You are on page 1of 18

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/230546542

Satellite‐based ET estimation in agriculture


using SEBAL and METRIC

Article in Hydrological Processes · December 2011


DOI: 10.1002/hyp.8408

CITATIONS READS

69 1,242

6 authors, including:

Richard G Allen Ayse Kilic


University of Idaho University of Nebraska at Lincoln
194 PUBLICATIONS 12,961 CITATIONS 39 PUBLICATIONS 643 CITATIONS

SEE PROFILE SEE PROFILE

Ricardo Trezza W.G.M. Bastiaanssen


University of Idaho UNESCO-IHE Institute for Water Education
66 PUBLICATIONS 1,839 CITATIONS 212 PUBLICATIONS 8,773 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ricardo Trezza on 03 January 2017.

The user has requested enhancement of the downloaded file.


HYDROLOGICAL PROCESSES
Hydrol. Process. 25, 4011–4027 (2011)
Published online in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/hyp.8408

Satellite-based ET estimation in agriculture using SEBAL and


METRIC
Richard Allen,1* Ayse Irmak,2 Ricardo Trezza,3 Jan M. H. Hendrickx,4 Wim Bastiaanssen5,† and
Jeppe Kjaersgaard6
1
Kimberly Research and Extension Center, University of Idaho, Kimberly, ID, USA
2
Civil Engineering/SNR, University of Nebraska Lincoln, Lincoln, NE, USA
3
Kimberly R&E Center, University of Idaho, Kimberly, ID, USA
4
Hydrology, New Mexico Tech, Socorro, NM, USA
5
Faculty of Civil Engineering and Geoscience, Delft University of Technology, Delft, The Netherlands
6
Biological Engineering, South Dakota State University, Brookings, SD, USA

Abstract:
Surface Energy Balance Algorithms for Land (SEBAL) and Mapping EvapoTranspiration at high Resolution with Internalized
Calibration (METRIC) are satellite-based image-processing models that calculate evapotranspiration (ET) as a residual of a
surface energy balance. Both models are calibrated using inverse modelling at extreme conditions approach to develop image-
specific estimations of the sensible heat flux (H) component of the surface energy balance and to effectively remove systematic
biases in net radiation, soil heat flux, radiometric temperature and aerodynamic estimates. SEBAL and METRIC express the
near-surface temperature gradient as an indexed function of radiometric surface temperature, eliminating the need for absolutely
accurate surface temperature and the need for air temperature measurements. Slope and aspect functions and temperature lapsing
are used in METRIC applications in mountainous terrains. SEBAL and METRIC algorithms are designed for relatively routine
application by trained professionals familiar with energy balance, aerodynamics and basic radiation physics. The primary inputs
for the models are short-wave and long-wave (thermal) images from satellite (e.g. Landsat and MODIS), a digital elevation
model and ground-based weather data measured within or near the area of interest. ET ‘maps’ (i.e. images) developed using
Landsat images provide means to quantify ET on a field basis in terms of both rate and spatial distribution. METRIC takes
advantage of calibration using weather-based reference ET so that both calibration and extrapolation of instantaneous ET to 24-h
and longer periods compensate for regional advection effects where ET can exceed daily net radiation. SEBAL and METRIC
have advantages over conventional methods of estimating ET using crop coefficient curves or vegetation indices in that specific
crop or vegetation type does not need to be known and the energy balance can detect reduced ET caused by water shortage,
salinity or frost as well as evaporation from bare soil. Copyright © 2011 John Wiley & Sons, Ltd.

KEY WORDS evapotranspiration; remote sensing; energy balance; SEBAL; METRIC

Received 2 July 2011; Accepted 9 November 2011

INTRODUCTION heat fluxes can severely impact the performance of optical,


infrared and acoustic sensors used by military, and ET and
Quantifying the consumption of water over large areas and
soil moisture affect operational mobility of military
particularly within irrigated agricultural areas is important
operations and detection of landmines and unexploded
for water resources planning, mitigation of impacts of
ordinance (Miller et al., 2004; van Dam et al., 2005).
reduced streamflow, establishment of hydrologic water
Traditionally, ET from agricultural fields has been
balances, water rights management and water regulation.
estimated by multiplying a weather-based estimate of
Differences between actual and potential ET at high spatial
reference ET by a crop coefficient (Kc) determined according
resolutions are of interest to agriculture, water resources and
to the crop type and the crop growth stage (Allen et al., 1998;
even national security as an indicator of crop water deficits
ASCE-EWRI, 2005). The reference ET estimate approx-
and depletions on scales of human activities and individual
imates an upper limit of ET from an extensive surface of
land holdings. Spatial estimates of ET are essential
healthy vegetation, fully covering the ground and not short of
components of general circulation and hydrologic models
water. This is especially the case with the ‘tall-reference’ of
(Wigmosta et al., 1994; Betts et al., 1997; Overgaard et al.,
ASCE-EWRI (2005) that approximates ET from full-cover
2006), and ET is used to infer soil moisture, a valuable input
alfalfa. In the case of the ‘short-reference’ of ASCE-EWRI
to weather and flood forecast models. Sensible and latent
(2005) and FAO (Allen et al., 1998), which approximates ET
from a short (0.12 m) tall cool-season grass cover, the
*Correspondence to: Richard Allen, University of Idaho, Kimberly reference ET is typically 20–30% lower than the maximum
Research & Extension Center, Kimberly, ID, USA. E-mail: rallen@ ET of taller and leafier vegetation. The Kc incorporates all
kimberly.uidaho.edu

Present address: eLEAF, Generaal Foulkesweg 28, 6703 BS, Wageningen, physical and time-based factors that cause a specific
The Netherlands. E-mail: wim.bastiaanssen@eleaf.com vegetation or surface to have different ET and surface energy

Copyright © 2011 John Wiley & Sons, Ltd.


4012 R. ALLEN ET AL.

partitioning than the reference surface. There is typically semi-arid locations. This phenomenon is illustrated in Figure 1,
some question whether actual vegetative and growing where the measured ET of alfalfa is plotted with 24-h net
conditions compare with the conditions represented by the radiation. In some cases, the ratios of ET to Rn, i.e. the EF,
idealized Kc values, especially in water short areas, where increase to more than 2.0 during late summer when advection
stomatal regulation can reduce ET to below potential of energy from desert areas upwind of the irrigated area
values for the particular physiological and phenological surrounding Kimberly, Idaho, substantially increases ET. The
conditions. In addition, it is often difficult to monitor or lysimeter measurements shown in Figure 1 can be reproduced
estimate the crop growth stage conditions over large using the ASCE-EWRI Penman–Monteith equation.
populations of crops and fields. The METRIC and SEBAL models are designed to
Satellite data are ideally suited for deriving spatially function as operational models for producing maps of ET
continuous fields of ET using energy balance techniques. for focused regions smaller than a few hundred kilometres in
Surface Energy Balance Algorithms for Land (SEBAL) and scale and at high resolution. This contrasts with some remote
Mapping EvapoTranspiration at high Resolution with sensing models that are more generally based and are
Internalized Calibration (METRIC™) are satellite-based designed for routine application over larger regions (for
image-processing tools that calculate ET as a residual of the example, over subcontinents), such as the Atmosphere–Land
energy balance at the earth’s surface. METRIC has, as its Exchange Inverse (ALEXI) model (Anderson et al., 2004,
foundation, principles and techniques used by the SEBAL 2007) and the feedback method of Granger (1989, 2000). The
model developed in the Netherlands and applied worldwide narrowed focus provides estimates of ET at relatively high
by Bastiaanssen and his associates (Bastiaanssen, 1995, resolution (~30 m), but does come at a cost, in the requirement
2000; Bastiaanssen et al., 1998a, 1998b, 2005). The for trained experts having good background in energy balance
innovative component of SEBAL is that the energy balance and radiation physics and adequate knowledge of vegetation
modelling uses a near-surface temperature gradient, dT, characteristics and agricultural practices.
which is indexed to radiometric surface temperature, which
has eliminated the need for absolute surface temperature Theoretical basis
calibration, which has been a major impediment to The theoretical and computational basis of SEBAL is
operational satellite ET. METRIC uses the SEBAL described in Bastiaanssen et al. (1998a; 2005) and
technique for estimating dT, thereby eliminating the need Bastiaanssen (2000). The basis and principles for METRIC
for absolutely accurate aerodynamic surface temperature and are described in Allen et al. (2007a). Although satellites
the need for air temperature measurements for estimating routinely measure surface reflectance and some measure sur-
sensible heat flux at the surface. Both models are calibrated face temperature, none measure near-surface vapour content or
using inverse modelling at extreme conditions to develop vapour flux. Therefore, ET is determined by applyingan energy
image-specific estimations of the sensible heat flux (H) balance at the surface, where energy consumed by the ET pro-
component of the surface energy balance and to effectively cess is calculated as a residual of the surface energy equation:
remove many systematic biases in net radiation, soil heat
flux, radiometric temperature and aerodynamic estimates. LE ¼ Rn  G  H (1)
Energy balance based models such as METRIC and
SEBAL can have advantages over conventional methods of where LE is the latent energy consumed by ET, Rn is net
estimating ET from crop coefficient curves or vegetation radiation (sum of all incoming and outgoing short-wave and
indices in that crop development stages do not need to be long-wave radiation at the surface), G is sensible heat flux
known, nor do specific crop types. In addition, energy balance conducted into the ground, and H is sensible heat flux
can detect reduced ET caused by water shortage, salinity or convected to the air. Energy absorbed into the canopy and
frost as well as increased ET caused by evaporation from bare by photosynthesis is generally less than a few % and is
soil or water intercepted by plant canopies following wetting ignored in Equation (1). A utility of using energy balance
events. METRIC departs from the SEBAL model in its use of
weather-based reference ET to establish energy balance
conditions at a ‘cold’ pixel, as described later, as compared
with traditional SEBAL approaches that assume that all
available energy (composed of net radiation less ground heat
flux) is converted to evaporation at a temperature similar to
that of a local water body. The use of reference ET estimates
establishes a ground reference for the satellite-based actual
ET estimate and takes advantage of confidence in the
standardized Penman–Monteith equation (ASCE-EWRI,
2005) to accurately estimate the reference ET. The use of
reference ET in METRIC for extrapolation of instantaneous
ET to 24-h and longer periods compensates for regional
advection effects and does not tie the 24-h ET to an Figure 1. Comparison of 24-h evapotranspiration from alfalfa, as
measured by lysimeter, near Kimberly, Idaho, USA, (data by Dr J. L.
evaporative fraction (EF) that is based on net radiation Wright, USDA-ARS (ret.)) over a growing season and during periods of
because ET can exceed daily net radiation in many arid or effective ground cover with estimated 24-h net radiation

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4013

over vegetation-based methods is that actual ET rather than RS# from automated weather stations because of common
potential ET (based on amount of vegetation) is computed problems associated with measurements such as sensor
so that reductions in ET caused by stresses caused by level, calibration drift and cleanliness. Bulk transmissivity is
disease, salinity or shortage of soil moisture are captured. A calculated using a general function from ASCE-EWRI
disadvantage of the energy balance approach is that the (2005) according to the estimated water vapour content of
computation of LE is only as accurate as the estimates for Rn, the atmosphere, sun angle and air mass by using similar
G, and H. METRIC and SEBAL attempt to overcome this functional forms as for narrowband reflectances (Tasumi
disadvantage by focusing internal calibration on LE and et al., 2008). The estimation of transmissivity is only
using H to absorb intermediate estimation errors and biases. moderately sensitive to the estimate for atmospheric water
The METRIC and SEBAL models are, in some ways, vapour, especially when used in the inverse-calibrated
‘open platforms’ where algorithms for estimating components METRIC and SEBAL, so that precipitable water vapour, W,
such as Rn, G, emissivity, aerodynamic roughness, and so on can be estimated from measured or estimated near-surface
can be replaced with alternative algorithms at the user’s vapour pressure from a representative weather station,
preference. The primary distinction of the models is in the way according to the equation by Garrison and Adler (1990):
that they are inverse calibrated and in how they use surface
temperature as a means to distribute, spatially, air temperature
W ¼ 0:14ea Pair þ 2:1 (3)
gradients during calculation of H so that absolute accuracy
requirements of surface temperature can be relaxed. The
following section describes general forms of equations used where ea is near-surface vapour pressure (kPa) and W is in
for Rn, G and subcomponents of H to provide an indication of millimetre. Tasumi et al. (2008) demonstrated Equation (3) to
how these components can be formulated and estimated and estimate similarly to retrievals for W from Moderate-resolution
how operational applications have been formulated. Imaging Spectroradiometer (MODIS). Trigonometric
In METRIC, Rn is computed from satellite-measured equations for estimating the influence of sun angle, terrain
narrowband reflectance and surface temperature using slope and aspect on incoming solar radiation are given by
algorithms described in Allen et al. (2007a). G is commonly Duffie and Beckman (1991) and repeated by Allen et al.
estimated in METRIC and SEBAL as a fraction of Rn, on the (2007a).
basis of surface temperature, for low amounts of vegetation
and as a function of vegetation indices when vegetation Surface albedo. Surface albedo, the ratio of reflected solar
cover is present (Bastiaanssen et al., 1998a; Allen et al., radiation to the incident solar (short-wave) radiation at the
2007a). More recent METRIC applications have estimated surface, represents the integrated reflectance across the short-
G as a function of H following suggestions by Brutsaert wave spectrum (0.2 to 3.2 mm). Albedo is generally calculated
(1982) and Stull (1988). Sensible heat flux in METRIC and in METRIC and SEBAL by integrating surface reflectivities
SEBAL is estimated from surface temperature, surface from representative satellite bands that fall within regions of
roughness and wind speed using buoyancy corrections. high atmospheric transmissivity, for example, bands 1–5 and 7
of Landsat, bands 1–7 of MODIS and bands 1–9 of Advanced
Net radiation. Net radiation flux at the surface (Rn) Spaceborne Thermal Emission and Reflection Radiometer.
represents the radiant energy at the surface that is partitioned Weighting coefficients are applied to each band when
into H, G and LE. Rn is computed by subtracting all outgoing estimating albedo (Liang, 2000; Tasumi et al., 2008).
radiant fluxes from all incoming radiant fluxes and includes ‘At-satellite’ or ‘top of atmosphere’ bidirectional reflect-
solar and thermal radiation: ance (rt,b), is converted into at-surface reflectance, rs,b, that is
needed to calculate surface albedo. Atmospheric correction
for the narrowband reflectances is generally made using
Rn ¼ RS#  aRS# þ RL#  RL#  ð1  eo ÞRL# (2)
simple humidity and sun angle based algorithms reported by
Tasumi et al. (2008). It is assumed that the bidirectional at-
where RS# is incoming short-wave radiation (W m2), a is surface reflectance is similar to the directional–hemispherical
surface albedo (dimensionless), RL# is incoming long-wave reflectance required for the surface energy balance. This is
radiation (W m2), RL" is outgoing long-wave radiation generally a good assumption (Tasumi et al., 2008), except for
(W m2) and eo is broadband surface thermal emissivity tall vegetation such as trees or corn when observed by a nadir-
(dimensionless). The (1  eo)RL# term represents the looking satellite such as Landsat. In those cases, the satellite
fraction of incoming long-wave radiation reflected from may view a disproportionate amount of shadows, causing an
the surface. RS# is estimated as a function of time, day of underestimation of reflectance and albedo. Empirical
year, location, terrain slope and aspect, assuming clear-sky corrections are derived and applied in some applications on
conditions, because clear sky is a prerequisite to a useable the basis of field studies of nadir to hemispherical albedo over
satellite image. In applications to mountainous terrain, RS# corn, wheat and alfalfa by Allen, Zhao and Li (pers. comm.,
varies widely, spatially, and must be calculated from a 2011). These corrections can be based on bidirectional
digital elevation model and transmissivity estimate that reflectance distribution factors. Atmospheric correction can
varies with atmospheric thickness. Studies by Allen (1996) be performed using more sophisticated models such as
and ASCE-EWRI (2005) have shown calculated RS# for MODTRAN (Berk et al., 1999) that account for impacts of
clear-sky conditions to have similar accuracy as measured humidity, temperature and aerosol profiles on scattering and

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4014 R. ALLEN ET AL.

absorption. Those models require good background know- applications of MODTRAN during spring, summer and
ledge of the correction model and the software programme, autumn. Atmospherically corrected Ts products are soon to
which is not common to all ET model operators. As described become available as standard, universal products from the
later, in METRIC and SEBAL, small biases in rs,b resulting USGS EROS data centre.
from using a simplified atmospheric correction procedure are
largely corrected for during the internal calibration of sensible Incoming long-wave radiation (RL#). Incoming long-
heat flux. wave radiation is the downward thermal radiation flux
originating from the atmosphere (W m2) and is tradition-
Outgoing long-wave radiation. Outgoing long-wave ally computed in simple operational models such as SEBAL
radiation, RL", emitted from the surface is driven by and METRIC using the Stefan–Boltzmann equation:
surface temperature and surface emissivity. RL" is
computed using the Stefan–Boltzmann equation: RL# ¼ ea sTa 4 (6)

RL" ¼ eo sTs 4 (4) where ea is the effective atmospheric emissivity (dimen-


sionless) and Ta is near-surface air temperature (K). An
where eo is the broadband surface emissivity (dimension- empirical equation for ea by Bastiaanssen (1995) has been
less), s is the Stefan–Boltzmann constant (5.67  108 W applied in many applications of METRIC using coefficients
m2 K4) and Ts is surface temperature (K). Surface developed by Allen et al. (2000) using data collected over
emissivity is usually computed in METRIC and SEBAL alfalfa in Idaho:
using empirical functions based on estimated leaf area or
ea ¼ 0:85 ðlntsw Þ0:09 (7)
fraction of ground covered by vegetation (Allen et al.,
2007a). The range in both broadband and narrowband e is
where tsw is the broadband atmospheric transmissivity for
generally estimated to be small, lying between 0.95 for
short-wave radiation. The original coefficients by Bas-
many soil types and 0.97 for vegetation and water (MODIS
tiaanssen (1995), derived for western Egypt, were ea = 1.08
UCSB Emissivity Library, 2004; Hulley and Hook, 2009).
 (lntsw)0.265. Equation (7) works well with remote
Surface temperature (Ts) is computed for Landsat images
sensing of energy balance because the transmissivity term
using a modified Plank equation following Markham and
changes automatically with change in elevation for images
Barker (1986) with atmospheric and surface emissivity
having mountainous terrain and with atmospheric water
correction using a corrected thermal radiance from the
vapour when the term is estimated using algorithms
surface, Rc, based on spectral radiance Lt,6 from band 6 of
containing pressure and humidity components (ASCE-
Landsat, which is the thermal band. Rc is calculated following
EWRI, 2005). Other methods for estimating RL# or tsw can
Wukelic et al. (1989) as
be employed at the discretion of the user. The Ta parameter
Lt;6  Rp in Equation (6) is assumed to be highly correlated with
Rc ¼  ð1  eNB ÞRsky (5) effective (radiometric) atmospheric temperature. In some
tNB applications of SEBAL and METRIC, the surface
temperature, Ts, of each image pixel has been used as a
where Lt,6 the spectral radiance of Landsat band 6 (W m2 surrogate for Ta in Equation (6), which suggests that
sr1 mm1), Rp is the path radiance in the 10.4–12.5 mm band incoming long-wave radiation varies across an image in
(W m2 sr1 mm1), Rsky is the narrowband downward proportion to the underlying surface temperature. However,
thermal radiation from a clear sky (W m2 sr1 mm1) and this can cause estimates for RL# to vary over heterogenous
tNB is the narrowband transmissivity of air (10.4–12.5 mm vegetation by as much as 100 W m2 over short distances,
range). Units for Rc are W m2 sr1 mm1. Values for Rp and which is unrealistic because atmospheric conditions (and
tNB can be improved through the use of an atmospheric effective temperature) tend to be relatively homogeneous
radiation transfer simulation model such as MODTRAN. In within a region. Allen et al. (2007a) recommended
the absence of an atmospheric correction model, the standard estimating the Ta in Equation (6) from the ‘cold pixel’ used
corrections for Rc are invoked, on the basis of a central during calibration of METRIC because at that condition, H
calibration of Equation (5), by setting Rp = 0, tNB = 1 and is generally small and therefore Ta ~ Ts. This has been
Rsky = 0. If these standard values are used, then biases confirmed using incoming RL# measurements from desert
introduced in this step are accounted for during the calibration and irrigated settings (Allen and Kjaersgaard, 2009,
of the dT function as described later. Allen et al. (2007a) unpublished data and analyses). The cold pixel Ts value
found the use of uncorrected Rc, (Rp = 0, tNB = 1 and Rsky = 0) was found by Allen and Kjaersgaard to be a good predictor of
to have good accuracy in the range 260 < Ts < 285 K in the effective atmospheric temperature, even for desert
Idaho, with underestimation of Ts by about 3 to 4 K at Ts portions of an image. Therefore, the cold pixel Ts value is
around 315 K. However, the bias was relatively linear with Ts used throughout a Landsat sized image (160  160 km2), but
and was therefore generally compensated for during the with adjustment for lapse effects in mountainous terrain to
establishment of the dT function. Allen et al. (2007a) simulate the net cooling of temperature aloft with elevation.
suggested mean values of Rp = 0.91 W m2 sr1 mm1, tNB =
0.866 and Rsky = 1.32 W m2 sr1 mm1 for applications in Soil heat flux (G): Soil heat flux is the rate of heat storage
the western USA (having a semi-arid climate) based on into the soil and vegetation due to conduction. Often, G is

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4015

estimated as a function of net radiation and amount of


vegetation that functions as both a shading mechanism and G ¼ maxð0:4H; 0:15Rn Þ (11)
insulative cover (Choudhury et al., 1987; Bastiaanssen
et al., 1998a; Kustas et al. 1998). A wide range of functions The ‘max’ function selects the larger of the two values.
are in existence for application to a variety of soil types or Equation (11) forces G to be at least as large as 0.15Rn, which is
vegetation conditions or because of varying experiences of a common minimum for moist bare soil and is estimated by
the appliers. General functions for G that are common to Equation (9a) when LAI = 0.0 and the 0.10 coefficient is used,
SEBAL and METRIC applications and that estimate a ratio representing moist soil conditions. The ‘max’ function causes
G/Rn include that by Bastiaanssen (2000) representing the estimate for G to go higher when H is greater than 0.38Rn,
values near midday: representing drier soil conditions. Kustas et al. (1998) and
Anderson et al. (2007) have estimated larger G/Rn ratios for
G  
¼ ðTs  273:15Þð0:0038 þ 0:0074aÞ 1  0:98NDVI 4 bare soil than Equations (8) -11) in applications of the ALEXI
Rn model, where often G/Rn = 0.31.
(8) Equations (8) -11) represent G/Rn for general agricultural
where Ts is the surface temperature (K), a is surface albedo soils that have been tilled within the last few hundred days
and NDVI is the normalized difference vegetation index and do not exhibit a large amount of cracking or delaminated
computed from bands 3 and 4 of Landsat. G is then crust (Allen et al., 2007a). When estimating G for desert
calculated by multiplying G/Rn by Rn. An alternative that soils, near-surface thermal conductivities may be smaller
has also been applied in METRIC was developed by Tasumi than for tilled soils because of cracks, delaminated crust,
(2003) using soil heat flux data collected by Dr J. L. Wright, lack of structure or very low soil-water contents, causing the
USDA-ARS for irrigated crops near Kimberly, Idaho: ratio of G/Rn or G/H to reduce, even when these conditions
may cause the Ts of the soil skin to increase. The cause is
reduced contact of soil particles at depth or very low thermal
G
¼ 0:05 þ 0:18e0:521LAI ðLAI≥0:5Þ (9a) conductance due to very dry soil. In these situations,
Rn standard METRIC applications reduce the estimate for G in
proportion to elevation of Ts above that of dry, bare, tilled
G agricultural soils, on the basis of the assumption that reduced
¼ 1:80ðTs  273:15Þ=Rn þ 0:084 ðLAI < 0:5Þ G of the soil is a partial cause of the elevation of Ts. In
Rn
(9b) situations where the surface is covered by senesced
vegetation such as harvested grain or wetland vegetation,
Later applications of Equation (9a) (Allen, 2010, G can be small, even for low LAI or NDVI conditions,
interoffice memo) have used a coefficient of 0.10 rather because the senesced vegetation functions as an insulator of
than 0.18 to estimate G for a ‘cool’ or ‘moist’ surface the surface.
condition on the basis of data from Wright (USDA-ARS, The heat storage in water estimated at the satellite
Kimberly, ID, pers. comm., 2002) for a silt loam soil and overpass time can have large uncertainties, as G varies with
southern Idaho irrigated row crops. Singh et al. (2009, 2008) water depth, turbidity and time of year. These parameters
developed a function similar to (9a), except using NDVI, on impact the depth of penetration and conversion of solar
the basis of data from Nebraska corn for use in SEBAL: radiation into thermal storage. In some applications of
METRIC to water bodies, time-based functions for G have
G been applied, where G/Rn can be very large, even exceeding
¼ 0:3811e0:232NDVI (10)
Rn 0.5 for clear, deep water bodies (Allen and Tasumi, 2005).
G/Rn will be less than 0.5 for turbid or shallow water bodies
Equation (8) suggests that G/Rn increases with increasing because of the absorption of short-wave radiation nearer the
albedo (indicative of bare soils that often have high reflection) water surface for turbid water and reflection of solar
and decreases with increasing vegetation (due to shading by radiation from and warming by the bottom for shallow
the canopy). Equations (9a) and (10) suggest that G/Rn systems. For applications where the evaporation from open
decreases with increasing leaf area, for the same reason, and water and snow is of particular interest, an aerodynamic
Equation ((9b)) suggests that for bare soil, G increases in procedure as described by Allen (2010) has been used in
proportion to surface temperature. An alternative approach place of the energy balance.
explored by Allen (2010, interoffice memo) for bare or nearly
bare soil conditions follows the suggestion of Brutsaert (1982)
Sensible heat flux: In both METRIC and SEBAL, H
and Stull (1988) to express G as a function of H. This approach
(W m2) is estimated from an aerodynamic function where
allows H to function as a surrogate for Ts, but with wider
applicability, because the Ts used in Equation (9b) will show dT
seasonality that may be independent of G. Values for H are H ¼ rair Cp (12)
rah;1;2
relatively high for dry, bare soil, and lower for wet, cool soil, as
is generally the case for G (more energy is consumed by LE). where rair is air density (kg m3), Cp is specific heat of air at
METRIC applications now use the following method when constant pressure (J kg1 K1) and rah,1,2 is aerodynamic
LAI < 0.5 based on the data by Wright at Kimberly: resistance (s m1) between two near-surface heights, z1 and z2

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4016 R. ALLEN ET AL.

(generally 0.1 and 2 m above the zero-plane displacement


height) computed as a function of estimated aerodynamic
roughness of the particular pixel. In METRIC, the rah,1,2
calculation uses wind speed extrapolated from some blending
height above the ground surface (typically 100 to 200 m) and
an iterative stability correction scheme based on the Monin–
Obukhov functions (Allen et al., 1996). The dT parameter (K)
represents the near-surface temperature difference between z1
and z2. dT is used in Equation (12) because of the difficulty in
estimating surface temperature (Ts) accurately from satellites
due to uncertainty in atmospheric attenuation or contamin-
ation and radiometric calibration of the sensor and due to the
unknown value for air temperature, Ta, above any particular
surface in an image, where Ta can vary by more than 5  C
between cold and dry conditions. Equation (12) is relatively
unique to SEBAL and METRIC and contrasts with classical
approaches where H is estimated using Ts and Ta as

Ts  Ta
H ¼ rair Cp (13)
rah

where rah is aerodynamic resistance (s m1) between the


surface and some height z2. Both Ts and Ta in Equation (13)
can have uncertainties that are greater than their differences,
causing large error in the estimate for H. In addition, the
calculation of rah from the surface requires estimation of the
aerodynamic roughness length for sensible heat transfer, zoh,
which is often uncertain. In contrast, the temperature
gradient dT in Equation (12) is specified to describe a
vertical temperature gradient having endpoints that are both
above the surface, beyond the heights for sensible heat
roughness and zero-plane displacement. The intent is to
estimate dT and H in a blended zone lying above both soil
and vegetation. The blending zone is proposed to be
sufficiently mixed, regarding the underlying blend of
vegetation and soil, so that a single temperature gradient, Figure 2. Interactive, two-layer, five-source model (top); Patch, two-source
dT, may exist or at least can be formulated to describe the model (middle); Interactive, two-source model (bottom) for sensible heat
flux, where soil is partially wetted. The top figure contains overlays of
blend of localized dT within a 30-m satellite pixel, for temperature profiles (not to scale) associated with each resistance point at a
example. The location of the dT function is shown in the top surface and their presumed blending to Ta above the canopy
figure of Figure 2 as the vertical distance between the two
large circular symbols. The z1 and z2 heights have often been
set at 0.1 and 2 m above the zero-plane displacement + zom constant temperature at a blended height well above the
height. This can be varied, however. The idea of elevating surface, for example at 100–200 m, where temperature is
dT above the surface is to eliminate the need to estimate zoh nearly independent of H, and with all instability effects
and to avoid the need to estimate the partitioning of LE incorporated into rah,1,2, Equation (12) suggests that dT, and
between E and T and degree of clumping of vegetation. It is thus Ts, will be largely proportional to H for a fixed
the blended dT that Bastiaanssen et al. (1998a) found to be aerodynamic condition. Therefore, one can anticipate a
linearly related to radiometric surface temperature, Ts. dT is segment of the lower temperature profile, represented by dT,
approximated as a relatively simple linear function of Ts as to be proportional to both H and to Ts. The application of
pioneered by Bastiaanssen (1995): ((14)) appears to extend well across a range of surface
dT ¼ a þ bTs datum (14) roughnesses because as roughness increases and rah
decreases, dT, for the same value of H, decreases because
where a and b are empirically determined constants for a of more efficient transfer of H, and Ts reduces for the same
given satellite image and Ts datum is surface temperature reason. The determination of a and b is described in a
adjusted to a common elevation datum for each image pixel following section. The use of dT indexed to Ts that does not
using a digital elevation model and customized lapse rate. rely on absolute values for Ts substantially reduces error in the
Bastiaanssen (1995) provided empirical evidence for using calculation of H. This approach constitutes one of the
the linear relation between dT and Ts. Allen et al. (2007a) pioneering components of the SEBAL model of Bastiaanssen
suggested that theoretically, assuming some essentially (1995). The use of Ts datum in Equation (14) compensates for

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4017

cooling impacts on Ts because of increasing elevation within can vary by as much as 100 W m2 among the four conditions
an image that are not related to dT and sensible heat flux. because of large differences in Ts of individual components.
The lower part of Figure 2 is a common interactive two-
The blending nature of the dT formulation. The layer model, such as is used in the popular Two-Source
resistance/temperature gradient based model of Equation Energy Balance (TSEB) submodel of ALEXI (Anderson
(13) assumes that a blended air temperature condition and et al., 2008). This model contains similar simplifications
profile exists at some point above a vegetation canopy– through the use of a single bulk soil surface temperature and a
soil surface where the two endpoints of dT can be single bulk canopy temperature. The interactive two-source
established. This assumption permits the application of a model connects H from soil and canopy through two near-
‘single-layer–blended model’ that is used in SEBAL and surface aerodynamic resistances (rh1 and rh2) to a canopy
METRIC. Figure 2 shows resistance diagrams for three level, blended air temperature, Tac, similar to the more
sensible heat flux models for a sparse canopy condition complete five-source model of Figure 2(top). In some
where soil between and beneath the canopy comprises four applications of the two-source model, Rn at the soil surface
conditions: shaded or sunlit and wet (or moist) or dry. These
is estimated as some sort of extinction function based on LAI.
conditions often exist with row crops, forests and riparian
In reality, depending on the architecture of the canopy and
systems, and the four combinations of shaded–dry, shaded–
clumping, significant portions of the soil may be in direct
wet, sunlit–wet and sunlit–dry are often observed even in
nonirrigated conditions, especially a few days after a rain sunlight, for example with row crops and trees. In this case,
event when exposure to direct solar beam and shielding of the Rn at the surface will likely be underestimated. Most
rain by the canopy create the possibility for these conditions. currently applied versions of TSEB use a two-stream radiative
Differences in Ts between the four conditions can be as great transfer model that does account for substrate shading as a
as 20 K. The top diagram of Figure 2 is a more complete and function of solar angle. The net long-wave radiation at the soil
true resistance model for this type of situation, where four surface can be 100 W m2 smaller than that of a canopy,
different endpoints of Ts exist, connected to a canopy level, depending on the soil temperature, which can be 20–30 K
blended air temperature, Tac, by four aerodynamic resis- hotter than canopy when exposed and dry. Depending on how
tances, rhsD, rhsW, rheD and rheW, where subscript s the estimation of G is tied to Rn or assumptions on relative
corresponds to shaded, e to exposed, D to dry and W to wetness or dryness of the soil surface in the model, the G
wet. Tac is presumed to be a blending of within canopy of air estimate will likewise be impacted.
temperature that is influenced from the above canopy Although the two-source models shown in Figure 2 have
blended air temperature, Ta, and the four soil suface more terms than the blended single-layer model of SEBAL
temperature and a bulk vegetation canopy temperature, Tc. and METRIC, they still must simplify the simulation of soil
This may not be a valid assumption under calm wind surface temperature distribution by combining various source
conditions where buoyancy-induced transfer dominates. areas shown in Figure 2. Applications must overcome
The top figure contains overlays of temperature profiles (not uncertainties in describing and modelling the ‘clumping’ of
to scale nor correct shape) associated with each resistance vegetation on the surface, especially when applied to satellite-
point at a surface and their presumed blending to Ta above scale images containing many vegetation types and commu-
the canopy. In Equation (12), the near-surface dT nities and the impact on the partitioning of the bulk
temperature gradient is presumed to ‘float’ within the radiometric temperature, Ts, of the satellite image into Tc
blended zone above the canopy, or slightly into the and Ts bulk soil. The partitioning of Ts is sometimes carried out
canopy, where a single gradient and a single aerodynamic by assigning LE estimated by the net radiation-based
resistance, rah, is assumed to describe the conditions of Priestley–Taylor (1972) equation to vegetation and then
sensible heat transfer. estimating bulk soil surface temperature via iterative
The SEBAL–METRIC blended layer model contrasts estimation of sensible heat fluxes from soil and canopy.
with resistance models commonly used in ‘two-source’ The LE estimated by the Priestley–Taylor equation, however,
models such as the ‘patch’ type of model shown in Figure 2 rarely exceeds the value of Rn, which, as shown in Figure 1,
(middle), where a bulk vegetation canopy temperature is can underestimate LE from vegetation by a factor of 2 under
separate from a bulk soil surface temperature. The within strong advective conditions (Figure 1), and can thereby
canopy temperatures above soil and below canopy top are assign incorrect temperatures to the vegetation and soil
allowed to have separate values in the patch model, where surfaces. In a recent study of multipliers on the Priestley–
different near-surface aerodynamic resistances (rhs and rhc) Taylor equation, Agam et al. (2010) found similar
and different boundary layer resistances (ras and rac) are underestimation when used in a two-source model under
estimated. Care is needed with this type of two-source patch high vapour pressure deficit conditions. This underestimation
model when combining the four possible soil surface could induce substantial error in estimation of buoyancy-
temperatures into a single bulk temperature that matches that induced convective transfer from soil and canopy in the more
viewed by satellite. This precludes estimating that bulk simple two-source models because of the incorrect estima-
temperature using near-surface soil-water content or some tion of surface temperatures as well as substantial error in the
means of energy balance partitioning. It can also preclude partitioning of ET into E and T.
accurate estimation of Rn at the soil surface because of Although they are more simple than many two-source
difficulties in estimating outgoing long-wave radiation that models, the blended temperature gradient models such as

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4018 R. ALLEN ET AL.

SEBAL and METRIC use inverse modelling during All units for z are the same. The value for u200 is assumed
calibration, described later, to reduce biases in the final constant for the satellite image. This assumption is
ET estimate. In these models, the aerodynamic and required for the use of a constant relation between dT and
gradient components are essentially blended to fit similar Ts to be extended across the image (Allen et al., 2007a).
blending presented by the satellite (30 to 120 m Tasumi et al. (2005b) showed the calibration of
resolution) (Rn, G, Ts and zom are all derived from METRIC to be relatively insensitive to the systematic
blended information). METRIC differs from previous error in the value for zom. In METRIC, the momentum
applications of SEBAL principally in how the roughness length (zom) is estimated for each pixel
‘H function’ is calibrated for each specific satellite image, according to land use type or amount of vegetation.
where METRIC uses reference ET as a calibration General land use type is used where the variation in zom as
vehicle, whereas SEBAL uses Rn. a function of vegetation amount is not well established.
General values are given in Brutsaert (1982) and Allen
Aerodynamic transport. The value for rah,1,2 is calcu- et al. (1996). In agricultural areas, zom, and in turn crop
lated between the two heights z1 and z2 in SEBAL and height, is generally proportional to leaf area index (LAI),
METRIC. The value for rah,1,2 is strongly influenced by and is therefore calculated as a function of LAI for ‘short’
buoyancy within the boundary layer driven by the rate of agricultural crops (less than about 1 m in height) using a
sensible heat flux. Because both rah,1,2 and H are relationship from Tasumi (2003):
unknown at each pixel, an iterative solution is required.
During the first iteration, rah,1,2 is computed assuming zom ¼ 0:018 LAI (18)
neutral stability: where zom has units in metres and LAI is dimensionless. A
 
ln zz21 minimum value for zom = 0.005 m represents roughness
rah;1;2 ¼ (15) typical of bare agricultural soils. Equation (18) tends to
u k underestimate zom for corn, which is typically 2 to 3 m in
height (Tasumi, 2003), and a different function may be
where z1 and z2 are heights above the zero-plane
preferred for vines and trees. Because of the difficulties in
displacement of the vegetation where the endpoints of dT
estimating zom from a general function of LAI for all
are defined, u* is friction velocity (m s1) and k is von
vegetation types within an image, Bastiaanssen (2000)
Karman’s constant (0.41). Friction velocity u* is computed
suggested customizing a function for zom based on NDVI
during the first iteration using the logarithmic wind law for
that is developed for each satellite image:
neutral atmospheric conditions:
k u200 zom ¼ exp½ða1 NDVI Þ þ b1  (19a)
u ¼   (16)
ln 200
zom or, as modified by Allen (2002):

where u200 is the wind speed (m s1) at a blending height zom ¼ exp½ða1 NDVI=aÞ þ b1  (19b)
assumed to be 200 m and zom is the momentum roughness
length (m). zom is a measure of the form drag and skin where a1 and b1 are regression constants derived from a
friction for the layer of air that interacts with the surface. u* is plot of ln(zom) versus NDVI or ln(zom) versus NDVI/a for
computed for each pixel inside the process model using a two or more conditions in the image representing specific
specific roughness length for each pixel but with u200 vegetation types. Equations (19a) and (19b) must be fitted
assumed to be constant over all pixels of the image because to local vegetation and conditions. Generally, a is lower
it is defined as occurring at a ‘blending height’ unaffected by for taller vegetation such as trees because of shading.
surface features. Equations (15) and ((16)) support the use of Therefore, the use of albedo helps to distinguish between
a temperature gradient defined between two heights that are tall and short vegetation types that may have similar
both above the surface. This allows one to estimate rah,1,2 NDVI but different albedo. Variations on Equations (19a)
without having to estimate a second aerodynamic roughness and (19b) can be explored by the user to improve the
for sensible heat transfer (zoh) because height z1 is defined to predictive accuracy, including the use of other para-
be at an elevation above zoh. This is an advantage because zoh meters. Current applications of METRIC use a function
can be difficult to estimate for sparse vegetation. by Perrier (1982) to estimate zom for forested land uses
The wind speed at an assumed blending height (200 m) and agricultural orchards (where they can be identified),
above the weather station, u200, is calculated as which provides more accurate estimates for both sparse
  systems and tall vegetation. The Perrier function uses
uw ln z200 estimates of LAI and tree height to estimate impacts of
u200 ¼  
omw
(17) sparseness and height (Kjaersgaard and Allen, 2010;
zx
ln zomw Santos et al., 2011) and estimates similarly to functions
by Raupach (1994) but with fewer data requirements.
where uw is wind speed measured at a weather station at zx The effects of mountainous terrain and elevation on
height above the surface and zomw is the roughness length for wind speed are complicated and difficult to quantify
the weather station surface, similar to Allen and Wright (1997). (Oke, 1987). In METRIC, zom or wind speed for image

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4019
 
pixels in mountains are adjusted using a suite of 1 þ xð2mÞ 2
Ψ h ð2mÞ ¼ 2ln (24a)
algorithms to account for the following impacts: 2
 
• Terrain roughness – the standard deviation of 1 þ xð0:1mÞ 2
elevation within a 1.5-km radius is used to estimate Ψ h ð0:1mÞ ¼ 2ln (24b)
2
an additive to zom to account for vortex and
channelling impacts of turbulence. where
• Elevation effect on velocity – the relative elevation  
within a 1.5 km radius is used to estimate a relative 200 0:25
increase in wind speed based on slope. xð200mÞ ¼ 1  16 (25a)
L
• Reduction of wind speed on leeward slopes – when
the general wind direction aloft can be estimated in a  0:25
2
mountainous terrain, then a reduction factor is made xð2mÞ ¼ 1  16 (25b)
to wind speed on leeward slopes using relative L
elevation and amount of slope as factors.  
0:1 0:25
xð0:1mÞ ¼ 1  16 (25c)
These algorithms have been developed for western L
Oregon and are being tested in Idaho, Nevada and Montana
and are described in an article in preparation (Allen and
Trezza, 2011). Allen and Trezza (2011) also refined the Values for x(200m), x(2m) and x(0.1m) have no meaning when
estimation of diffuse radiation on steep mountainous slopes. L ≥ 0 and their values are set to 1.0.
For L > 0 (stable conditions),
Iterative solution for rah,1,2. During subsequent iterations  
for the solution for H, a corrected value for u* is computed as 2
Ψ mð200mÞ ¼ 5 (26)
L
u k
u ¼   200 (20)  
zom  Ψ m ð200mÞ
ln 200 2
Ψ hð2mÞ ¼ 5 (27a)
L
where cm(200m) is the stability correction for momentum  
transport at 200 m (Equation (23) or (26)). A corrected value 0:1
for rah,1,2 is computed each iteration as Ψ hð0:1mÞ ¼ 5 (27b)
L
 
ln zz21  Ψ h ðz2 Þ þ Ψ h ðz1 Þ
rah;1;2 ¼ (21) When L = 0, the stability values are set to 0. Equation (26)
u  k uses a value of 2 m rather than 200 m for z because it is
where ch(z2) and ch(z1) are the stability corrections for heat assumed that under stable conditions, the height of the stable
transport at z2 and z1 heights (Equations (24a, 24b) or (27a, inertial boundary layer is on the order of only a few metres.
27b)) that are updated each iteration. Using a larger value than 2 m for z can cause numerical
instability in the model. For neutral conditions, L = 0, H = 0
Stability correction functions. The Monin–Obukhov and cm and ch = 0.
length (L) defines the stability conditions of the atmosphere
in the iterative process. L is the height at which forces of
The use of inverse modelling at extreme conditions
buoyancy (or stability) and mechanical mixing are equal and
during calibration
is calculated as a function of heat and momentum fluxes:
In METRIC, the satellite-based energy balance is internally
r cp u 3 Ts calibrated at two extreme conditions (dry and wet) using
L ¼  air (22)
kgH locally available weather data. The auto-calibration is
performed for each image using alfalfa-based reference ET
where g is gravitational acceleration (=9.807 m s2) and
(ETr) computed from hourly weather data. Accuracy and
units for terms cancel to metre for L. Values of the integrated
dependability of the ETr estimate has been established by
stability corrections for momentum and heat transport (cm
lysimetric and other studies in which we have high confidence
and ch) are computed using formulations by Paulson (1970)
(ASCE-EWRI, 2005). The internal calibration of the sensible
and Webb (1970), depending on the sign of L. When L < 0,
heat computation within SEBAL and METRIC and the use of
the lower atmospheric boundary layer is unstable, and when
the indexed temperature gradient eliminate the need for
L > 0, the boundary layer is stable. For L < 0,
atmospheric correction of surface temperature (Ts) and
    reflectance (albedo) measurements using radiative transfer
1 þ xð200mÞ 1 þ xð200mÞ 2
Ψ m ð200mÞ ¼ 2ln þ ln models (Tasumi et al., 2005b). The internal calibration also
2  2 reduces impacts of biases in estimation of aerodynamic
 2ARCTAN xð200mÞ þ 0:5p (23) stability correction and surface roughness.

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4020 R. ALLEN ET AL.

Determination of constants in the dT function. In based on a representative ‘cold’ pixel selected from the
traditional applications of SEBAL (Bastiaanssen et al. satellite scene and is assumed to be 1.05  ETr  l, where l
1998a, 1998b), parameters a and b in Equation (29) are is the latent heat of vaporization (from Equation (35)).
computed by setting dT = 0 for Ts equal to the surface Exceptions to the assumption that ETcold = 1.05 ETr occur
temperature of a local water body (or in its absence, a well during nongrowing season periods and early in growing
vegetated field) and where SEBAL expects H to be zero. seasons where the amount of vegetation for the greenest and
This assumption has been modified in some recent coolest pixels is less than that of the 0.5-m tall standardized
applications of SEBAL by Bastiaanssen (pers. comm. alfalfa reference. In these cases, a ratio of ETcold/ETr = f(NDVI)
2010) by applying an advection correction factor can be developed for the image based on the judgement of the
determined by examining behaviour of reference ET at operator or based on local data. METRIC applications for
nearby weather stations. The dry end of the dT function is agricultural crops in southern Idaho indicate that ETcold/
estimated for a selected ‘hot’ pixel in the image where the ETr ~ 1.25 NDVI to 1.30 NDVI during development periods,
soil is assumed to be dry enough that LE = 0: where NDVI is calculated without atmospheric correction
(Tasumi et al., 2005a).
ðRn  GÞrah hot The near-surface temperature gradient over the cold pixel,
dThot ¼ (28)
rair hot Cp dTcold, is computed using the inverse of Equation (14):

Hcold rah cold


where rah hot is rah computed for the roughness and stability dTcold ¼ (31)
conditions of the hot pixel and rair hot is rair calculated at the rair cold Cp
hot pixel. The values for a and b in Equation (14) are then
estimated from these two pairs of dT and associated Ts. where rah cold is rah computed for the roughness and stability
In METRIC, a similar approach and assumptions are conditions of the cold pixel and rair cold is rair calculated at
made for determining dThot for the hot pixel, although a the cold pixel.
daily surface soil-water balance such as that of the FAO56 The a and b coefficients are determined using the two
publication (Allen et al., 1998) is run for a bare soil pairs of values for dT and Ts, where
condition to confirm that ET = 0 or to supply a nonzero value dThot  dTcold
for ET if there is residual evaporation (Suleiman and Ritchie, b¼ (32)
Ts datum hot  Ts datum cold
2003) from antecedent precipitation or other wetting event.
Thus, in METRIC, and

Hhot ¼ ðRn  GÞhot  LEhot (29) a ¼ dThot  bTs datum hot (33)
where Ts datum hot and Ts datum cold are the surface temperatures
from Equation (1), where LEhot is the latent residual at the hot and cold pixels adjusted to a common elevation
evaporation at the hot pixel and Rn and G are values for datum for each image pixel using a digital elevation model
the hot pixel. and customized lapse rate.
At the ‘cold’ pixel, we define the sensible heat flux in In applications with MODIS, it is often difficult to find
METRIC as 1000-m thermal pixels containing homogenous vegetation
Hcold ¼ ðRn  GÞcold  LEcold (30) that is at sufficient ground cover to represent the ‘cold pixel’
condition (where ET ~ 1.05 ETr). Under these conditions, one
where LEcold is the estimated latent heat flux at the cold can use METRIC applications based on and calibrated using
pixel. Experience shows that the coldest (and wettest) Landsat to determine ET and ETrF for specific MODIS-scale
agricultural fields in a satellite image that are at full cover pixels and to develop associated ETrF versus NDVI
(LAI > 4) have ET rates that are typically about 5% greater relationships for a target region or some fraction of the
than the rate from the alfalfa reference crop ETr (Tasumi, population of pixels having maximum NDVI that can in turn
2003; Tasumi et al., 2005a). This occurs because, for a large be used to determine calibration for the coarse MODIS-based
population of fields, some fields will have a wet soil surface images (where pixels having ETrF = 1.05 do not exist).
beneath a full vegetation canopy that will tend to increase the
total ET rate to about 5% above that of the ETr standard Calibration via reference evapotranspiration
(Wright, 1982). In addition, when viewing a large The METRIC model uses the standardized ASCE
population of fields containing full-cover alfalfa, a specific Penman–Monteith equation for the alfalfa reference ETr
subpopulation of fields will have somewhat wetter condi- (ASCE-EWRI, 2005) to calibrate the energy balance
tions because of recent irrigation and therefore slightly functions. ETr is typically 20 to 30% greater than grass
higher ET and slightly cooler temperature than the ‘mean’ reference ET (ETo). ETr is used to approximate the ET of
full-cover condition represented by the standardized alfalfa the cold pixel calibration condition because METRIC can
reference. Also, the ETr definition represents the ‘mean’ be applied without crop classification so that the specific
evaporative condition for a 0.5 m tall alfalfa. Some of the crop type at any pixel is generally not known. Because the
alfalfa population (or other crops at full ground cover) will crop coefficient, Kc, typically peaks at 1.0 for many crops
have ET rates above the mean. Hence, the value for LEcold is when the alfalfa reference basis is used (Wright, 1982;

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4021

Jensenetal.,1990),thecoldpixelcanbenearlyanyagricultural equations include those for leaf area index, aerodynamic


crop having complete ground cover and LAI greater than roughness length, soil heat flux and air temperature gradient.
approximately 4.0. The freedom from requiring crop classi- Users may be concerned whether these component estimates
fication constitutes a significant cost savings, as accurate crop or the METRIC calibration process may break down under
classification from satellite can be relatively costly. specific circumstances. The empirical equations applied are,
Accuracy and dependability of the standardized ASCE in principle, considered to be relatively robust in their
Penman–Monteith equation for calibration of the energy transferability. The use of ETr to calibrate each application
balance under the well-watered condition has been well corrects for much of the individual bias introduced into a
documented. The equation has exhibited good sensitivity to component estimate. For example, Tasumi (2003) found
hourly changes in weather (Jensen et al., 1990; Allen et al., less than a few % error in METRIC ET when estimates for
2005; ASCE-EWRI, 2005), where the equation provides a zom were systematically halved and doubled prior to
robust approximation of the ET expected from extensive METRIC calibration. The calibration process adjusted the
surfaces of full vegetation cover. Each satellite image is calibration of the dT function to account for the decreased or
auto-calibrated during the METRIC process using ETr increased bias in roughness, and therefore estimates for H
(calculated using hourly or shorter period weather data), and ET were little impacted. In addition, the energy balance
where LEcold = 1.05 ETr  l. The Hcold from Equation (30) process is only slightly sensitive to roughness for areas
is in turn used in Equation (31) to derive coefficients from having ET within 30% of ETr because the total value for H is
Equations (32) and (33) to derive a general function to small. The process is only slightly sensitive to roughness at
estimate dT for all pixels of the image. low values for ET because boundary layer buoyancy
The calibration of the sensible heat process equations, and dominates the sensible heat exchange.
in essence the entire energy balance, to ETr corrects the Several applications of METRIC have required modifica-
surface energy balance for lingering systematic computation- tion of the empirical functions for G. In an application to
al biases associated with empirical functions used to estimate Imperial Valley, California, soil heat transfer into desert soils
some components and uncertainties in other estimates as was reduced by crusts detached from the underlying soil. In
summarized by Allen et al. (2005), including this situation, desert soils had significantly higher surface
temperature than bare agricultural soils because of the reduced
• atmospheric correction G. Values for G estimated empirically were reduced in
• albedo calculation proportion to surface temperature above a threshold value.
• net radiation calculation In applications in southern Idaho, sparsely vegetated
• surface temperature from the satellite thermal band desert systems, although aerodynamically rough, have
• air temperature gradient function used in sensible heat characteristically high surface temperature. The combin-
flux calculation ation of high zom and high Ts causes overestimation of H
• aerodynamic resistance including stability functions when the same dT versus Ts function developed for
• soil heat flux function agricultural regions of the satellite image is used. The
• wind speed field overestimation of H has been largely overcome by including
an ‘excess resistance’ term to rah to account for the sheltering
This list of biases plagues essentially all surface energy effect of desert canopy. The function for the excess
balance computations that utilize satellite imagery as the resistance was empirically developed using eddy covariance
primary spatial information resource. Most polar orbiting and temperature gradient measurements made over the
satellites orbit about 700 km above the earth’s surface, yet the sagebrush environment and was expressed as a closed
transport of vapour and sensible heat from land surfaces is function of wind speed.
strongly impacted by aerodynamic processes including wind In applications along the Middle Rio Grande of New
speed, turbulence and buoyancy, all of which are essentially Mexico, two separate dT functions were developed and
invisible to satellites. In addition, precise quantification of applied for each image. One dT function was applied to
albedo, net radiation and soil heat flux is uncertain and agricultural areas within the Rio Grande valley, which tends
potentially biassed. Therefore, even though best efforts are to be incised below the regional plain, and the other dT
made to estimate each of these parameters as accurately and function was applied to areas outside the valley. The two
as unbiased as possible, some biases do occur, and calibration functions were required because of substantially different
to ETr helps to compensate for this by introducing a bias surface temperatures exhibited for the ‘dry pixels’. In some
correction into the calculation of H. The end result is that applications containing desert systems, the slope of the dT
biases inherent to Rn, G and subcomponents of H are function in Equation (29) is reduced when Ts datum exceeds
essentially cancelled by the subtraction of a bias-cancelling the Ts datum hot of the hot pixel. These deviations from routine
estimate for H. The result is an ET map having values ranging application of METRIC equations are examples of
between near zero and near ETr for images having a range of modifications to the empirical method that may be required
bare or nearly bare soil and full vegetation cover. to improve accuracy of estimates of ET from natural systems
and illustrate the value of specific knowledge of native
Adjustment to empirical equations: The METRIC model systems, principles of aerodynamic transfer and boundary
utilizes a number of empirical equations to estimate values layer development, soil physics and energy transfer limi-
to support various components of the energy balance. These tations. Future study and additional sensitivity assessments

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4022 R. ALLEN ET AL.

of impacts of empirical estimates for various energy balance relative ET between the two approaches averaged about 0.2,
components are warranted. indicating that about 20% of values exceed 20% difference.
Figure 3 also shows the high value of Landsat images and its
Calculation of evapotranspiration relatively high resolution thermal sensor. This permits
retrieval of ET for individual agricultural fields and even
ET at the instant of the satellite image is calculated for each some indication of within-field variation.
pixel by dividing LE from Equation (1) by latent heat of
vaporization: 24-h evapotranspiration (ET24). Daily values of ET
LE (ET24) are generally much more useful than the instantaneous
ETinst ¼ 3600 (34) ET that is derived from the satellite image. In the METRIC
lrw
process, ET24 is estimated by assuming that the instantaneous
where ETinst is instantaneous ET (mm h1), 3600 converts ETrF computed at image time is the same as the average ETrF
from seconds to hours, rw is the density of water (~1000 kg over the 24-h average. The consistency of ETrF over a day has
m3) and l is the latent heat of vaporization (J kg1) been demonstrated by various studies, including those of
representing the heat absorbed when a kilogramme of water Romero (2004), Colaizzi et al. (2006) and Allen et al.
evaporates and is computed as (2007a).
The assumption of constant ETrF during a day has been
l ¼ ½2:501  0:00236ðTs  273:15Þ  106 (35) shown to be valid for agricultural crops that have been
developed to maximize photosynthesis and thus stomatal
conductance. The advantage of using ETrF is to account for
The reference ET fraction (ETrF) is calculated as the the increase in 24-h ET that can occur under advective
ratio of the computed instantaneous ET (ETinst) from each conditions. The impacts of advection are represented well by
pixel to the reference ET (ETr) computed from weather the Penman–Monteith equation. However, in nonirrigated or
data: water short situations, ETrF may decrease during afternoon,
ETinst especially for some native vegetation where plants endeavour
ETr F ¼ (36) to conserve soil water through stomatal control (Glenn et al.,
ETr
2011). In addition, by definition, when the vegetation under
where ETr is the estimated instantaneous rate (interpolated study is the same as or similar to the vegetation for the
from hourly data) (mm h1) for the standardized 0.5-m tall surrounding region and experiences similar water inputs
alfalfa reference at the time of the image. Generally, only one (natural rainfall only), then (by definition) no advection can
or two weather stations are required to estimate ETr for a occur. This is because as much sensible heat energy is
Landsat image that measures 180  180 km, as discussed generated by the surface under study as is generated by the
later. ETrF is the same as the well-known crop coefficient, Kc, region. Therefore, the net advection of energy is nearly zero.
when used with an alfalfa reference basis and is used to Therefore, under these conditions, the estimation by ETr that
extrapolate ET from the image time to 24-h or longer periods. accounts for impacts of advection to a wet surface do not
One should generally expect ETrF values to range from 0 occur, and the use of ETrF to estimate 24-h ET may not be
to about 1.0 (Wright, 1982; Jensen et al., 1990). At a valid. Instead, the use of evaporative fraction, EF, that is used
completely dry pixel, ET = 0 and therefore ETrF = 0. A pixel with SEBAL applications may be a better time-transfer
in a well-established field of alfalfa or corn can occasionally approach for rainfed systems. Various schemes of using EF
have an ET slightly greater than ETr and therefore ETrF > 1, for rainfed portions of Landsat images and ETrF for irrigated,
perhaps up to 1.1 if it has been recently wetted by irrigation or riparian or wetland portions were explored by Kjaersgaard
precipitation. However, ETr generally represents an upper and Allen (2010). When used, the EF is calculated as
bound on ET for large expanses of well-watered vegetation.
ETinst
Negative values for ETrF can occur in METRIC because of EF ¼ (37)
systematic errors caused by various assumptions made earlier Rn  G
in the energy balance process and due to random error where ETinst and Rn and G have the same units and represent
components so that error should oscillate about ETrF = 0 for the same period.
completely dry pixels. In calculation of ETrF in Equation Finally, the ET24 (mm day1) is computed for each
(36), each pixel retains a unique value for ETinst that is derived image pixel in SEBAL as
from a common value for ETr derived from the representative
weather station data. ET24 ¼ ðEF ÞðRn24 Þ (38)
Figure 3 compares ETrF developed using METRIC with
ETrF derived from NDVI as ETrF = 1.25 NDVI. Similarity and in METRIC as
in estimates is apparent, indicating the strong relationship ET24 ¼ Crad ðETr F ÞðETr24 Þ (39)
between relative ET and vegetation amount. What is
different between the two estimates is that the energy where ETrF (or EF) is assumed equal to the ETrF (or EF)
balance was able to detect evaporation from fields having determined at the satellite overpass time, ETr24 is the
low vegetation cover, which stemmed from post harvest cumulative 24-h ETr for the day of the image and Crad is a
irrigation and tillage. In addition, the standard deviation of correction term used in sloping terrain to correct for variation

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4023

Figure 3. Comparison of the ratio of actual ET to reference ET (ETrF) on 22 July 2006 near Twin Falls, Idaho, using full energy balance (METRIC) (top)
and using a simple ETrF = 1.25 NDVI relationship (bottom)

in 24-h versus instantaneous energy availability. Crad is image where Ts exhibits a similar range. Investigations in
calculated for each image and pixel as Idaho (Trezza, 2002; Tasumi, 2003) have indicated that
where wind speed is higher and thus ET higher in a
RsoðinstÞHorizontal Rsoð24ÞPixel
Crad ¼  (40) subregion relative to the region of dT calibration, that
RsoðinstÞPixel Rsoð24ÞHorizontal values for Ts tend to run slightly lower. The lower Ts causes
lower dT estimates that in turn produce higher ET.
where Rso is clear-sky solar radiation (W m1), the ‘(inst)’ Therefore, even though wind speed and ETr for the
subscript denotes conditions at the satellite image time, ‘(24)’ subregion deviate from that for the image subregion of
represents the 24-h total, the ‘Pixel’ subscript denotes slope calibration, the ET is more accurately predicted using a
and aspect conditions at a specific pixel and the ‘Horizontal’ single, general dT function (Equation (14)) established for
subscript denotes values calculated for a horizontal surface the image date. The coefficients in Equation (14) are highly
representing the conditions impacting ETr at the weather correlated to wind speed. Therefore, if wind speed over an
station. For applications to horizontal areas, Crad = 1.0. image is varied, then Equation (14) must be recalibrated.
The 24-h Rso for horizontal surfaces and for sloping After ET and ETrF have been determined using the
pixels is calculated as energy balance, and the application of the single dT
Z 24 function, then, when interpolating between satellite
Rso ð24Þ ¼ Rsoi (41) images, a full grid for ETr is used for the extrapolation
0 over time to account for both spatial and temporal variation
where Rso_i is instantaneous clear-sky solar radiation at in ETr. The ETr grid is generally made on a 3- or 5-km base
time i of the day, calculated by an equation that accounts using as many quality-controlled weather stations located
for effects of slope and aspect. In METRIC, ETr24 is within and in the vicinity of the study area as available.
calculated by summing hourly ETr values over the day of Depending on data availability and the density of the
the image. weather stations, various gridding methods including
There is some question regarding whether the dT kriging, inverse distance, and splining can be used.
function responds correctly across regions of a satellite
image where wind and thus ETr have different values. In Seasonal evapotranspiration (ETseasonal). Monthly and
application of METRIC, a single dT function is generally seasonal evapotranspiration ‘maps’ are often desired for
developed for an entire image or for a subregion of an quantifying total water consumption from agriculture.

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4024 R. ALLEN ET AL.

These maps can be derived from a series of ETrF images acquisition. In the case of Landsat, the return interval is
by interpolating ETrF on a pixel by pixel basis between 16 days. As a result, monthly ET estimates are based on only
processed images and multiplying, on a daily basis, by the one or two satellite image snapshots per month. In the case of
ETr for each day. The interpolation of ETrF between clouds, intervals of 48 days between images can occur. This
image dates is not unlike the construction of a seasonal Kc can be rectified by combining multiple Landsats (five with
curve (Allen et al., 1998), where interpolation is done seven) or by using data fusion techniques, where a more
between discrete values for Kc. frequent but more coarse system such as MODIS is used as a
The METRIC approach assumes that the ET for the entire carrier of information during periods without quality Landsat
area of interest changes in proportion to change in gridded images (Gao et al., 2006; Anderson et al., 2010).
ETr. This is a generally valid assumption and is similar to the Trezza and Allen (2009) developed a type of data fusion
assumptions used in the conventional application of Kc  ETr. method that used MODIS as a transfer mechanism in between
This approach is effective in estimating ET for both clear and Landsat images that were more than 32 days apart, and
cloudy days in between the clear-sky satellite image dates. therefore, interpolation of ETrF was considered to be too
Tasumi et al. (2005a) showed that the ETrF was consistent speculative. In that application, they developed ratios of ETrF
between clear and cloudy days using lysimeter measurements from Landsat (30 m) to the underlying NDVI from MODIS
at Kimberly, Idaho. Because ETr is used only as an index of (250 m) on the same or similar days as the Landsat images.
the relative change in weather, specific information at each These ratios were then interpolated between Landsat images,
pixel is retained through the ETrF. and for every available MODIS date, ETrF was estimated by
Cumulative ET for any period, for example, month, season multiplying MODIS NDVI by the interpolated ratios. The
or year is calculated as result was a time integrated ETrF that followed the evolution
of NDVI as monitored by MODIS but retained much of the
X
n
30-m resolution of Landsat. This technique, however, was
ETperiod ¼ ½ðETr Fi ÞðETr24 i Þ (42) unable to account for any evaporation from precipitation
i¼m
occurring between Landsat image dates and effectively only
preserved the amount of soil evaporation that was captured in
where ETperiod is the cumulative ET for a period beginning on the endpoint Landsat images.
day m and ending on day n, ETrFi is the interpolated ETrF for A potential shortfall in basing integrated ET averages on
day i and ETr24i is the 24-h ETr for day i. Units for ETperiod will periodic snapshots from satellites is in sampling a represen-
be in mm when ETr24 is in mm day1. The interpolation tative or ‘average’ impact of precipitation on ET. Precipitation
between values for ETrF is best made using a curvilinear events immediately antecedent to a satellite image may
interpolation function, for example a spline function, to better unduly dominate the ET image, or conversely, if no
fit the typical curvilinearity of crop coefficients during a precipitation event has occurred for 5 to 10 days prior to a
growing season (Wright, 1982). Generally, one satellite Landsat image, then effects of evaporation from precipitation
image per month is sufficient to construct an accurate ETrF may be absent from the image. Therefore, the image-based
curve for purposes of estimating seasonal ET (Allen et al., product may not represent evaporation from rainfall averaged
2007a). During periods of rapid vegetation change, a more over the monthly period. It is usually important, hydrologic-
frequent image interval may be desirable. Examples of ally, that the average, or integrated, effect of evaporation from
splining ETrF to estimate daily and monthly ET are given by precipitation be represented by the ET image so that the
Allen et al. (2007a) and Singh et al. (2008). product is more useful in hydrologic and water balance
If a specific pixel must be masked out of an image because studies. Methods for accounting for precipitation events
of cloud cover, then a subsequent image date must be used occurring between satellite images are described in a
during the interpolation and the estimated ETrF or Kc curve companion paper by Kjaersgaard et al. (2011) including a
will have reduced accuracy. soil-water balance procedure that adjusts the ET derived from
the satellite overpass date for background evaporation from
Average ETrF over a period. An average ETrF for the soil caused by rainfall over monthly or longer integration
period can be calculated as periods. The result of the adjustment is an ET image and
consequently final ET map that better represents the average
P
n
evaporative conditions over the period. Figure 4 shows a map
½ðETr Fi ÞðETr24i Þ
ETr Fperiod ¼ i¼m (43) of ETrF for the Central Platte Natural Resources District,
P
n
Nebraska, where application of the procedure to adjust for
ETr24i
i¼m background evaporation assisted in ETrF from adjacent
images (in time) to fill in clouded areas relatively seamlessly.
Figure 5 shows growing season (April–October)
Moderately high resolution satellites such as Landsat evapotranspiration during 2006 over portions of the
provide the opportunity to view evapotranspiration on a field Eastern Snake River plain and surrounding mountains
by field basis, which can be valuable for water rights of SE Idaho as produced by METRIC energy balance and
management, irrigation scheduling and discrimination of ET splining between Landsat image dates using gridded
among crop types (Allen et al., 2007b). The downside of reference evapotranspiration. Irrigated areas show sea-
using high resolution imagery is less frequent image sonal ET ranging from 500 to 1200 mm and nonirrigated

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4025

Figure 4. ETrF product for 20 August 2007 over the Central Platte Natural Resources District, Nebraska, with clouded areas masked (top) and filled
(bottom) using a procedure that adjusted for background evaporation from antecedent precipitation events

areas ranging from 100 to 400 mm in lower lying areas model Surface Energy Balance Algorithms for Land
and as high as 1000 mm in mountains. (SEBAL), an important energy balance model developed
in the Netherlands and applied worldwide by Bastiaanssen
and his associates. In particular, METRIC uses the SEBAL
technique for estimating the near-surface temperature
SUMMARY AND CONCLUSIONS
gradient, dT, as an indexed function of radiometric surface
Mapping EvapoTranspiration at high Resolution with temperature, thereby eliminating the need for absolutely
Internalized Calibration (METRIC™) is a satellite-based accurate aerodynamic surface temperature and the need for
image-processing tool for calculating ET as a residual of the air temperature measurements for estimating sensible heat
energy balance at the earth’s surface. METRIC has, as its flux at the surface. METRIC utilizes internal calibration of
foundation, principles and techniques used by the pioneering the satellite-based energy balance at two extreme conditions

Figure 5. Growing season (April–October) evapotranspiration during 2006 over portions of the Eastern Snake River plain and surrounding mountains as
produced by METRIC energy balance and splining between Landsat image dates using gridded reference evapotranspiration

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
4026 R. ALLEN ET AL.

(dry and wet) using locally available weather data. The auto- Priestley–Taylor approach in a two-source surface energy balance
model. Journal of Hydrometeorology 11(1): 185–198.
calibration is performed for each image using alfalfa-based Allen RG. 1996. Assessing integrity of weather data for use in reference
reference ET (ETr) computed from hourly weather data. The evapotranspiration estimation. Journal of Irrigation and Drainage
internal calibration and the use of the indexed temperature Engineering, ASCE 122(2): 97–106.
Allen RG. 2002. Evapotranspiration for southwest Florida from satellite-based
gradient eliminate the need for atmospheric correction of Ts energy balance. Consulting report to Tampa Bay Water, Tampa Bay, FL; 45.
and reflectance (albedo) measurements using radiative Allen RG. 2010. Aerodynamic Procedure for Estimating Evaporation
transfer models. The internal calibration also reduces from Open Water and Snow in METRIC. Memo, Kimberly R&E
Center: University of Idaho; 10.
impacts of biases in estimation of aerodynamic stability Allen, RG, Tasumi, M. 2005. Evaporation from American Falls reservoir
correction and surface roughness. in Idaho via a combination of Bowen ratio and eddy covariance.
The SEBAL and METRIC models represent a maturing Proceedings of the EWRI World Water and Environmental Resources
Congress 2005: Impacts of Global Climate Change. EWRI, 1801
technology for deriving a satellite-driven surface energy Alexander Bell Drive, Reston, VA; 17.
balance for estimating ET from the earth’s surface and have Allen RG. Wright JL. 1997. translating wind measurements from weather
the potential to become widely adopted by water resources stations to agricultural crops. Journal of Hydrologic Engineering, ASCE
2(1): 26–35.
communities. METRIC is designed to produce high quality Allen RG, Trezza R. 2011. Sensitivity of aerodynamic characteristics in
and accurate maps of ET for focused regions smaller than a mountainous terrain when estimating evapotranspiration from satellite-
few hundred kilometres in scale and at high resolution. based energy balance. Interoffice memo and paper in preparation.
University of Idaho; 12.
Thus, use of reference ET for calibration and for Allen RG, Pruitt WO, Businger JA, Fritschen LJ, Jensen ME, Quinn FH.
extrapolating from instantaneous to 24-h ET helps to 1996. Evaporation and transpiration, Chap. 4, 125–252. In ASCE
account for regional advection effects on ET and improves Handbook of Hydrology, Wootton et al. (ed.). ASCE: Reston, VA.
Allen RG, Pereira LS, Raes D. Smith M. 1998. Crop evapotranspiration:
congruency with the traditional reference ET  Kc approach. guidelines for computing crop water requirements. United Nations FAO,
Requirements for applying METRIC are trained experts Irrigation and Drainage Paper 56. 1998. Rome, Italy; 300. http://www.fao.
having good background in energy balance and radiation org/docrep/X0490E/X0490E00.htm (accessed 5 February 2007)
Allen RG, Hartogensis O, de Bruin H. 2000. Long-wave radiation over
physics and adequate knowledge of vegetation characteristics alfalfa during the RAPID field campaign in southern Idaho. unpublished
as well as high quality hourly (or shorter) weather data. The Univ. Idaho Research Report, Kimberly, ID; 20.
requirement for trained experts having good physics Allen RG, Tasumi M, Trezza R. 2005. Benefits from tying satellite-based
energy balance to ground-based reference evapotranspiration. In Earth
background and high quality weather data is a distinct Observation For Vegetation Monitoring and Water Management. AIP
disadvantage of METRIC, and in some regard of SEBAL, in Conference Proceedings Vol. 852. ISBN: 0-7354-0346-5, AIP, College
that highly specialized personnel are required to operate the Park, MD; 127–137.
Allen RG, Tasumi M, Trezza R. 2007a. Satellite-based energy balance for
model with credibility. It is evident that METRIC is highly mapping evapotranspiration with internalized calibration (METRIC) – model.
reliant on the operator’s ability to select appropriate cold and Journal of Irrigation and Drainage Engineering, ASCE 133: 380–394.
hot pixels. Current development work has tested the use of a Allen RG, Tasumi M, Morse A, Trezza R, Wright JL, Bastiaanssen W, Kramber
W, Lorite I, Robison C. 2007b. Satellite-based energy balance for mapping
statistical procedure to automate the selection of the hot and evapotranspiration with internalized calibration (METRIC) – applications.
cold pixels with relatively good success (Burnett et al., 2008, Journal of Irrigation and Drainage Engineering, ASCE 133: 395–406.
Kjaersgaard et al., 2009). However, METRIC is currently not Anderson MC, Norman JM, Mecikalski JR, Torn RD, Kustas WP, Basara JB.
2004. A multi-scale remote sensing model for disaggregating regional fluxes
a complete hands-off model. Continued comparisons with to micrometeorological scales. Journal of Hydrometeorology 5: 343–363.
other remote sensing-based approaches are needed to Anderson MC, Norman JM, Mecikalski JR, Otkin JA, Kustas WP. 2007.
understand the advantages/disadvantages of the different A climatological study of evapotranspiration and moisture stress across
the continental United States based on thermal remote sensing: 1 Model
techniques and to gain a better understanding of the formulation. Journal of Geophysical Research 112: D10117–D10134.
uncertainties in various model components and their impacts Anderson MC, Kustas WP, Dulaney W, Feng G, Summer D. 2010.
on final ET estimates. Integration of multi-scale thermal satellite imagery for evaluation of
daily evapotranspiration at the sub-field scale. Remote Sensing and
Hydrology Symposium 2010. Jackson Hole, Wyoming, USA; 62.
ASCE-EWRI. 2005. The ASCE standardized reference evapotranspiration
ACKNOWLEDGEMENTS
equation. ASCE-EWRI Standardization of Reference Evapotranspiration
The authors acknowledge the long-term lysimeter data Task Comm. Report, ASCE Bookstore, ISBN: 078440805, Stock Number
40805; 216.
collection by Dr James L. Wright, USDA-ARS, Kimberly, Bastiaanssen WGM. 1995. Regionalization of surface flux densities and
Idaho, that provided the means for comparing ET measure- moisture indicators in composite terrain: a remote sensing approach under
ments with ET estimated by METRIC and for confirming the clear skies in Mediterranean climates. PhD Diss., CIP Data Koninklijke
Bibliotheek, Den Haag, the Netherlands; 273.
relative constancy of ETrF during daylight hours. Funding for Bastiaanssen WGM. 1998. Remote sensing in water resources management: the
the various applications and development came from NASA, state of the art. International Water Management Institute: Colombo, Sri
Raytheon, USDA-CSREES, USGS, USDA, Idaho Depart- Lanka.
Bastiaanssen WGM. 2000. SEBAL-based sensible and latent heat fluxes
ment of Water Resources, Idaho Agricultural Experiment
in the irrigated Gediz Basin Turkey. Journal of Hydrology 229: 87–100.
Station, Idaho Engineering Experiment Station, National Bastiaanssen WGM, Menenti M, Feddes RA, Holtslag AAM. 1998a. A
Science Foundation Idaho EPSCoR and University of remote sensing surface energy balance algorithm for land (SEBAL): 1.
Nebraska Experiment Station. Formulation. Journal of Hydrology 212–213: 198–212.
Bastiaanssen WGM, Pelgrum H, Wang J, Ma Y, Moreno J, Roerink GJ,
van der Wal T. 1998b. The surface energy balance algorithm for land
(SEBAL): Part 2. Validation. Journal of Hydrology 212–213: 213–229.
REFERENCES Bastiaanssen WGM, Noordman EJM, Pelgrum H, Davids G, Thoreson
BP, Allen RG. 2005. SEBAL model with remotely sensed data to
Agam N, Kustas WP, Anderson MC, Norman J, Colaizzi PD, Howell improve water-resources management under actual field conditions
TA, Prueger JH, Meyers TP, Wilson TB. 2010. Application of the ASCE. Journal of Irrigation and Drainage Engineering 131(1): 85–93.

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)
SATELLITE-BASED ET ESTIMATION WITH SEBAL AND METRIC 4027

Berk A, Anderson GP, Bernstein LS, Acharya PK, Dothe H, Matthew MW, Oke TR. 1987. Boundary Layer Climates, 2nd edn. Methuen: London;
Adler-Golden SM, Chetwynd JH, Jr., Richtsmeier SC, Pukall B, Allred CL, 435, ISBN: 0-415-04319-0.
Jeong LS, Hoke ML. 1999. MODTRAN4 radiative transfer modeling for Overgaard J, Rosbjerg D, Butts MB. 2006. Land-surface modeling in
atmospheric correction. Proceedings of SPIE Optical Spectroscopic hydrological perspective – a review. Biogeosciences 3: 229–241.
Techniques and Instrumentation for Atmospheric and Space Research III, Paulson CA. 1970. The mathematical representation of wind speed and
3756, 1999. Bellingham WA 98227–0010 USA temperature profiles in the unstable atmospheric surface layer. Applied
Betts AK, Chen F, Mitchell KE, Janjic ZI. 1997. Assessment of the land Meteorology 9: 857–861.
surface and boundary layer models in two operational versions of the NCEP Perrier A. 1982. Land surface processes: vegetation. In land surface
Eta model using FIFE data. Monthly Weather Review 125: 2896–2916. processes in atmospheric general circulation models, Eagelson P (ed.).
Burnett B, Allen RG, Robison CW, Tasumi T, Lorite I. 2008. Estimating Cambridge University Press, Cambridge. 395–448.
the soil surface evaporation and transpiration components from satellite Priestley CHB, Taylor RJ. 1972. On the assessment of surface heat flux
images in the absence of a thermal band. Paper presented at the 2008 and evaporation using large scale parameters. Mon. Weath. Rev. 100:
World and Environmental Resources Congress of ASCE and EWRI, 81–92.
Honolulu, HI, May 12-16, 2008. Published on CD-ROM, ASCE. 18. Raupach MR. 1994. Simplified expressions for vegetation roughness length
Brutsaert W. 1982. Evaporation into the atmosphere. D. Reidel Publishing and zero plane displacement height as functions of canopy height and area
Co: Dordrecht, Holland; 300. index (Research note). Boundary-Layer Meteorology 71: 211–216.
Choudhury BJ, Idso SB, Reginato RJ. 1987. Analysis of an empirical Romero MG. 2004. Daily evapotranspiration estimation by means of
model for soil heat flux under a growing wheat crop for estimating evaporative fraction and reference ET fraction. PhD Diss., Utah State
evaporation by an infrared-temperature-based energy balance equation. Univ., Logan, Utah.
Agricultural and Forest Meteorology 39: 283–297. Santos C, Lorite IJ, Tasumi M, Allen RG. 2011. ET estimation for olive
Colaizzi PD, Evett SR, Howell TA, Tolk JA. 2006. Comparison of five orchards using satellite-based energy balance. Irrigation Science Accepted.
models to scale daily evapotranspiration from one-time-of-day measure- Singh RK, Irmak A, Irmak S, Martin DL. 2008. Application of SEBAL
ments. Transactions of the ASABE 49(5): 1409–1417. model for mapping evapotranspiration and estimating surface energy
Duffie JA, Beckman WA. 1991. Solar engineering of thermal process, fluxes in south-central Nebraska. Journal of Irrigation and Drainage
2nd edn. John Wiley and sons: NY. Engineering 134(3): 273–285.
Gao F, Masek J, Schwaller M, Hall F. 2006. On the blending of the Singh RK, Irmak A, Walter-Shea, EA, Verma SB. 2009. Performance
Landsat and MODIS surface reflectance: predicting daily landsat evaluation of soil heat flux models in Great Plains. World Environmental
surface reflectance. IEEE Transactions on Geoscience and Remote and Water Resources Congress 2009: Kansas City, MO; 10.
Sensing 44: 2207–2218. Stull RB. 1988. An introduction to boundary layer meteorology, Kluwer
Garrison JD, Adler GP. 1990. Estimation of precipitable water over the Academic Publishers, Dordrecht, Boston, London. 666. ISBN 90 277
United States for application to the division of solar radiation into its 2768 6; 90 277 2769 4.
direct and diffuse components. Solar Energy 44(4): 225–241. Suleiman AA. Ritchie JT. 2003. Modeling soil water redistribution during
Glenn E, Neale C, Hunsaker D. 2011. Vegetation-index-based crop second-stage evaporation. SSSAJ 67: 377–386.
coefficients to estimate evapotranspiration by remote sensing in agricul- Tasumi M. 2003. Progress in operational estimation of regional
tural and natural ecosystems. Hydrological Processes 25(26): 4050–4062. evapotranspiration using satellite imagery. PhD Dissertation, University
DOI: 10.1002/hyp8392. of Idaho, Moscow, ID; 357.
Granger RJ. 1989. Evaporation from natural nonsaturated surfaces. Tasumi M. 2005. A review of evaporation research on Japanese lakes.
Journal of Hydrology 111: 21–29. Proceedings of the ASCE/EWRI World Water and Environmental
Granger RJ. 2000. Satellite-derived estimates of evapotranspiration in the Resources Congress, May 15–19, 2005. 10 pages. ASCE, Reston, VA.
Gediz Basin. Journal of Hydrology 229(1–2): 70–76. Tasumi M, Allen RG, Trezza R, Wright JL. 2005a. Satellite-based energy
Hulley G, Hook SJ. 2009. The North American ASTER Land Surface balance to assess within-population variance of crop coefficient curves.
Emissivity Database (NAALSED) Version 2.0. Remote Sensing of ASCE Journal of Irrigation and Drainage Engineering 131(1): 94–109.
Environment 113(9): 1967–1975. Tasumi M, Trezza R, Allen RG, Wright JL. 2005b. Operational aspects of
Jensen ME, Burman RD, Allen RG (eds.). 1990. Evapotranspiration and satellite-based energy balance models for irrigated crops in the semi-
irrigation water requirements. ASCE Manuals and Reports on Engineering arid U.S. Irrigation and Drainage Engineering 19: 355–376.
Practice No. 70. ISBN: 0-87262-763-2, 332 p. ASCE, Reston, VA. Tasumi M, Allen RG, Trezza R. 2006. Estimation of at-surface reflectance
Kjaersgaard J, Allen RG. 2010. Remote sensing technology to produce and albedo from satellite for routine, operational calculation of land
consumptive water use maps for the Nebraska Panhandle. Final surface energy balance. ASCE Journal of Hydrologic Engineering 22:
completion report submitted to the University of Nebraska; 88. 197–208.
Kjaersgaard J, Allen RG, Garcia M, Kramber W, Trezza R. 2009. Automatic Tasumi M, Allen RG, Trezza R. 2008. At-surface albedo from Landsat
selection of anchor pixels for Landsat-based evapotranspiration estimation. and MODIS satellites for use in energy balance studies of evapotrans-
ASCE World Environmental and Water Resources Congress 2009. Kansas piration. Journal of Hydrologic Engineering 13: 51–63.
City, KS. Proceedings (CD); 11. Trezza R. 2002. Evapotranspiration using a satellite-based surface energy
Kjaersgaard J, Allen RG, Irmak A. 2011. Improved methods for balance with standardized ground control. PhD Diss., Utah State Univ.,
estimating monthly and growing season ET using METRIC applied Logan, Utah.
to moderate resolution satellite imagery. Hydrological Processes Trezza R, Allen RG. 2009. Estimation of Monthly and Annual ET for
25(26): 4028–4036. DOI: 10.1002/hyp8394. Morocco areas. Report to Riverside Technology, Inc., Ft. Collins, CO. 9.
Kustas WP, Zhang Z, Schmugge TJ. 1998. Combining optical and Van Dam RL, Borchers B, Hendrickx JMH. 2005. Strength of landmine
microwave remote sensing for mapping energy fluxes in a semiarid signatures under different soil conditions: implications for sensor
watershed. Remote Sensing of Environment 64: 116–131. fusion. International Journal of Systems Science 36: 573–588. DOI:
Liang S. 2000. Narrowband to broadband conversions of land surface 510.1080/00207720500147800
albedo. I: Algorithms. Remote Sensing Environ. 76: 213–238. Webb EK. 1970. Profile relationships: the log-linear range, and extension
Markham BL, Barker JL. 1986. Landsat MSS and TM post-calibration to strong stability. Quarterly Journal of the Royal Meteorological
dynamic ranges, exoatmospheric reflectances and at-satellite temperatures. Society 96: 67–90.
EOSAT Landsat Technical Notes 1:3–8. Earth Observation Satellite WigmostaMS,VailL,LettenmaierDP.1994.Adistributedhydrology-vegetation
Company, Lanham, Maryland. model for complex terrain. Water Resources Research 30: 1665–1679.
Miller TW, Hendrickx JMH, Borchers B. 2004. Radar detection of buried Wright JL. 1982. New evapotranspiration crop coefficients. Journal of the
landmines in field soils. Vadose Zone J 3: 1116–1127. Irrigation and Drainage Division (ASCE) 108: 57–74.
MODIS UCSB Emissivity Library. 2004. University of California Santa Wukelic GE, Gibbons DE, Martucci LM, Foote HP. 1989. Radiometric
Barbara. http://www.icess.ucsb.edu/modis/EMIS/html/em.html. Last calibration of Landsat thematic mapper thermal band. Remote Sensing
visited, 5 February 2007. of Environment 28: 339–347.

Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 25, 4011–4027 (2011)

View publication stats

You might also like