficult to define an adequate averaging time. enhancing the dissipation rate (ε). in which the active turbulence in the surface layer forms a self-organizing system that resists change by local instability. In the general framework of MOS theory.g. 2002). This result supports the interpretation that lowfrequency motions indeed affect the structure of smaller scale turbulent structures and.8) – v* is a parameter that represents the additional momentum transported down from the variable motions of the outer layer to the surface layer – we found that when v* is of comparable magnitude as u* or higher (v*/u* ≥ 1). in Chapter 4 we also tested the use of uε to scale the standard deviation of w.5. 1994). Consistent with the fundamental proposition of McNaughton’s model. McNaughton and Brunet. Further investigating the scale dependence of rwT and of u-w correlation (ruw). Another aspect of studying low-frequency variations is related to whether the averaging time is appropriate to sample the active eddies passing by the tower. 1990. In a situation where large convective cells persist for hours locked into fixed positions in the landscape. McNaughton (2004. the outer convective motions (larger boundary-layer eddies) modulate the surface layer structure. Further investigating the relationship between v*/u* and ruw (figure 4. surface heat flux and buoyancy – but there is evidence that some properties are also influenced by ‘outer’ parameters – parameters of the mixed layer. and in these cases the surface layer is different from the textbook descriptions. Hogstrom. and new scales for temperature (Tε) and humidity (qε). According to McNaughton theory. With the objective of making a qualitative test of the application of this model. An appropriate velocity scale to be used in the surface layer would be then uε = (kzε)1/3. the characteristics of surface layer turbulence is described only by ‘local’ parameters – parameters of the surface layer itself.the relationships provide reasonable results only when the correlation coefficient between vertical wind and temperature (rwT) is above 0.35 in the surface layer over flat uniform terrain (Kaimal and Finnigan. such as the height of the boundary layer and the convective velocity scale (e. a measure of the 140 . such as friction velocity. we found that these quantities decrease under conditions of low-frequency variations. the scaled turbulent statistics became independent of stability. induce deviations from textbook descriptions of the surface layer. In these conditions.
In Chapter 5. we obtained the biggest differences between the LAS and the EC fluxes for 10-min-averages. less so for 30-minaverages. Evaluating the energy balance for different ranges of rwT (table 5. our analysis in chapter 5. (2003) also show that motions on scales longer than 30 min substantially contribute to the exchange in other forests. play a role. As the LAS provides a measurement that represents a weighted spatial average of the turbulent eddies along the path. A general result from our studies is that motions on scales of the order of 30 min to 1 h clearly transport part of the energy and mass over the Amazonian forest. These results are attributed to the spatial averaging effect of the LAS. comparing 141 . Using different averaging times on EC calculations. also analyzing conditions of low-frequency motions. (2001) and Finnigan et al.1). topography.spatial variability of turbulence is necessary. and the averaging time scale can be reduced. we found that the closure improved when data with increasingly higher rwT were used. Moreover. especially because many factors. results from chapter 4 and 5 suggest that rwT can be used as an indicator of the importance of low-frequency motions in the surface layer. Generally. it is not necessary to use a long time period to sample a number of eddies. a Large Aperture Scintillometer (LAS) is used to estimate the sensible heat fluxes. and 1-hour averages gave the smallest differences. The subject of low-frequency transport is still poorly understood. in combination with the EC system. self-organized turbulent structures and mesoscale flows. such as surface heterogeneity. 6. The results show that the EC fluxes are often lower than the LAS. Sakai et al.2 Perspective and recommendations The analyses and results of the studies presented in this thesis provide improvement to our knowledge of the influence of low-frequency (convective) motions on the turbulent exchange of mass and energy in the surface layer in the tropics. The difference is observed to increase with both increasing non-stationarity conditions and decreasing rwT.
extending the averaging time to very low frequencies should be done with caution. (2004) can be applied for this purpose. it is important to quantify the uncertainties of eddy flux calculations. especially caused by different averaging / rotation timescales. Based on the large variability observed at the mesoscale. the contributions are highly variable at the lowest frequencies and are prone to large random errors. the general scheme proposed by McNaughton (2004. we believe these factors explain the failure to close the energy balance at our sites. most of the exchange. because the transport mechanisms for momentum and scalars are not the same (Kays. and in many others. however. such as the multi-resolution decompositions applied in this thesis (chapters 3 and 4). 142 . However. Therefore. to evaluate an appropriate averaging timescale that captures. A proper description of the modulation of scalar fields is still necessary. For future studies of the surface-atmosphere interaction in complex sites. Apart from possible horizontal flux divergence at Rebio Jaru.EC with the spatially-integrating estimate from the LAS suggests that the eddy covariance system may not be capturing quasi-steady eddies that can also significantly contribute to the transport. we suggest a careful analysis of the contributions of different scales of motion to the total transports. 2006) provides a foundation for a new similarity model of momentum exchange in the surface layer. we also suggest trying to separately classify and quantify the contributions of ‘turbulent’ and ‘mesoscale’ components of the total flux. Although we recommend including the low-frequency components in the flux calculations. on average. 1994. an averaging time of at least 30 min will be necessary. Methodology similar to Mahrt (1998) or Kruijt et al. This classification is useful for modeling purposes. Regarding the modeling issue. Future development of McNaughton’s theory or similar models that include explicit dependences on relevant outer layer parameters would be extremely important for a better understanding of the unstable atmospheric surface layer. a criterion is suggested to define the separation between the two classes (Chapter 3). As shown in chapter 3. 1970) and for this reason a simple extension of McNaughton’s model to scalar fields is not adequate. accounting for the effects of low-frequency motions from the outer layer. Tennekes. We expect that over tall forests.
. R. Lelieveld.. 6.L. 59. A. Vesala.... Morgenstern.... A. Atmos Sci.... O. Carswell. A.1029/2001JD000676. Estimates of the annual net carbon and water exchange of forests: The EUROFLUX methodology.. J... A. On the Kolmogorov constants for the temperature-humidity cospectrum and the refractive index spectrum. E. T. Dallarosa... C.. 2000. Snijders. F. T. R... Hicks.. Artaxo. Manzi.. Oceanic Tech. Grünwald. water. 1987. P. 8090. Ü.. Atmos. 102. C.. Kowalski. C. B. B.. G.. 2605-2607. Ruivo. J. Kruijt... Gash. Biogeochemical cycling of carbon. Foken. P. 107(D20).. Andreas.... D. J. Berbigier. M. Granier. P. J Geophys Res. P. 2002. J. Malhi. Geophysical Res.. A. E. 13423-13432. C.. 143 . 8066. Parlange. F. 113-175. B. M. M. O. Valentini.. Valentini. J. A. Stefani.. O. The average dissipation rate of turbulent kinetic energy in the neutral and unstable atmospheric surface layer..S. 30.. Meixner. 44.. Grace. W. D. R. P. Kiely. and aerosols in Amazonia: The LBA-EUSTACH experiments. 1997... J. E. C. Adv Ecol. Gash. von Jouanne.. J... J. W. Brandão. C.. J Geophys Res. Pilegaard. Manzi.. Atmos.. Kabat. Esteves.References Albertson. L.. Eichinger. P. M.1029/2001JD000524.. Andreae. A. A.... A.. Res. M. and Kabat. da Costa. D. C. doi: 10. M. R.. Nobre. A. Elbers. A. J. Andreas. P. Culf. J. 280-292.. Y. A. 2002. Ch. J. Two-wavelength method of measuring path-averaged turbulent surface heat fluxes.. Sci. H. D. L. trace gases. Araujo. K. Ibrom... T. Clement. K. A. Grelle. 2399-2406. L. Waterloo.. Comparative measurements of carbon dioxide fluxes from two nearby towers in a central Amazonian rainforest: The Manaus LBA site. P. Rebmann.. von Randow. Valentini. Bernhofer. Ciccioli. L. J. J. Elbers.. Rannik. Comments on “Critical Test of the Validity of MoninObukhov Similarity During Convective Conditions”. E. Andreas. 1989. Stefani. R. J.. X. J. Martin.. doi: 10. L. F. E. B. Silva Dias.. P.. Moncrieff. A.H. Nobre. Culf. Aubinet. H. Nobre... 2002. 107(D20).. D. energy.
W. Atmos Sci.. J. 107 (D20).. 63. C... Palheta. doi: 10. J. Dallarosa.. Exploring SelfCorrelation in Flux-Gradient Relationships for Stably Stratified Conditions. Susskind.. 139-163. Brunet. E. The Control of Coherent Eddies in Vegetation Canopies: Streamwise Structure Spacing.. 1998.. Grace. 49-73. M. Steeneveld. W. and surface radiation over cleared and undisturbed amazonian forest. Digirolamo. F. Shuttleworth. Carswell. G. Darnell.. Observations of climate. J. J. R. Meir.. 2002. J. Collineau. J.. Opt... J.. 2433-2442. 107(D20). Y. doi:10. M. Irvine.. Canopy Shear Scale and Atmospheric Stability. 2002. Leal.. N.. 783-796. A. Preface to special issue on the Large-Scale BiosphereAtmosphere Experiment in Amazonia (LBA). homogeneity.. Res. P.K... 8076. 186-204.E. Nobre..L. Malhi. Boundary Layer Meteorol. 2000. M. 8034. 2006. J.. Detection of Turbulent Coherent Motions in a Forest Canopy. S. 1993. albedo. Int J Climatol.. Brutsaert. Part II: Time-scales and Conditional Averages. Geophys. Bastable. 205..Avissar.. Clifford.... van de Wiel.. A biophysical process-based estimate of global land surface evaporation using satellite and ancillary data – I. Boundary Layer Meteorol.. Regional and global patterns of seasonal and annual variations. Land-surface water vapor and sensible heat flux: Spatial variability. S. Brunet. Hydrol... Y. Costa. Fisch. 94 (1). H. 1998.G. A. P..L.. R. J Geophys Res. 1971. 34 (10).J. Asrar. Costa. M. R. C.A. L. W. J. Costa.. 1993. G.. Holtslag. B. Choudhury. B. L. Temporal-frequency spectra for a spherical wave propagating through atmospheric turbulence. P. R. Seasonality in CO2 and H2O flux at an eastern Amazonian rain forest. A. S. Y. Ruivo. 66 (1-2).. Water Resources Res. R. M. L. Gupta. Baas.G.. Amer. N. Soc... 3045-3054.1029/2000JD000284.F. 13. 1285-1292. M. G.. A.J. S.1029/2002JD002507. and measurement scales. Nobre. 144 . 61 (10).. Clement.
R. Boundary-Layer Meteorol.. Boundary-Layer Meteorol.Culf. Springer Verlag. J. The scintillation method tested over a dry vineyard area. J.). P. in Amazonian Deforestation and Climate. W. A. R. D. renaissance of scintillometry. 1995.. Dirmeyer. B. G. J. Germany. Heidelberg. 1999. On the TemperatureHumidity Correlation and Similarity. A Verification of Some Methods to Determine the Fluxes of Momentum. Gash. Philadelphia. W.. 193-198. C. C. Water.G..L. De Bruin. De Bruin. Fisch.. Esteves. C. Sensible Heat. Humans and the Climate. M. M.L. M. M. and Water-Vapor Using Standard-Deviation and Structure Parameter of Scalar Meteorological Quantities. H. Gash.. J. H. Boundary Layer Meteorol. 93. Deguenni. 1995.O..R... Van Den Hurk. 453-468. J.M.. Kohsiek. J.. Hydrol. 2002. H. van den Hurk. A. Claussen. NewYork.. 1991. T..D. M. R. in Land Surface Evaporation. R. 175-191.R. 4(1). 1-4.D. H. 76. Earth System Sci. 25-40. H. Schmugge T. I. De Bruin. H. Marques Filho. Fluxes in the surface layer under advective conditions. J. 357 pp. (eds). Culf. H. Victoria (eds. Pielke Sr... Culf. C. H.. Rocha.. John Wiley. Meybeck.R. 2004: The energy balance closure problem.A. P. J.. A. 1996.1544-1554. L. R. L. B. De Bruin.. 157-171. Bink. Boundary-Layer Meteorol. Kabat. N. H.J. J. SIAM. A. Chichester. H. Ten Lectures on Wavelets. H. Foken.. Kohsiek.. Nobre. 231-257... 2000. Radiation. Springer Verlag. Van den Hurk. 145 . Climate. Kroon. Vörösmarty.J. Roberts and R. In: Vegetation. J.. Hodnett. B. S. A. M. Examples of the effects of different averaging methods on carbon dioxide fluxes calculated using the eddy correlation method. 1992. A. 8. 1993. L. A. Culf. Daubechies. André. J. Introduction. J. The Albedo of Amazonian Forest and Rach Land. Gash. Lütkemeyer (Eds)..A.. 159-166. D. B. temperature and humidity over forest and pasture in Amazonia. A. J. C. C. Hutjes. 63.. J.. A. W.A.. A.. A. 105.. De Bruin. J. Kroon..
Evapotranspiration models with canopy resistance for use in climate model, a review.
Agric. Forest Meteorol., 54, 373-388.
Elbers, J.A., 1998. Eddy Correlation System Winand Staring Centre, user manual version 2.0, internal report, Alterra, Wageningen, The Netherlands, 39 p. Fan, S.-M., Wofsy, S.C., Bakwin, P.S., Jacob, D.J., Fitzjarrald, D.R., 1990. Atmospherebiosphere exchange of CO2 and O3 in the central-Amazon-forest. J. Geophys. Res.,
Farge, M., 1992. The Wavelet Transform and its Applications to Turbulence. Annu. Rev.
Fluid Mech., 24, 395-457.
Finningan, J., 1999. A comment on the paper by Lee (1998) On Micrometeorological observations of surface-air exchange over tall vegetation. Agric. Forest Meteorol., 97, 55-64. Finnigan, J. J., Clement, R., Malhi, Y., Leuning, R., and Cleugh, H. A., 2003. A re-evaluation of long-term flux measurement techniques - Part I: Averaging and coordinate rotation.
Boundary-Layer Meteorol., 107, 1-48.
Fitzjarrald, D.R., Stormwind, B.L., Fisch, G., Cabral, O.M.R., 1988. Turbulent transport observed above the Amazon forest. J. Geophys. Res., 93 (D2), 1551-1563. Fitzjarrald, D. R., Moore, K. E., Cabral, O. M. R., Scolar, J., Manzi, A. O., Sá, L. D. A., 1990. Daytime Turbulent Exchange Between the Amazon Forest and the Atmosphere. J.
Geophys. Res., 95 (D10), 16825-16838.
Gao, W., Li, B. L., 1993. Wavelet Analysis of Coherent Structures at the Atmosphere-Forest Interface. J. Appl. Meteorol., 32 (11), 1717-1725. Garratt, J.R., 1993. Sensitivity of climate simulations to land-surface and atmospheric boundary layer treatments. A review. J Climate, 6, 419-449. Garstang, M., Fitzjarrald, D. R., 1999. Observations of Surface to Atmosphere Interactions in
the Tropics. Oxford-University-Press, New York, 405 pp.
Gash, J. H. C., Huntingford, C., Marengo, J. A., Betts, R. A., Cox, P. M., Fisch, G., Fu, R., Gandu, A. W., Harris, P. P., Machado, L. A. T., von Randow, C., and Silva Dias, M. A., 2004: Amazonian climate: results and future research. Theor. Appl. Climatol., 78, 187–193. Gash, J. H. C., Nobre, C. A., Roberts, J. M., Victoria, R. L., 1996. Amazonian Deforestation
and Climate, John Wiley and Sons, Chichester, UK, 611 pp.
Gash, J. H. C., Dolman, A. J., 2003. Sonic anemometer (co)sine response and flux measurement: I. The potencial for (co)sine error to affect sonic anemometer-based flux measurements. Agric. Forest Meteorol., 119, 195–207. Goulden, M. L., Miller, S. D., Menton, M. C., da Rocha, H. R., Freitas, H. C., 2004. Diel and Seasonal Patterns of Tropical Forest CO2 Exchange. Ecological Applications, 14(4) Supplement, S42–S54. Grace, J., Lloyd, J., McIntyre, J., Miranda, A., Meir, P., Miranda, H., Moncrieff, J., Massheder, J., Wright, I., Gash, J. H. C., 1995. Carbon dioxide uptake by an undisturbed tropical rain forest in south-west Amazonia, 1992-1993. Science, 270, 778-780. Grace, J., Lloyd, J., Miranda, A.C., Gash, J.H.C., 1998. Fluxes of carbon dioxide and water vapour over a C4 pasture in south western Amazonia (Brazil). Australian J. Plant
Physiol., 25, 519-530.
Gu, L., Fuentes, J. D., Garstang, M., Tota da Silva, J., Heitz, R., Sigler, J., Shugart, H. H., 2001. Cloud Modulation of surface solar irradiance at a pasture site in southern Brazil,
Agric. Forest Meteorol., 106, 117-129.
Hartogensis, O. K., 2006. Exploring Scintillometry in the Stable Atmospheric Surface Layer. PhD Thesis, Wageningen University, Wageningen, 227 pp. Hartogensis, O.K., Watts, C.J., Rodriguez, J.-C., De Bruin, H.A.R., 2003. Derivation of the effective height for scintillometers: La Poza experiment in Northwest Mexico, J.
Hydrometeorol., 4, 915-928.
Hartogensis, O. K., De Bruin, H. A. R., 2005. Monin-Obukhov similarity functions of the structure parameter of temperature and turbulent kinetic energy dissipation rate in the stable boundary layer. Boundary-Layer Meteorol., 116, 253-276. Hicks, B.B., 1978. Some Limitations of Dimensional Analysis and Power Laws. BoundaryLayer Meteorol., 14, 567-569.
Hicks, B.B., 1981. An Examination of Turbulence Statistics in the Surface Boundary Layer.
Boundary-Layer Meteorol., 21, 389-402.
Hildebrand, P. H., 1991. Error in Eddy Correlation Turbulence Measurements from Aircraft: Application to HAPEX-MOBILHY. in Land Surface Evaporation - Measurement and
Parameterization, T.J.Schmugge and J.-C. André. (eds), Springer-Verlag, New York,
231-243. Hill, R. J., 1989. Implications of Monin-Obukhov Similarity Theory for Scalar Quantities, J.
Atmos. Sci., 46 (14), 2236-2244.
Hill, R. J., 1992. Review of Optical Scintillation Methods of Measuring the Refractive-Index Spectrum, Inner Scale and Surface Fluxes. Waves in Random Media, 2, 179–201. Hill, R. J., 1997. Algorithms for Obtaining Atmospheric Surface-Layer Fluxes from Scintillation Measurements. J. Atmos. Oceanic Tech., 14, 456–467. Hill, R. J., Ochs, G. R., Wilson, J. J., 1992. Measuring Surface-Layer Fluxes of Heat and Momentum Using Optical Scintillation. Boundary-Layer Meteorol., 58, 391–408. Hodnett, M. G., Oyama, M. D., Tomasella, J., Marques Filho, A. O., 1996. Comparisons of long-term soil water storage behaviour under pasture and forest in three areas of Amazonia. in Amazonian Deforestation and Climate, J. H. C. Gash, C. A. Nobre, J Roberts and R. L. Victoria (eds.), John Wiley, Chichester, 57-77. Hogstrom, U., 1990. Analysis of Turbulence Structure in the Surface-Layer with a Modified Similarity Formulation for near Neutral Conditions. J. Atmos. Sci., 47, 1949-1972. Holtslag, A. A. M., Moeng, C.-H., 1991. Eddy Diffusivity and Countergradient Transport in the Convective Atmospheric Boundary Layer, J. Atmos. Sci., 48 (14), 1690-1698.
G. Sigmon. Atmos. 1994. S. 2002. Kader... A. T. S. 107-128.).. J. 18-24. F. J. B. Critical test of the validity of Monin-Obukhov similarity during convective conditions. 51 (15). 23. W.. Reply to Comments on “Critical Test of the Validity of Monin-Obukhov Similarity During Convective Conditions”. 1978.. 116. Parlange... 1994. S. M. C.. Hsieh. Johansson. New York.. J. Horizontal Velocity Spectra in an Unstable Surface-Layer. G.. San Diego. 149 . Atmospheric Boundary Layer Flows: Their Structure and Measurement. Katul. A. Howell. L. Johansson. J. 637-662. Kumar and Efi Foufoula-Georgiou (eds. B. Kaimal. Finnigan. G.. A. Mahrt. Yaglom. 117-137.. L. Chi. Hogstrom. 2608-2614. Atmos Sci. On the Active Role of Temperature in Surface-Layer Turbulence. 2001. Smedman. Hsieh. 35. G... 289 pp. Oxford University Press. 1997. 2181-2195. J. Energy-Inertial Scale interactions for Velocity and Temperature in the Unstable Atmospheric Surface Layer.. 49-80. 765–772... 1994. Boundary-Layer Meteorol.. A. J.. Hogstrom. 83. Mahrt. J. in Wavelets in Geophysics. C. P. U. C. J. Turbulent Prandtl Number – Where are we? J. J. 1549-1566..-I... 284295. Brasseur. An approximate analytical model for footprint estimation of scaler fluxes in thermally stratified atmospheric flows... Atmos. Brasseur. 2000. Academic-Press. Khanna. Smedman. G. Water Resour. Kaimal. M.Howell. 1994. J.. J.. M. 1997. C. C. Mean fields and fluctuation moments in unstable stratified turbulent boundary layers. Fluid Mech.. J. Kays.-I... J. Katul. Sci. 58.. C. 1990. Sci.. Katul. 82 (1). Multiresolution Flux Decomoposition. 59. U. Heat Transfer. Sci. An Adaptative Decomposition: Application to Turbulence.. Atmos.. Boundary Layer Meteorol. 212. G. F. Khanna. S. J. Adv...
Boundary-Layer Meteorol. S101-S113. Dolman. von Randow. 1991. IGBP Newsletter.. A. 1973. P. B. J.. 14(4) Supplement. 2004. Atmos. Miller. 39-49. Alencar.. 1976. R... Dokl. B. Atmos... A. Lloyd. Kruijt. Asner. A. B. Boundary-Layer Meteorol. Atmos. Priante. Markewitz... 30. O. S. H. Moors. Sci. J.. G. G. Lloyd.. Keller. Trumbore. B. M. Sci. J. D. 57. A. Nobre. Oliveira.A. Bustamante. 14(4) Supplement. P. A. G.R.. B.. M.. 2001... 91. 2000. C. Ecological Applications. A. C... C. N. D. E. A. Culf. 150 . M.. LeMone. J. Agric Forest Meteorol.. J. Culf. C. Rocha. 299-303. Kruijt. Nobre.. H.. Kolmogorov.... D. Kabat.. A. E. Culf. Ecological Applications. Turbulence statistics above and within two amazon rain forest canopies. 33. Luizao. Akad.. Lee. Von Randow. Estimates of Sensible Heat Flux from Observations of Temperature Fluctuations. Ecological Research in the Large Scale Biosphere Atmosphere Experiment in Amazonia (LBA): Early results.. 1998... D.. N. S. M.. 297-331... Feldpausch. 2004.. X. Braswell. Silva Dias. 43. Length scales in the convective boundary layer. A. Investigating the Carbon Cycle of the Amazon Forests.. Nobre.. 15-19. Gash... C. A. J.. On micrometeorological observations of surface-air exchange over tall vagetation.. C. Davidson... 1986. Kruijt. 1308-1320. The local structure of turbulence in an incompressible viscous fluid for very large Reynolds numbers.... A. Sci. Y. E. 311-322. Elbers. Nobre. Kruijt.. J. The structure and dynamics of horizontal roll vortices in the planetary boundary layer. P. 30. C.D. 1941.. Stankov. P. F.. M. T. Rocha.. 1198-1209. Lenschow. Modulation of turbulent energy by longitudinal rolls in an unstable boundary layer. Manzi. The robustness of eddy correlation fluxes for Amazon rain forest conditions.Keller. J. 1308–1320. D.. H. Araujo. A. Malhi. Miranda. A.. J.. A. M.. M. Grace. Goulden.. S3–S16.. B.. Pereira. 94.. H.P.. Kabat. Vourlitis. LeMone. J. Nauk. issue No. 45. Fernandes.. B.
Malhi. Observations of Fluxes over Heterogeneous Surfaces. Res. The Structure of Atmospheric Turbulence. Boundary-layer moisture regimes. Jarvis.. Malhi.. Geophys. and B. J. 8061. Mahrt. D... 78. Nobre. 1991.. Mahrt. McNaughton. G. D. Res.. 151 .. A. 332-337. Pederson. Mahrt... 715-740. Vickers. Y. R.. L. Boundary Layer Meteorol. Pereira. Mahrt. J. 1964. M. A. D. D. temperate and boreal forests. Q. L. Atmos.).. D. W. Geophys. R. Wiley. Culf.. 15. Malhi. 1999. 345-367. Carbon dioxide transfer over a Central Amazonian rain forest. J. H. Panofsky. Law (eds.. Boundary Layer Meteor. 151-176. 1994. G. Flux Sampling Errors for Aircraft and Towers. 51. Meteorol. Trends Ecol.. The carbon balance of tropical.1029/2001JD000623. G. R. 101-118.. Cell and Environment. L. Macpherson.. J. Grace. P. Macpherson. Tropical forests and atmospheric carbon dioxide.. J. The Bulk Aerodynamic Formulation for Surface Heterogeneity. Sun. Y.. G. J.. and Desjardins. Kluwer Academic Publishers. 2000.. von Randow. 1994. Evol. Grace.. 31593-31612. 1996. Desjardins... P. K. Observations of Fluxes and Inland Breezes over a Heterogeneous Surface. 2004. Energy and water dynamics of a central Amazonian rain forest. E. Mahrt L.. L. J. 1998.. Pereira. L. A. Plant. Grace. I. 87-119.. Oceanic Technol. I. S. 22. 103. 107(D20).. J.. Sci. J. 1998. Y.. R. Scott. A... Atmos. Pegoraro. Y. doi: 10. Soc.. Culf.Lumley. Clement... Nobre. J. J. Y. Massman. A.. Low frequency atmospheric transport and surface flux measurements. 2484–2499. 117. M. 2002. The Netherlands. Baldocchi. L. Kruijt. 416-429.. 239 pp. R.. New York. D. J. in Handbook of micrometeorology: a guide for surface flux measurement and analysis. 15.. Malhi. X Lee. C.. B.. 67. Malhi.. Dordrecht.... J.
Boundary Layer Meteorol. H. 2004. C.).. Laubach. Freitas. 143-185. J.. 1972. Chichester. S114– S126. 98. H. Power Spectra and Cospectra for Wind and Scalars in a Disturbed Surface Layer at the Base of an Advective Inversion. Roberts and R . Turbulence structure of the unstable atmospheric surface layer and transition to the outer layer. Nobre. R. Soc. A.. M.-L. G.. 2002. C Gash. M. PAMI. C. L.... O. R.. An Eddy-Correlation Technique with Extended Applicability to NonSimple Terrain. McNaughton.. Boundary-Layer Meteorol. 37-62.. A theory for multiresolution signal decomposition: the wavelet representation... Gomes. Boundary-Layer Meteorol. Meijninger. C. G. 105. D. C. W.. Goulden. A. B. Menton. 83-107. IEEE Trans. R. H. Silva Figueira. Meteorol. McWilliam. De Bruin. 112. S. Q. Boundary-Layer Meteorol. Cabral. R. Hartogensis. K.Mallat... K. G. R. 231-245. J.. M. M. Zuurbier. 11. McNaughton. Biometric and micrometeorological measurements of tropical forest carbon balance. Kohsiek.. A. 1989. Boundary-Layer Meteorol. B. John Wiley. A. J... 1988. McMillen. 2002. 161-175. J. L. da Rocha. 96. 265-285. G. Townsend's hypothesis. Boundary-Layer Meteorol. K. Miyake. W. 43. Ecological Applications. Y. M. McBean. 1996. G. K.. Brunet. 2006. Roberts. 2004. S. L.. 152 .. A. 14(4) Supplement. M. de Sousa. D. Forest and pasture leaf-gas exchange in south-west Amazonia. J. 383-398. Hoedjes. Determination of area averaged sensible heat fluxes with a large aperture scintillometer over a heterogeneous surface – Flevoland field experiment.. C.. J. J.. 118.. coherent structures and Monin-Obukhov similarity.. Victoria (eds. 674-693. On the Kinetic Energy Budget of the Unstable Atmospheric Surface Layer. M. in Amazonian Deforestation and Climate. McNaughton.. C.. K. R.. McNaughton.. 102.L. Miller. M. A.. T. 199-221. Turbulent transfer mechanisms in the atmospheric surface layer. M.. 2000. O.. Esteves. H.
. The relationship of subcloud fluxes and structure to the distribution and enhancement of cumulus clouds. M. R. 1975. J.. J. H. Basdevant. A system to measure surface fluxes of momentum. A. Atmos... MA. J. 87. 41. J. sensible heat. 1984. 38. Turbulence in and above Plant Canopies.-H. Fluid Mech. Raupach. M. USA. J. Moore. 13. Statistical Fluid Mechanics. S. Gash. L. Sci. Scott. M. Kohsiek..). S.. 97-129. J. water vapour and carbon dioxide. J. T. Wyngaard. Fisch. C. Thom. 37. H.. H..Moeng. 1981.. Wavelet spectra compared to fourier spectra. New York. Models and Methods for Engineering Applications. Nobre. 3161-3169. Gash. 2051-2067.. C. 1971. Statistics of conservative scalars in the convective boundary layer. C. Phys. 769 pp. Statistical Fluid Mechanics... 874 pp. 577-595. Verhoef. 1984. Cambridge. Using a large-aperture scintillometer to measure absorption and refractive index fluctuations. Chichester. Dutton. Agric. The fair weather boundary layer in GATE. C.. 1998. De Bruin. Victoria. USA. Monin. A.. Heusinkveld.. Vol 1.. Conclusions from ABRACOS.. Forest Meteorol. J.. 1980. Vol 2. 1997. in Amazonian Deforestation and Climate. J. J. L. Annu. 1996. C. H.. Roberts. Atmos. MIT Press. Hydrol. A... B. M. G. C. Perrier. A. R. 1986. P... J.. Atmospheric Turbulence. Roberts and R. 153 . 189. Moncrieff. Monin. J. Kabat.. A. A.. A. A. J. 147-169.. Yaglom.. Panofsky. Soegaard. John Wiley. H. Boundary-Layer Meteorol. A. W. Estimating heat storage in Amazonian tropical forests. LeMone.. 1506-1519. 36. T. Math.. M. A. Yaglom.. A. Cambridge. Sci. E. Nicholls. A. Rev... J. Nobre.. Friborg. 397 pp. V. S. 101-116. M. Victoria (eds.. 589-611... C.... Philipovitch. A. Massheder.. Green. C. R.. J. B. 1995. S. P.. John Wiley & Sons. Nieveen. Elbers.. MA. S. MIT Press.
doi:10. J.. Plana-Fattori. D. M. O. 1-16. M. Rutledge... Marengo. 1983. 2997-3012. C. A. R. A. Gash. Rocha.. P. E. A. D. Garstang.. de O. R.. A.. K. H. R.. Fuentes. Ribeiro.. C... A.. N. K.. 1996.. Silva. C.. Q. Scale Dependence of Air-Sea Fluxes over the Western Equatorial Pacific. A. Artaxo. 26. L. Sakai. Patel.. J. R. LeMone. K. Schotanus.. 1988.. Marques. H. 182p. L.. Importance of low-frequency contributions to eddy fluxes observed over rough surfaces. J. J. C. S. Freitas. Lloyd. Zipser. S. A. Silva Figueira. State University of New York at Albany. Ecological Applications. Pinto. J. C.. F.. 14(4) Supplement.. Grossman. Atmos. Meteorol. Sá. 2178– 2192. V. Roberts. C. De Bruin.. M. Manzi. Manzi.. Howell. R. F. S... L.. Moraes. 107 (D20). Dolman. A. 2004. M.1029/2001JD000335.. G. 2000. H. J Appl Meteor. de P. A. H. D.Rocha.. Molion.. Silva Dias. Andreae. Climatol. J. P.. Observational study of turbulent exchange between the surface and canopy layer over several forest types. M. 8072. R.. Moore. C. Mahrt. 40. C.. C. Nobre. J. M. 81-93.. Y. Eddy correlation measurements of energy partition for Amazonian forest. C.. J. K. R. R. Sakai. D. L.. B. D. D.... 110. 1143-1162. O. Miller. J... Energy flux partitioning over the Amazon forest. L. Kabat. Temperature-Measurement with a Sonic Anemometer and Its Application to Heat and Moisture Fluxes. D. P..... 1984. Gatti.... Fitzjarrald. 2002: Clouds and rain processes in a biosphere-atmosphere interaction context in the Amazon Region. Nieuwstadt. 2001. J. Cabral. L. Res. Sun. Theor. M. S.. A.. Moore. J. R. W. M. Fisch. Fisch. Viswanadham. P... 39. Menton. C. Soc. Sá. Alvalá. A. Geophys. O. PhD Thesis.. Sá. V.. Gielow. M. G. 53 (21). R. A... A. Greb.. O. S.. T. H. J. Appl. BoundaryLayer Meteorol. S22–S32. Sci. E. R. M.. R. Silva Dias.. O. Esbensen.. L.. Shuttleworth. F. Seasonality of water and heat fluxes over a tropical forest in eastern Amazonia... M.. Goulden. Nobre. M. J. L. 154 .
L. Free Convection in Turbulent Ekman Layer of Atmosphere. 2000. Mahrt. Andreas.. Wang. D. J. Oceanic Technol.. 27. 11. Boundary Layer Meteorol. D. 1961.. 1997. Oncley. Stull.. J. H. 29. Prueger. J. 1976. P. 1970. H. I. Oceanic Tech... Crawford. Meyers.. Kustas.0007 to 900 cycles per hour.. 271-288. Applied Meteorol. 2003. Twine. Treviño. 660-672.. Correcting eddy-covariance flux underestimates over a grassland.. Kluwer academic publishers. and Desjardins. J. 1027-1034. Sci. S.. 1972. Power spectrum of horizontal wind speed in the frequency range from 0.. Agric... P. An introduction to boundary layer meteorology. I. J. Quality Control and Flux Sampling Problems for Tower and Aircraft Data. M..... The effect of introducing an angle of attack dependent calibration. J.. 11. 160-164. Q.Sun. P.. Atmos. L. 2004. J. Agric.. Gash. H. Tillman. L. 14. K. Norman. The Structure of Turbulent Shear Flow. 1997: Lake-induced atmospheric circulations during BOREAS. Sci. 1957. 102 (D24).. Vickers. Press. J. C. H. Vickers. J. Meteorol.. E.R. Cook. D. A.. Elbers. Van der Molen. Townsend. L. 279-300. M. 512-526. R. 95–109. J. T. A.. 1988. 97-120. 122. D. Townsend.. T.. MacPherson. Cambridge. The indirect determination of stability..B.. Meteorol. Dordrecht. 14. A. Mahrt. heat and momentum fluxes in the atmospheric boundary layer from simple scalar variables during dry unstable conditions... Atmos.. U.. Wesely. 429 pp. R. J. Atmos. Van der Hoven. K.. 20. Mahrt. On Wavelet Analysis of Nonstationary Turbulence. L. For.. J. T.. Atmos. 155 . 1996.. Lenschow.. J. Davis. P. 783-792. A. For. G.. The cospectral gap and turbulent flux calculations. Starks.155-29. J. Geophys Res. J. W. 666 pp. L. 81 (3-4). Fluid Mech.166. Equilibrium layers and wall turbulence. P. R. A... L. 103. E... M. R. J. Sonic anemometer (co)sine response and flux measurement: II. J.. Houser. E. Tennekes. Dobosy. C.
Campelo Jr. 1995. A. 2004... M. 1976b. Oliveira. J. P. M. 5-26. J Climate.. J. B. 15... J. J. J. An improved Land Surface Parameterization Scheme in the ECMWF Model and its validation. J. 1187-1202. Zamparoni. The combined effect of temperature and humidity on the refractive index. Zanchi. A. Boundary-Layer Meteorol. Applied Climatol. T. A. Kraus. 8. 1992.192–207.. Hoegel. J. M. Kabat. Res. 1177–1188. C... Kruijt. H. 53. C. L. Kruijt. 2716-2748. C. Williams. Meteorol.. von Randow. Manzi. C. Kruijt... 2004. P. A.. G. L. A. R. von Randow. 141.. P. L.. Vourlitis. Weaver. G.. Scale variability of atmospheric surface layer fluxes of energy and carbon over a tropical rain forest in southwest Amazonia – I.. L. Wright... 43-49. Raiter. Atmos.. Cardoso. F. Wesely...1029/2001JD000379. Comparative measurements and seasonal variations in energy and carbon exchange over forest and pasture in South West Amazonia.. Shuttleworth. Dry season 156 . P.. A. Gannabathula. F. R. H. 78. 15. M. O. H.Viterbo. I. P. Meteorol. D. Transport Processes in the Tropical Warm Pool Boundary Layer. Gash. de S. da Rocha. Gash. Appl. Theor. 2002. C.. C... H. Ecological Applications. Low-frequency modulation of the atmospheric surface layer over Amazonian rain forest and its implication for similarity relationships. 53 (8)........ O. Elbers. Waterloo. G. 8062.. A. R. Effects of meteorological variations on the CO2 exchange of a Brazilian transitional tropical forest. L. F. L... B. M. P. M. Sa. J. M. J. A. Geophys. G. Beljaars. A. 1990. R.. Temperature and Humidity Flux-Variance Relations Determined By One-Dimensional Eddy Correlation.. Carvalho.. 2006. Holtslag. B. G. J. Meteorol. S.. J.. B. A comparison of two optical methods for measuring line averages of thermal exchanges above warm water surfaces. Arlino. Hacker. W. P. 14(4) Supplement. Nobre. Hayashi. W. 1996.. N. For. S89– S100. 1976a. Nogueira.. P. 107(0). von Randow. H. 77-91. R. Part I: Spectral Composition of Fluxes. Agric.. Diurnal conditions.. H.. M.. Filho. Hodnett. Sci. A. Appl. J. A. J. doi: 10. Maitelli.. C.. Silva. M.. M.. Manzi. C. Wesely. A. J.
. N. J. A. Wyngaard. Atmospheric Turbulence. C.. C. Modelling surface conductance for Amazonian pasture and forest. A. Roberts. 473-504. J. J.. 1083-1099.. Wright. Wyngaard. Rev.. 1177-1188. D. S. 1995..). H. Behaviour of the refractive index structure parameters near the ground.. Gash. C. Wright I. Meteorol. J. J. I. H. 52.. H. Chichester. S. da Rocha. Sci. H. Soc. V... Towards a GCM surface parameterisation of Amazonia. C. Y. Victoria (eds. Atmos. Nobre. Dynamics and Energetics of Convective Plumes in the Atmospheric Surface Layer. J. Opt.. Victoria (eds. M. Annu. J. Gash... E. 15. 438-458. in Amazonian Deforestation and Climate.. C. Y. Hodnett.. Roberts. Am.. John Wiley. M. L. Roberts and R. Gash. 205-233. Tomasella. J. C. A. Vertamatti. Q.. Zhuang. H. Culf. R.). J.. John Wiley.. Roberts and R. R. A.. J. C. R. H. L.. Nobre. 1992. J.. 1712-1722. C. Collins Jr. G. Chichester. in Amazonian Deforestation and Climate. J. Nobre. Izumi. 1996.. Ubarana. 24.micrometeorology of central Amazonian ranchland. R. 1971. J. R. 118. da Rocha. 157 .. C. Alvala. Soc. 1996. A. Fluid Mech. R.
Brazil. Utrecht) and Dr. logistics and bureaucracy matters. and The Netherlands Foundation for the Advancement of Tropical Research (WOTRO). Han Dolman (VU Amsterdam). who was always open for discussions and always willing to contribute to the research. Dr. Bart van den Hurk (Univ.Acknowledgements “There are places I'll remember all my life though some have changed Some forever not for better. Prof. First of all. Much more than this. I am also grateful to my promotor Prof. Thanks for your nice suggestions that certainly improved the quality of this thesis. Dear reader. and all contributed to the work reported here. he wouldn’t tell me to follow his path. for Bart’s enthusiasm for science might be contagious. Thanks to my doctoral committee composed by Prof. 159 . Pavel Kabat (WUR). Prof. Everyone listed below (and maybe some others that for some unfortunate reason are not mentioned) has. Bart Kruijt. for financially supporting my PhD project. for his dedication for this project. This work (and the work of many others) would never been complete without the support of many people from the LBA program in the field work. Whenever I got stuck in the research. influenced my life in the last 4 years. Bert Holtslag. some have gone and some remain…” (“In my life”. The Beatles) Although only one name appears in the cover of this thesis. Antonio Manzi certainly deserves most of the credit for leading the LBA office with dedication and competence. I also acknowledge Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES). Leonardo Sá (Museu Goeldi. but encouraged me to find my own. guidance and inspiration along these 4 years. to some extent. Belém) for carefully reading the thesis. I am grateful to my supervisor and friend Dr. he deserves most of the credit for guiding me in the beginning of my scientific career and for teaching me how scientific research should be carried out with rigor and honesty. it was actually written by many people. watch out.
don’t worry. Rubia. Glenda and Ana Luiza. I thank Maria Betania de Oliveira and Hermes Xavier for their support with the LAS. Catharien. Betânia. Moreover. for several Sundays. and party everyday!). Many many thanks for countless sleep-over weekends.Special thanks go to Alessandro Araujo and Dr. Mario. But I must say: whenever I had to explain where I worked. I had a hard time. Herbert. Isabel. I’m sure life in Wageningen would be much more boring! I lift a toast to celebrate our friendship. The LBA staff in Manaus is also acknowledged for the current maintenance and constant improvement of the K34 field station. and many moments of great fun during coffee-breaks. 160 . several trips we’ve taken. Judith. Ronald. Ana Claudia (adoro vocês!). Jan. U2 concerts … . Marijn. we also kept moving around. Petra. Zé Márcio (I want rock n’ roll all nite. Eddy.-) I am very fortunate for also meeting good friends that will never be forgotten. Vassilis. Michael. especially because of the group I’ve joined in Wageningen University and Research Centre. Olaf. In particular. ice-creams. team uitjes. But it doesn’t matter. Isabella (ermã. Anabele. Antonio Nobre. whose work in the field data collection and maintenance of the K34 site in Manaus during several years were invaluable. René (to whom not only me. Ann-Marie. Ji-Paraná. And not only changing names. and others. be happy). Saskia. I am also in debt to the field crew from Unir. without you guys. Alessandro Araujo has been my brother. barbecues/parties we had and Formula-1 races we watched together. Jeroen. Lu Soler. (And thanks a lot for providing me. Flávio (best cook ever!). Working in the Netherlands was definitely a great pleasure. I actually can’t tell precisely. Glaciela (the best mother little Valentin could have). especially Fabricio Zanchi and Paulo Jorge Oliveira. More than just a friend. and his family has been my family (Jacqueline. but the whole Brazilian community in Wageningen is thankful. Was it Wageningen University? Alterra? Center for Water and Climate? Land atmosphere Interactions Team? Team Moors? (the latest is Earth System Sciences – Climate Change… be aware this might change in a few months). Wilma. Marcia (e aí. for great barbecues and parties in the sports center canteen). for the hard work operating the Rebio Jaru and FNS sites. Hester. vamô lá?). guria. my dear friends Odair (O-da! O-da! O-da!).). Fons. What will always be on my memories are the great people I worked with. André (who provided my first room in NL). Cor. with the only decent meal I had during the week. my dear ‘Familia Tra-la-lá’). May it last forever.
Rita de Cassia Silva von Randow. “(…) Though I know I'll never lose affection for people and things that went before I know I'll often stop and think about them… in my life I’ll love you more” (“In my life”. for their love and encouragement through all steps of my way. And finally I thank my beloved (soon-to-be and forever) wife.I’m also forever in debt to my parents Roberto (in memorian) and Maria do Carmo. I’m glad to always feel their support even across 10000 km distance. The Beatles) 161 . I look forward to the life we’re building together. for fulfilling what was missing in my life. and to my brothers and sisters.
About the author Celso von Randow was born in Belo Horizonte. being awarded with a scientific initiation grant from CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico). with the Earth Systems Science – Climate Change (ESS-CC) group. to work on field measurements of turbulent fluxes and analyses of a surface-atmosphere interaction model. he had his first contact with research on micrometeorology and surfaceatmosphere exchange processes. In 1994. E-mail: cvrandow@gmail. and it was a natural follow-up from this collaboration to pursue a PhD study in Wageningen University.com 163 . Brazil. 1976. he started working at CPTEC (Centro de Previsão de Tempo e Estudos Climáticos). Brazil. Wageningen. During the undergraduate course. on February 21st. With financial support from CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior). he worked on his PhD from December 2002 to April 2007 at Alterra/WUR. and from NWOWOTRO (The Netherlands Foundation for the Advancement of Tropical Research). Brazil. he joined the Bachelor in Meteorology program at the University of São Paulo. His work at CPTEC had strong collaboration with researchers from Alterra. including field experiments in Amazonia. where he developed many scientific activities related to the LBA program (Large Scale Biosphere Atmosphere Experiment in Amazonia). After graduating in 1998. in collaboration with the Meteorology and Air Quality group.
Statement of the Research School Utrecht. Director of the Buys Ballot Research School 165 . February 2007 Dear Mr.D. Prof.ir. von Randow.dr. Opsteegh. Best regards. 19. J. I wish you success with the defense of your PhD thesis. It is a pleasure for me to confirm that you have fulfilled all educational activities required by the Buys Ballot Research School with good results.
C. and by The Netherlands Foundation for the Advancement of Tropical Research (WOTRO). project number WB-76 224. von Randow’s PhD research was financed by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES). Brazil. Cover illustration: site locations photographs by Maarten Waterloo (FNS tower). project number 1271/01-6. 166 . Ana Claudia Lessa (Manaus K34 tower) and Antonio Manzi (Rebio Jaru tower).
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.