Professional Documents
Culture Documents
canopy dynamics at the rate limited by overpass frequencies and various other factors.
Captured radiometric signals are transmitted to the ground stations and processed into
inputs for geophysical models of net primary productivity (NPP – main characteristic of
vegetation biomass growth - Field et al 1998) and radiative (light) transfer in the canopy.
PAR=IPAR+TPAR (1)
IPAR=APAR+RPAR (2)
where PAR - photosynthetically active radiation (total), IPAR - PAR intercepted by the
canopy, TPAR – PAR transmitted through the canopy, APAR PAR absorbed by the
canopy, RPAR -PAR reflected by the canopy, ƒIPAR, ƒRPAR, ƒTPAR and ƒAPAR –
fractions of IPAR and APAR in PAR, ε* – maximum light use efficiency (LUE), g(T) –
temperature effects, h(W) – water stress effects, k – light extinction coefficient, LAI –
leaf area index (Field et al 1998, Nouvellon et al 2000). The fAPAR, LUE and LAI are
the ultimate candidates for inclusion. Surface temperature, and soil/surface moisture
content often have well pronounced seasonality and should contribute in handling g(T)
and h(W) issues influencing LUE. Close attention should be paid to presence and
duration of cycles in the dynamics of chosen variables. Surface properties with cycles
shorter then one growing season should be selected as they contribute to the within
season variability. Otherwise, the dynamics related to growing season is overlaid on the
bigger cycles. Hardly, can canopy height and fraction vegetation cover cycles appear
within growing season for coniferous or rain forests but for grasslands and arid
There are numerous methods for estimation of fraction of PAR absorbed by the canopy
(ƒAPAR) in the literature, but most of them rely on vegetation indices obtained from
reflectances in the visual and NIR portion of spectrum. Strong linear relationship of
ƒAPAR and NDVI made this index the most popular for the purpose. Yet, the relationship
differs for different instruments and land cover types, depending on sensor band location,
timing in the season (changes in canopy structure) and fraction of vegetation cover
Gobron et al (1999) used their own BRDF adjusted MGVI with uniform linear
relationship to ƒAPAR across the globe. In the production of MODIS MOD15A2, the
main and backup algorithms are used to keep continuous records. Main algorithm is
based on radiative transfer logic, and backup algorithm – on the empirical relationship
with NDVI (Knyazikhin et al 1999). In the near future, MODIS science team plans to
transfer the production to the use of linearly related EVI in the ƒAPAR backup algorithm.
Similarly to ƒAPAR, the most popular estimation methods of leaf area index rely on
empirical relationships with vegetation indices. There is much criticism for this approach
in the literature, especially for the use of NDVI that looses sensitivity when LAI reaches
2. Wide dynamic range vegetation index (WDRVI) was developed by Gitelson (2004)
and performed better for LAI values 2-4 (Vina et al 2004). In 2006, Deng et al used
reflectances in RED, NIR and SWIR bands to derive simple ratio(SR) and reduced simple
ratio (RSR) indices. Authors claim that optimized for land cover type and tuned with
BRDF parameters the relationships between two indices yield estimates of LAI that are
close to inversion of four-scale canopy reflectance model (Chen & Leblanc 1997).
transfer theory in canopies modeled for six different biomes (Knyazikhin et al 1999).
MODIS LAI/ƒAPAR product is the only source of global moderate spatial resolution
continuous records of ready to use ƒAPAR and LAI. However, there are several archives
–AVHRR, MERIS, VEGETATION – of vegetation indices and raw reflectance data that
Unlike ƒAPAR, leaf area index can be estimated from microwave measurements both
active and passive. In empirical retrieval methods, LAI is related to weighted mean
(proportional) and K-a band (inversely proportional) (Palocia 1995.) Radiative transfer
soil emissions (Renzullo et al 2006.) In this context, there are three ways of treating the
canopy (Kerr and Wigneron 1995.) First one has it as continuous homogeneous medium
characterized by opacity and single scatter albedo. Second splits the scattering into four
components representing leaves, branches, trunks and stalks. Third – the most complex
one – has the vegetation layer consisting of discrete plants/trees and takes into account
length and orientation of the four scattering components. The main concern here is
volumetric plant water content (PWC) of the canopy. Simple models have it directly
proportional to canopy opacity [τ=bPWC], but the coefficient b varies with incidence
angle and time in the season. PWC is well related to Microwave Polarization Difference
Index (MPDI) for some microwave bands (Wigneron et al 2003). Corrected with factors
that influence b, volume fraction of vegetation (VFV) can be estimated from MPDI
measurements. During the greenup phase, VFVis highly correlated with LAI (Wigneron
1995.) Global scale satellite retrievals of LAI are problematic for passive microwave
sensors because of large footprint and heterogeneity within one pixel. However, the
synergistic methods with visible and infrared data are being actively developed for
Light use efficiency is a property of green vegetation that summarizes the conditions for
energy (Guo & Trotter 2004.) Retrieval of LUE by the means of remote sensing not as
well developed as retrievals of LAI or ƒAPAR. In the ground based measurements LUE
(measured in mol CO2 per mol of photons) has found to be related with Photochemical
greenness indicators such as NDVI and red edge chlorophyll concentration index (Sims et
al 2006). In 2005, Drolet et al tried to come up with PRI algorithm for MODIS. Just for
having a better correlation, their attempt ended up with using top of the atmosphere
reflectances of bands that are far from those specified in Penuelas et al (1995). However,
they recognized the difficulty of atmospheric correction for the bands used in PRI. As a
remote sensing product light use efficiency has not been produced on the global scale.
LUE is specie specific and depends on many factors that including soil water availability,
Retrieval of the land surface (skin) temperature (LST) from satellite observations has
been done by many satellite missions including solar orbiting AVHRR and MODIS (Jin
M, 2004.) In the latter sensor the LST is obtained by inverting Planck’s Law equation for
several bands including ones in thermal diapason and by correcting it for known
emissivity (Wang 1999.) Jin (2004) identifies several problems with LST retrievals for
NOAA AVHRR and other polar orbiting platforms. The first problem is the drift of the
satellite’s orbit that causes inconsistency in timing for day and night observations. The
second problem is unknown emissivities of the observed surfaces that are highly variable
in space. The third complication is the atmospheric contamination that prevents direct
measurement of LST. Viewing geometry is the last but not the least source of
uncertainties for LST retrieval from orbiting equipment. In the production of MOD11
level 2 and 3 products the window split algorithm corrects for atmospheric and emissivity
effects with regard to viewing geometry (Wang 1999.) To capture diel dynamics, day and
night acquisitions are made by both Terra and Aqua. However, the products are produced
with daily, 8 day and monthly temporal resolutions. There is a possibility to obtain land
surface temperature from geostationary satellites such as GOES that have higher
acquisition frequency, but sacrifices include limitation in coverage and low spatial
resolution. The resolution is also sacrificed when estimating LST from microwave
Monitoring of soil moisture content (SMC) in the thin (1-5 cm) top layer of soil is well
spectrum. The basic principle of retrival is that observed brightness temperature of soil
varies with the water content (Wigneron et al 2003.) In addition to that, the discrepancy
between emissions with horizontal and vertical polarization from bare soil is higher than
the same discrepancy for emissions from vegetation (Paloscia 1995.)Vegetation related
effects on brightness temperatures are addressed in the same three manners described in
the LAI retrieval via microwave section. As an empirical shortcut, Wigneron (1995)
brightness temperatures. Some studies used vegetation indices such as PVI and NDVI for
this purpose (Wang and Choudhury 1995). Paloscia (1995), simply recommend using
lower frequencies that are less affected by the green foliage. The Antecedent Precipitation
Index (API) has been a popular empirical indicator of soil moisture in studies that worked
with spaceborne sensors (Wigneron et al 2003.) Over the last two decades, the index went
through numerous modifications changing the microwave bands and variables that
compensate for vegetation opacity. In the same review Wigneron mentions a cohort of
methods of surface moisture retrievals based on regression, neural networks and lookup
tables. The author points out that these stocastic models suffer form inability to address
the issue of spatiatial variability and can be used for local/regionals studies only.
The utility of remotely sensed data in thermal, infrared and visible portions of spectrum
for retrieval of SMC related parameters has also been studied extensively. Sandholt et al
(2002) used so called “land surface temperature/ vegetation index surface space” to
resistance to evaporation (rc). They assumed that rc is driven mostly by soil moisture
availability and reported high negative correlation between TVDI and SWC. Though,
their methodology is scene limited and hence suffers greatly from heterogeneity of land
cover within the scene. Studies of Gao (1996), Serano et al (2000), Dennison et al (2003),
Sims and Gamon (2003) used vegetation water content indices derived from NIR
based on the same principles but took into account some additional factors. They also
proposed the evepotranspiration product for MODIS land suite where they had
evaporation split in two components -vegetation and soil. The latter one is clearly a proxy
of soil moisture content. There is no operational SMC product with global coverage.
Studies that involved large scale SMC in data assimilation experiments (Walker and
Houser 2001, Renzullo et al 2006) had their surface moisture observations simulated. Yet
with limited consistency, the product can be generated using data from SMMR, AMSR,
Monitoring of temporal variations of percent vegetation cover (PVC) and canopy height
(CH) within one growing season has not been considered for development /
implementation at global scale yet. The methodologies used currently for estimation of
PVC and CH are probabilistic and rather categorical (close to those of land cover
mapping) (Hansen, 2003) and therefore they use the whole year of remotely sensed data
vegetation types and their particularities in spatial distributions of crown and canopy
deciduous, non-tree, shrubs, crops and other herbaceous vegetation fields presenting
those in great spatial details. However, the temporal variation is lost as well as canopy
height information. The MOD44B production does not use a deterministic canopy model
and there is no radiative transfer model to invert or even proxy variable to relate PVC to.
However, there have been many alternative attempts to derive PVC and CH and some of
that experience can be used to generate observations of canopy height and percent
vegetation cover.
Usually canopy height is the task for active remote sensing - radar and lidar sensors, but
those can hardly be used on the global scale with high temporal resolution. Correlation of
PVC and CH with other canopy characteristics brought the idea of using similar passive
remote sensing methods with care for species/biome and pheno-phase dependency of
these relationships. Shoshany et al (1996) proposed using reflectances in red and infrared
bands to obtain percent vegetation cover with empirical linear model. A threshold based
GVI product suite. These methods, as well as derivation of percent cover based on linear
mixing of vegetation indices (Purevdorj et al 1998, Montandon and Small 2008) has yet
several problems to overcome - sensor dependence, land cover dependence and soil type
effects.
Talking about all these vegetated surface properties, numerous vegetation indecis (VI)
were mentioned almost in every paragraph. Therefore, the reason dictates authors to
provide some insight on VI matter. The evolution of spectral vegetation indices can be
partitioned into three stages or generations. First stage - the indices are developed based
transformations on spectral band responses that are some how related to certain
biophysical characteristic of the foliage. The important moment here is that the
relationship with surface attributes is the one and only utility of indices. They do not
measure the actual physical property, rather they capture the relative dynamic of the
canopy development. After the index is tested in multiple studies, the second stage
comes, where the original index is corrected for encountered problems. Basically, it
evolves into a new index (next generation) that utilizes sensitivities of other bands to the
setbacks of original index. Numerous examples of this include ARVI (Kaufman and
Tanré 1992), SAVI (Huete 1988), GEMI (Pinty and Verstraete 1992 ) and AFRI (Karnieli
relationships to the core biophysical variables. Such resolution - adding and tuning the
correcting coefficients - was used in EVI (Liu and Huete 1995) and WDRVI (Gitelson
2003). A more sophisticated approach came from Gobron et al (1999) where they used
MERIS sensor. A similar approach was taken by Deng et al (2006) for alternative
estimation of LAI from VEGETATION sensor reflectances. In 2000 paper, Gobron et al
showed that vegetation indices from different sensors derived in this manner give
dependence of index values on band characteristics but the tuned slope makes index
derived product less dependent on the sensor characteristics. The linearity in relationship
with actual physical property of the surface gives a great advantage and simplicity to the
third generation indices for usage in data assimilation. It opens the possibility for efficient
fusion of products from multiple sensors in derivation of one land surface variable