Journal of Hydrology 212–213 (1998) 213–229

A remote sensing surface energy balance algorithm for land

(SEBAL)
2. Validation
W.G.M. Bastiaanssen a,*, H. Pelgrum a, J. Wang b, Y. Ma b, J.F. Moreno c,
G.J. Roerink a, T. van der Wal a
a
DLO-The Winand Staring Center for Integrated Land, Soil and Water Research (SC-DLO), P.O. Box 125, 6700 AC,
Wageningen, The Netherlands
b
CAS-Lanzhou Institute of Plateau Atmospheric Physics, 316 Dong-gang Xi-lu, Lanzhou, Gansu 730000, P.R. of China
c
University of Valencia, Faculty of Physics, Department of Thermodynamics, Campos de Burjassot, 46100 Valencia, Spain

Abstract
The surface fluxes obtained with the Surface Energy balance Algorithm for Land (SEBAL), using remote sensing informa-
tion and limited input data form the field were validated with data available from the large-scale field experiments EFEDA
(Spain), HAPEX-Sahel (Niger) and HEIFE (China). In 85% of the cases where field scale surface flux ratios were comparedwith
SEBAL-based surface flux ratios, the differences were within the range of instrumental inaccuracies. Without any calibration
procedure, the root mean square error of the evaporative fraction \Delta (latent heat flux/net available radiation) for footprints of
a few hundred metres varied from LRMSE ˆ 0.10 to 0.20. Aggregation of several footprints to a length scale of a few kilometres
reduced the overall error to five percent. Fluxes measured by aircraft during EFEDA were used to study the correctness of
remote sensed watershed fluxes (1,000,000 ha):The overall difference in evaporative fraction was negligible. For the Sahelian
landscape in Niger, observed differences were larger (15%), which could be attributed to the rapid moisture depletion of the
coarse textured soils between the moment of image acquisition (18 September 1992) and the moment of in situ flux analysis
(17 September 1992). For HEIFE, the average difference in SEBAL estimated and ground verified surface fluxes was
23 W m ⫺2, which, considering that surface fluxes were not used for calibration, is encouraging. SEBAl estimates of evaporation
from the subsealevel Qattara Depression in Egypt (2,000,000 ha) were consistent with the numerically predicted discharge from
the groundwater system. In Egypt’s Nile Delta, the evaporation from a distributed field scale water balance model at a 700,000
ha irrigated agricultural region led to a difference of 5% with daily evaporative fluxes obtained from SEBAL. It is concluded
that, for all study areas in arid zones, the errors average out if a larger number of pixels is considered. Part 1 of this chapter deals
with the formulation of SEBAL. 䉷 1998 Elsevier Science B.V. All rights reserved.
Keywords: Surface fluxes, validation, remote sensing

1. Introduction available for measurements of natural variations in

surface fluxes over heterogemeous land surfaces, the
Due to the limited technical and financial means performance of remote sensing flux algorithms cannot
be straightforwardly assessed. Sometimes, surface
* Corresponding author. Corresponding address. ITC, P.O. Box flux maps estimated by the use of remote sensing
6, 7500 AA, Enschede, The Netherlands. algorithms have not been verified at all (e.g. Gurney
0022-1694/98/$ - see front matter 䉷 1998 Elsevier Science B.V. All rights reserved.
PII: S0022-169 4(98)00254-6
214 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

and Hall, 1983; Ottle et al., 1989; Smith and Choudh- relatively small source area, viz. the footprint with
ury, 1990). Diak (1990) determined regional fluxes length x in the upwind direction which contributes
from radio soundings, which provide useful informa- to the establishment of a certain flux. The measured
tion on the bulk exchange processes of complex land- flux F0 relates to the orientation and length of this
scapes that can be used to calibrate regionally footprint in the upwind direction, which vary with
averaged surface energy balances estimated from the height above the surface at which the fluxes are
remote sensing data. Such data sets are, however, measured, zmeas, as well as with surface parameters
less suitable for validating field scale fluxes. In ideal such as the surface roughness z0m, the displacement
situations, surface fluxes should be measured in situ height d, and wind variables such as speed u and
and simultaneously on all dominant land use types. direction (e.g. Itier et al., 1994). Schuepp et al.
Taconet et al. (1986) and Kustas et al. (1990) used (1990) showed analytically that the contribution to
three to four stations to validate their meso scale the measured flux of land surface elements adjacent
surface flux map, trading off between what was to the tower is significantly higher than that from land
required and what was technically and financially surface elements located at a greater distance (x) from
feasible for a relatively small team of scientists and the tower:
technicians. The availability of data from large-scale Z
field experiments held over the last ten years should F0 x† dx  
uavg zmeas
be considered the best option for verifying the ˆ exp ⫺ ⫺† 1†
F0 uⴱ kx
regional surface energy balance estimated from R
remote sensing data. Examples are given for FIFE where F0 x†dx=F0 represents the cumulative flux at a
(Sellers and Hall, 1992), EFEDA (Bastiaanssen et al., distance x relative to a flux measurement at height
1997) and HAPEX-Sahel (Chehbouni et al., 1997). zmeas representing a spatially constant flux and u
Evaporation values from remote sensing data can (m s⫺1 ) is the friction velocity. Eq. (1) is a tool for
also be validated with regionally calibrated water integreating the pixel wise fluxes in a non-linear
balances. Hydrological models are classically cali- fashion in such a way that they become compatible
brated against time series of pressure heads, soil with tower-based fluxes. The average wind speed uavg
moisture and surface runoff. The present paper of the profile to with u applies (see Eq. 1) can be
assesses and summarizes the potential of several calculated from:
procedures to validate remote sensing algorithms in ln zmeas =z0m † ⫺ 1 ⫹ z0m =zmeas
a range of landscapes and climates. Case studies in uavg ˆ u m s⫺1 † 2†
k 1 ⫺ z0m =zmeas †
different agro-ecosystems were used to reveal how the
accuracy of SEBAL changes with the spatial scale and Unless the landscape is strictly uniform, the exact
the type of landscape. The validation procedures field orientation and the wind direction of the source
tested include: areas is crucial for validating pixel wise surface
fluxes. Although it is felt that the aspect of non-linear
flux integration is crucial in validation studies, most
1. Turbulent surface fluxes measured in the field;
studies have ignored this issue: pixel-based fluxes
2. Airborne turbulent flux measurements;
have been compared directly with tower-based flux
3. Soil moisture profiles measured in the field;
measurements without considering footprint aspects.
4. Hydrological models.
Preferably, the pixel size should be a small fraction of
the footprint, so that the areal integration pixel wise
2. Validation by the means of turbulent surface F0(x) values can be obtained according to their
fluxes measured in the field distance from the pixel on which the tower is located.
Validation of low resolution flux maps with pixels
2.1. Footprint issues larger than the footprint requires a nested calibration
procedure between tower-based fluxes, high resolu-
Field measurements of turbulent surface fluxes by tion surface flux maps and low resolution flux maps
means of instrumented towers are representative of a (Fig. 1).
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 215

Fig. 1. Multi-scale validation procedure required for the verification of remote sensing-based surface flux maps with different spatial resolu-
tions. The NS001 is a Daedalus AAD1268 twelve channel multi-spetral scanner mounted on an aircraft.

2.2. NS001 measurements of the Barrax and Since H changes almost continuously with the solar
Tomelloso super-sites (EFEDA) elevation, the instantaneous remote sensing based H-
flux could not be compared with half-hourly or hourly
A plateauthreatened by land degradation between integrated H-fluxes. To overcome this inconsistencies,
2⬚–3⬚30 0 W and 39⬚–40⬚ N in Central Spain was the evaporative fraction (see Eq. 3) was computed
selected for an in-depth analysis of water and energy (Table 1). Experimental evidence indicates that for
balances during EFEDA (Fig. 2). The EFEDA study homogeneous (Shuttleworth et al., 1989) and hetero-
area consists of three super-sites, situated at distances genous land surfaces, L, is temporarily stable between
of 70 km from one another: the dry farming area of 1000 and 1500 hours local time (Bastiaannssen et al.,
Tomelloso, the irrigated area of Barrax and the hilly 1996). The evaporative fraction has therefore been
Rada de Haro areas where agriculture is only margin- selected as a basis to validate the instantaneous
ally feasible (Bolle et al., 1993). A joint field surface energy balance.
programme was organised to conduct simultaneous lE lE
field scale flux measurements. Surface fluxes were L ˆ ⫺† 3†
Q ⴱ ⫺G0 lE ⫹ H
measured at 21 stations, most of which were located
at the Tomelloso super-site. The turbulent surface fluxes at the 13 stations were
During the airborne NS001 mission on June 29 measured using eddy correlation, Bowen-ratio and
(1991), visible and infrared measurements were flux profile-energy budget methods. The H and lE
made from flying altitude. The NS001 pixel size fluxes measured in the field have an error due to
was 18.5 m, which is suitable for footprint and inte- instrument uncertainty and calculation procedures to
gration. As a consequence of irrigation practices, The obtain fluxes. In accordance with the technique
Barrax region is characterised by a patchy type of applied to measure H and lE, the error orginating
agriculture, with a mixture of rainfed and irrigated from inaccurate H and lE fluxes and propagated
fields (see Fig. 11). The validation of the SEBAL- into L was computed for each station. The allocation
based fluxes is demonstrated in Fig. 3. A total of 13 of the pixels in the surroundings of the tower affecting
individual flux towers were used for this purpose. the in-situ flux was done using the theories incorpo-
Most of the stations (10) were situated 5 to 15 km rated in Eq. (1). In the absence of sufficiently accurate
east of Tomelloso. The Barrax stations (3) were situ- data on u , d, u and z0m on June 29, it was impossible
ated at La Gineta, 5 km east of the Barrax village. to calculate the 13 different footprints. hence Eq. (1)
The H and lE fluxes measured in situe by the could not be applied. Two rectangular integration
EFEDA participants had different integration times. regions of (i) 5*5 and (ii) 7*13 pixels were chosen
216 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 2. Location of the EFEDA grid in Castilla la Mancha, Central Spain. The Barrax, Tomelloso and Rada de Haro super-sites are indicated.

instead, and all pixels were given equal weight. This is 2.3. Landsat thematic mapper measurements at
not correct, but it was one option to include the avail- Castilla la Mancha (EFEDA)
able flux data in the validation procedure. It was
concluded that all towers in Fig. 3, without exception, EFEDA’s ultimate goal was the development of a
lie within an envelope of one times the standard devia- set of techniques to assess the bulk surface energy
tion. Using the centre of each error bar, the root mean balance at grid square scale of numerical atmospheric
square error became LRMSE ˆ 0.10. The slope of the models. Since a Thematic Mapper image covers 100%
fitted line for 5*5 pixels was 0.997 (R 2 ˆ 0.79) while of the EFEDA-grid, SEBAL was executed from the
7*13 pixels induced a clear cut shift in the slope of cloud free overpass on June 12. Terrain elements
1.488 (R 2 ˆ 0.76). Table 1 shows that the selection of comprising agricultural crops and natural vegetation,
the size of the footprint (25 or 91 pixels) had conse- which were not the subject of intensive field research
quences mainly for flux stations 4 and 10. The signif- during the Special Observation Period (SOP), could in
icant deviation of station 4 from the 1:1 line is purely this way be incorporated in the determination of the
related to the selection of the integration pixels. regional surface energy balance. The evaporative
Station 5 (CNRM-2) tended to have a poor perfor- fraction based on TM data was validated against in-
mance regardless of the footprint selected. situ flux data (Fig. 4). Again, 13 flux towers could be
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 217

Fig. 3. Validation of the SEBAL-derived evaporative fractions, \Lambda SEBAL, against tower measurements of evaporative fractions, Ltower
for June 29, 10 21 0 GMT, 1991 (NS001). Error bars are indicated. EFEDA participants are indicated by number. The footprint of each flux
tower is 92.5*92.5 m 2.

Table 1
Data on evaporative fraction derived using the SEBAL parameterization and in-situ tower measurements collected by various EFEDA
participants during the Special Observation Period. The values between brackets are the standard deviations of the areally integrated at-
pixel values (in the case of SEBAL) or standard measurement errors (in the case of tower data), NS001. 10:21 h GMT, June 289, 1991.
Flux stations 9, 10 and 11 are situated in Barrax. The other stations were set up in Tomelloso

Institution Land use Coordinates Ground-based SEBAL SEBAL

UTM L-values 25 pixels 91 pixels

Wageningen Univ. Fallow 508919.50, 4333460.00 0.12 (0.06) 0.14 (0.08) 0.17 (0.09)
(Wat)
Inst. of Hydrology Vetch 504440.39, 4333523.69 0.21 (0.03) 0.22 (0.04) 0.20 (0.06)
Inst. of Hydrology Vineyard 509309.50, 4333010.00 0.31 (0.04) 0.33 (0.08) 0.30 (0.13)
CNRM-4 Vetch 504089.50, 4333580.00 0.29 (0.15) 0.08 (0.07) 0.17 (0l24)
CNRM-2 Vineyard 506309.50, 4332350.00 0.43 (0.21) 0.19 (0.06) 0.17 (0.07)
Wageningen Univ Vineyard 506009.50, 4332290.00 0.25 (0.03) 0.20 (0.11) 0.21 (0.10
(Met)
DLO Winand Staring Vineyard 505859.50, 4332320.00 0.26 (0.09) 0.21 (0.12) 0.20 (0.11)
Centre
Copenhagen Vineyard 506399.50, 4330940.00 0.22 (0.03) 0.26 (0.05) 0.25 (0.06)
University a
CNRM-6 Fallow 576989.50, 4325240.00 0.14 (0.07) 0.14 (0.05) 0.14 (0.05)
Karlsruhe University Maize 575129.50, 432070.00 0.82 (0.10) 0.72 (0.36) 0.54 (0.44)
Karlsruhe University Fallow 575039.50, 4324460.00 0.03 (0.01) 0.02 (0.13) 0.09 (0.13)
Reading University a Vineyard 509249.50, 4333250.00 0.31 (–) 0.31 (–) 0.31 (–)
Reading University a Vetch 504299.50, 4333580.00 0.28 (–) 0.25 (–) 0.25 (–)
a
For June 28, taken from Bolle and Streckenbach, 1992
218 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 4. Validation of the SEBAL-derived evaporative fractions, L SEBAL, against tower measurements of evaporative fractions, L tower for
June 12, 12, 10:12 h GMT, 1991 (Thematic Mapper). Error bars are indicated. EFEDA participants and indicated by number. The footprint of
each flux tower was calculated by means of Eq. (1).

included in the comparison,although they were not the the left hand side of Eq. (1) represents the aggregated
same towers as presented in Fig. 3 (see Table 2). The flux from heterogeneous terrain, assuming that the
footprints for the selection of the pixel-fluxes to be theory holds for such terrain The total horizontal
areally integrated were calculated according to the distance in the upwind direction for which the cumu-
guidelines provided by Eq. (1). In this calculation, lative flux contribution is approximately 100% was

Table 2
Data on evaporative fraction derived with the SEBAL parameterization and in-situ tower measurements collected by various EFEDA parti-
cipants during the Special Observation Period. The values between brackets are the standard deviations of the areal integrated at-pixel values (in
the case of SEBAL), or standard measurement errors (in the case of tower data), Landsat Thematic Mapper, 10:12 h GMT, June 12, 1991. Flux
stations 1 and 2 were near Barrax, 18, 19 and 21 at Rada de Haro and the remaining stations were set up at the Tomelloso super-site

Institution Land use Coordinates Ground-based SEBAL

UTM L-values Schuepp
method

Karlsruhe University Maize 575129.50, 4324070.00 0.86 (0.10) 0.85 (0.57)

Karlsruhe University Fallow 575039.50, 4324460.00 0.19 (0.02) 0.54 (0.23)
Reading University Vetch 504299.50, 4333580.00 0.33 (0.11) 0.45 (0.07)
Reading University Vineyard 509249.50, 4333250.00 0.39 (0.20) 0.30 (0.22)
CNRM-2 Vineyard 506309.50, 4332350.00 0.48 (0.25) 0.32 (0.19)
CNRM-4 Vetch 504089.50, 4333580.00 0.48 (0.25) 0.51 (0.05)
DLO Winand Staring Centre Vineyard 505859.50, 4332320.00 0.24 (0.07) 0.32(0.10)
Wageningen Univ. (Met) Vineyard 506009.50, 4332290.00 0.22 (0.11) 0.34 (0.08)
Wageningen Univ. (Wat) Fallow 508919.50, 4333460.00 0.32 (0.12) 0.32 (0.08)
Copenhagen University Vineyard 506399.50, 4330940.00 0.17 (0.02) 0.31 (0.12)
Berlin University Field 2 533189.50, 438680.00 0.58 (0.07) 0.56 (0.28)
Berlin University Field 3 533369.50, 4380560.00 0.32 (0.12) 0.69 (0.16)
Copenhagen University Sunfl. 534149.50, 4378490.00 0.27 (0.03) 0.65 (0.13)
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 219

Fig. 5. Area variation in evaporative fraction within the footprint of an individual flux tower. Part A: Tiger bush, Part B: Millet, Part C: Fallow
Savanah and Part D: Weighing scheme for the non-linear integration of the upwind fluxes for each individual flux tower where RSS is Relative
Source Strength.

estimated as an average of 1425 m for the 13 stations. improves for a larger set of heterogeneous land
The hight of the eddy correlation systems varied surface elements (DL eff/L eff ˆ 0.94/0.45 ˆ 0.10).
between 10 and 25 m.
Of the 13 stations, 4 were situated outside the 2.4 Thematic mapper measurements at sahel
envelope based on one time the standard deviation. (HAPEX)
As such, the algorithm’s performance was not as
good as it was for the NS001 images. The root mean The success of predicting the atmospheric circula-
square error relative to the centre of the error bar was tion processes over the Sahel is greatly affected by the
also much larger: LRMSE ˆ 0.19. The bulk behaviour description of the land surface fluxes (Xue and
of a larger area encompassing all 13 footprints yielded Shukla, 1990). An intensive international expedition
LTower
eff
ˆ 0.45 while SEBAL gave LSEBAL eff
ˆ 0.49, was organized in the autumn 1992 to measure the
indicating that the overall SEBAL performance fluxes from these scattered natural bushlands
220 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 6. Measured and SEBAL estimated surface energy balances in a desert-oases system on July 9, 1991, in the Heihe river basin, Gansu,
China.

(Goutorbe et al., 1994). September 14 and 15 were and L eff ˆ 0.63 for the whole TM scene. The 11 flux
rainy days. Application of SEBAL for this region on stations gave on average L ˆ 0.66 for millet, L ˆ 0.79
September 18, 1992, as 2 to 3⬚E and 13 to 14⬚N (see for fallow savannah and L ˆ 0.76 for patterned wood-
Fig. 12) led to the conclusion that a difference of land. The systematic deviation with remote sensing
LRMSE ˆ 0.14 between LSEBAL and in-situ measure- based estimates of L can be explained by the rapid
ments of L was unavoidable (Kabat et al., 1997). soil moisture depletion of coarse and bare sands
The equivalent relative error then becomes DL/ between September 17 and 18.
L eff ˆ 0.20. This conclusion, was, however, based
on3 flux stations only, which is insufficient to draw
general conclusions. Fig. 5 shows the frequency distri- 2.5 Thematic mapper measurements of the Heihe river
bution of the fluxes withinthe footprint of the flux basin (HEIFE)
towers placed in tiger bush, millet and fallow savan-
nah. The effect of linear and non-linear averaging The Heihe river basin, located at 100⬚E and 39⬚N in
especially affected the millet field, while the tiger the gansu Province in the North-western part of the
bush and fallow savannah were less sensitive to the People’s Republic of China, has been the focus of a
flux weighing procedure. Each tower had different special research programme on the water and energy
integration lengths, vis. 6200, 1650 and 2000 m for cycle between mountains, oases and desert since
tiger bush, millet and fallow savannah, respectively. 1988. This HEIFE project (Mitsuta, 1994) involves
Gash et al. (1997) thoroughly analysed the flux permanent observations of essential hydro-meteor-
measurements on September 17 (one day before ological parameters dispersed across the area and
Thematic Mapper overpass) and their data were includes regular field campaigns in which surface
included in the current validation procedure because fluxes are measured with eddy correlation devices.
of the larger number of stations (n ˆ 11). SEBAL Basically, the flux stations have been set up in a topo-
estimates L eff ˆ 0.57 for the West Central SAupersite sequence going from the irrigated oases fed by water
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 221

Fig. 7. Temporal behaviour of the aggregated evaporative fraction L eff for the entire 900,000 ha EFEDA grid according to SEBAL and aircraft
flux measurements made Falcon 20 and Dornier 128.

from the Heihe river, via sand desert to a stony desert 3. Validation by means of airborne turbulent flux
surface (Gobi type of desert surface). SEBAL was measurements
tested with a Thematic Mapper image of July 9,
1991 (Wang et al., 1995, Ma, 1996) with the aim to 3.1. Iberian peninsula, using NOAA-AVHRR and
extrapolate point measurements obtained. Field data METEOSAT measurements (EFEDA)
of hemispherical surface reflectance and surface
temperature were used to atmospherically correct A description of macro scale surface energy
the spectral radiance’s measured by a satellite situated balances of the entire Iberian Peninsula was obtained
at the top of the atmosphere. Some estimates of by retrieving surface temperature T0 and hemispheri-
surface roughness (z0m) on the basis of wind profile cal surface reflectance r0 from METEOSAT and
measurements (u(z)) were used to fit the emperical NDVI from NOAA-AVHRR. Since the METEOSAT
relationship between z0m and NDVI. The areal distri- flux maps cannot be compared with field measure-
bution of the latent heat flux is depicted in Fig. 13. The ments directly (pixel size larer than footprint, see
fluxes resulting from SEBAL were compared after- Fig. 1), a comparison with aircraft flux measurements
wards with corresponding values measured at ground has been worked out.
level (Fig. 6). As the figure shows the soil heat flux The DLR Falcon 20 aircraft was flown in the
gave the poorest results. This was not considered a EFEDA study area on June 19, 21, 25 and 28, 1991.
serious problem, because microscale soil heat flux Eddy correlation apparatus were onboard the aircraft
measurements are representative for a very small to measure H and lE fluxes at flighing altitude. This
sphere of influence and therefore incompatible with independent data source was used to validate a set of
the size of one TM-pixel anyhow. Since field 520 METEOSAT pixels covering the EFEDA grid
measurements of Q*, G0, H and lE all have different square. The flight pattern on June 28 consisted of
footprints, the energy budget from measurements in parallel legs to obtain the horizontal variability of
composite terrain is seldom neutral. hence, the the fluxes at 350 m height across the full EFEDA
results in Fig. 6 for complex terrain conditions domain (Jochum et al., 1993; Jochum, 1993). As the
should, with an average RMSE of 23 W m ⫺2, be spatial patterns of fluxes change with height, aircraft
perceived as encouraging. The difference of fluxes cannot be compared with surface fluxes result-
25 W m ⫺2 in latent heat flux is equivalent to a ing from remote sensing. Instead, evaporative frac-
relative error of 25/244 ˆ 10% which, for one tion, L, of an area can be taken. The average
single footprint, is within the range of uncertainty evaporative fraction, based on linear averaging of
of the field instruments. the H and lE fluxes for legs ML1, ML2, ML4, ML6
222 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 8. Non-linear trend between wetness indicators and volumetric soil water content at a depth of 10 cm (Q10) for Tomelloso and Barrax,
enclosing a range of surfaces (vineyard, vetch, bare soil, maize, barley, alfalfa) and soils (reddish sandy loam, loamy sand, limestone) with
different soil moisture conditions (0.05–0.30 cm 3 cm ⫺3): Bowen ration b (part A), evaporative fraction L (part B), Priestly and Taylor a (part
C), bulk surface resistance (part D)), June 29, 10:21 h GMT.

and ML7 was calculated as LFalcon eff

ˆ 0.24.The 4. Validation by means of soil moisture profiles
METEOSAT June 29 images for an area of 85 km ⴱ measured in the field
125 km covering almost the entire EFEDA grid
yielded an average of LMeteosat
eff
ˆ 0.21. Since the In addition to ground observations of surface
zone around the Embalse de Alarcon reservoir contri- fluxes, a soil moisture monitoring programme was
butes significantly to 具lE典 and is not including the executed during EFEDA (Droogers et al., 1993).
spatial integration of lE(x,y) because of clouds The energy partitioning in arid conditions relates
above the reservoir, the LMeteosat
eff
-value should be primarily to near-surface soil water content (e.g.
slightly higher, approaching the Falcon L eff ˆ 0.24 Owe and Van de Griend, 1990). Information on the
eff
value rather closely (Lmeteosat eff
–LFalcon ⬃ 0.0 to 0.02). regional distribution of soil moisture must therefore
Fig. 7 shows that the rapid changes in L eff during the match the partitioning between H and lE. Four
course of June from airborne turbulent flux measure- common expressions for energy partitioning or
ments. The SEBAL calculations agree rather well ‘wetness indicators’, derived from SEBAL output,
with this trend which indicates that L eff of a watershed have been compared with field scale near-surface
can be accurately assessed from meteorological satel- soil moisture measurements: Bowsen ratio b,
lites. The same conclusion was drawn by van den Priestley and Taylor a, evaporative fraction L and
Hurk et al. (1997) for the evalorative conditions of bulk surface resistance, rs. The relationship between
the entire Iberian Peninsula using synoptical stations these wetness indicators and the near-surface soil
for screen light air temperature and air humidity. water content provides another yardstick to check
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 223

conditions at a depth of 10 cm, Q10, against the

SEBAL-derived wetness indicators of that particular
field; they compare well. Evidently, rs shows a better
relationship with Q10 than b, a and L. There is a well-
known non-linearity between the resistance to bare
soil evaporation rsoil and Q (Kondo et al., 1990) and
between the resistance to crop evaporation rc and Q
(Jarvis, 1976) for homogeneous surfaces. Fig. 8
allows the conclusion that the non-lienarity between
a bulk surface resistance rs and Q also holds for
heterogeneous landscapes (vineyard, vetch, bare
soil, maize, barley and alfalfa at different develop-
Fig. 9. Relative difference in daily evaporation between SIWARE ment stages, different soil types and different moisture
and SEBAL for the six main irrigation ommand areas. The numbers depletion phases).
represents the (1) Ismaileya, (2) Rayah Tawfiki, (3) Mansureya, (4)
Sharqaweya, (5) Bassoseya and (6) Abu Managa areas.
5. Validation by means of hydrological models
the reliability of L or lE maps. The measurements of
Apart from aircraft fluxes, a general shortcoming of
the Time Domain Reflectometer (TDR) reflect the
the validation procedures based on in situ measure-
moisture situation of one particular pixel.
ments is that only a small number of selected point
Over 46 plots and 13 test fields (some fields include
observations could be used. An alternative way to
more than one plot), soil mosture was monitored at
obtain regional evaporation rates is by studying the
several standard depths between 5 and 50 cm. Four
water balance of watersheds, river basins or deltas.
different test fields were located in the Tomelloso area
(vineyard, vetch, bare soil), while the Barrax area 5.1. Qattara Depression, using Thematic Mapper
included nine test fields (maize, fallow, barley, measurements (Egypt)
alfalfa). The incorporation of moisture data in Barrax
was particularly valuable, since the surface fluxes in The floor of the 2,000,000 ha vast sub-sea level
Barrax were available only for three locations (see Qattara Depression in the Westerm Desert of Egypt
Table 1). Fig. 8 presents field-averaged moisture overlies the fresh water Nubian Sandstone aquifer.
Since human activities in this hyper-arid climate with-
out rainfall and runoff are negligible (no groundwater
extraction, no crop evaporation) and the phreatic level
is stationary throughout the year, the water balance of
the Qattara Depression becomes very simple: ground-
water inflow is equal to evaporation. Fig. 14 illustrates
the general evaporation features estimated with
SEBAL in the Eastern part of the Qattara Depression
on August 7, 1986. Higher evaporation spots coincide
with the locations of the sebkha’s and salt marshes
where the water table almost reaches the surface.
Numerical groundwater flow models were applied
to assess whether estimated groundwater losses by
evaporation were consistent with water flow into the
depression (Meneti et al., 1991). The model study
Fig. 10. Determination of the relative deviation DE/E for each of showed that the SEBAL estimations of evaporation
the 82 irrigation units of SIWARE on August 5, 1986. The average were consistent with the calculated groundwater
value turned out to be DE/E ˆ ⫺0.08 ^ 0.26. flow of 2100 Mm 3 yr ⫺1 (FEMSAT model) and
224 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 11. NS001-based evaporation map for Barrax estimated with the SEBAL parameterization, June 29, 1991. Time integration of instanta-
neous to 24 hour values was realized by holding evaporative fraction constant and adjust the net radiation.

2300 Mm 3 yr ⫺1 (TRIWACO model) using observed to correct for less optimal soil moisture and salinity
pressure heads and large values for the saturated conditions. When this category of models is locally
hydraulic conductivity ksat. Large ksat values, ranging calibrated against soil moisture, groundwater table
from 300 to 600 cm d ⫺1, could be physically obtained flucations and return flow, they give reasonable esti-
by the lower water viscosity at 60 to 70⬚C which mates of regional evaporation.
prevails in deep aquifers (⬎4000 m). Pelgrum (1992) focussed on the water consumption
of the irrigated Nile Delta. An attempt was made to
5.2. Nile Delta, using METEOSAT measurements determine the surface energy balance for each
(Egypt) daytime hour using METEOSAT-based T0 and r0
data. The SEBAL procedure was repeated for each
Field water balance models usually compute hourly image during daylight (12 images and thus
evaporation from vegetation and soil according to 12 SEBAL runs). A large set of instantaneous surface
the unstressed transpiration and bare soil potential energy balances between 7 AM and 6 PM was so
evaporation. Empirical reduction factors based on obtained Feddes et al., 1993). The 24 hour SEBAL
actual soil water content and salinity, are then applied evaporation rates obtained for the Nile Delta were
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 225

Fig. 12. Thematic Mapper-based evaporative fraction map of the HAPEX-Sahel grid estimated with the SEBAL parameterization, September
18, 1992.

compared by Bastiaanssen et al. (1992) with the 24 network in the Eastern Nile Delta distincts 82
hour evaporation figures obtained from the calibrated irrigation units. The total gross irrigated area covered
hydrological model SIWARE (Abdel Gawad et al., by the 82 irrigation units is 695,000 ha. The results
1991). At the time of image acquisition in August for six irrigation commands are demonstrated in Fig.
1986, maize, cotton, orchards and vegetables were 9. The relative error of the command areas DE24/
grown on the irrigated Nile delta. The SIWARE E24 ˆ (SIWARE-SEBAL)/SIWARE appeared to be
226 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Fig. 13. Thematic Mapper-based latent heat flux map of the Heihe river basin estimated with the SEBAL parameterization, July 9, 1991, 9:50 h
local time.

suprisingly small (Fig. 9), with a mean for the 6 cases 6. Summary and conclusions
of DE24/E24 ˆ ⫺0.08 (SD ˆ 0.06).
A further down-scaling allows extension of the The five different calibration options all have a
comparative study to the level of each of the 82 irriga- certain potential (although groundwater flow models
tion units. The average difference between SEBAL can only be used for a first order estimate; they are just
and SIWARE for the individual irrigation units was quantitative). Depending on the spatial scale at which
DE24/E24 ˆ ⫺0.08 (SD ˆ 0.26). With the exception of the remote sensing study is performed (field,
units 15, 19 and 27, the results were encouraging (see watershed, river basin, continent), a validation proce-
Fig. 10). For all 82 irrigation units together, daily dure can be selected using the summary provided in
values of SEBAL were found to be 5.1% higher Table 3.
than the SIWARE predictions. A 5.1% deviation is The difference between tower-based and remote
within the allowable range of the SIWARE model sensing-based L-data was found to range from
accuracy (10% on an annual basis). Hence, the error LRMSE ˆ 0.10 to 0.20 for single footprints of several
in evaporation from remote sensing increases as one hundreds of metres, independent of land use types
goes from extensive composite regions (5.1% devia- (vineyard, bare soil, vetch, irrigated maize, irrigated
tion) via command areas (8% deviation ^ 6%) to cotton, sunflower, patterned woodland, millet, Gobi
isolated irrigation units (8% ^ 26%). desert and sebkha’s). The error is reduced substantially
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 227

Fig. 14. Thematic Mapper-based latent heat flux map of the Qattra Depression estimated with the SEBAL parameterization and aggregated
towards a spatial resolution of 1 km * 1km (NOAA-AVHRR scale), August 7, 1986, 9:50 h local time.

if the individual footprints are areally aggregated to a Depression and Eastern Nile Delta, it was concluded
length scale to a few km (DL eff ⬃ 0.05). At watershed that the evaporation is within the confidence limits of
scale, where aircraft fluxes are meaningful, such as for the calibrated water balance models. The validation for
Castilla la Mancha (1,000,000 ha), the error was natural woodlands such as Rada de Haro and the wood-
DL eff ⬃ 0.01. For the large areas in the Qattara lands in Niger gave moderate results, which in the

Table 3
Evaluation of the validation procedure tested and accuracies obtained for the SEBAL evaporation studies in arid zones

Option Advantage Disadvantage SEBAL Scale

Accuracy
(%) (km)

1. Tower fluxes Direct flux data, field scale Non-linear averaging of fluxes 85 to 95 0.5 to 5
2. Aircraft fluxes Direct flux data, large area coverage, Expensive, advection, height 99 100
regional scale dependence of fluxes
3. Soil moisture Geometric coupling with pixel wise Fluxes are indirectly validated 90 0.5
fluxes is easy, field scale
4. Groundwater models Physical constraints can be assessed Uncertain geo-hydrological 81 100
regional scale schematisation
5. Field water models Multi-scale validation feasible Calibration compulsory for realistic 92 10
regional and field scale water balances 95 100
228 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229

Niger case is most probably related to rapid moisture and M. Badji (eds). Advances in Planning, Design and Manage-
depletion. ment of Irrigation Systems as Related to Substainable Land use,
Center for Irrigation Engineering and ECOWARM, Leuven,
These examples indicate that evaporation assess- Belgium: 473–483.
ments from remote sensing data using new type of Bastiaanssen, W.G.M., Pelgrum, H, Meneti, M., Feddes, R.A.,1996.
empirical relationships are feasible with a minimum Estimation of surface resistance and Priestley-Taylor a-para-
of collateral input data. SEBAL is not a profitable meter at different scales. In: J.B. Stewart et al. (eds). Scaling
solution under all circumstances; the attainable accu- up in hydrology using remote sensing. Institute of Hydrology,
Wallingford: 93–111.
racy changes with the degree of land surface hetero- Bolle, H.J. and Streckenbach, B., 1992. The ECHIVAL field experi-
geneity in relation to pixel size and correlation length ment in a desertification-threatened area EFEDA. First Annual
of a given landscape (see Bastiaanssen et al., 1996 for Report to EC.EFEDA-Secretariat, Free University of Berlin,
more details). Berlin, Germany.
Future large scale field experiments should pay Bolle, H.J. et al., 1993. EFEDA: European field experiments in a
desertification-threatened area. Annales Geophysica 11, 173-189.
more attention to an exact determination of the foot- de Bruin, H.A.R., van den Hurk, B.J.J.M., Kohsiek, W.,1995. The
print orientation and length, in order to calculate the scintillation method tested over a dry vineyard area. Boundary
fractional contribution of land surface elements to the Layer Met. 76, 25–40.
total flux measured by the towers. The prospects of Chehbouni, A., Lo Seen, D., Njoku, E.G., Lhomme, J.P., Monteny,
long-path scintillometers look promising (De Bruin et B., Kerr, Y.H., 1997. Estimation of sensible heat flux over spar-
sely vegetated surfaces, J. of Hydr. 188–189, 855–868.
al., 1995). The new generation of satellite platforms Diak, G.R., 1990. Evaluation of heat flux, moisture flux and aero-
(Earch Observing System, Envisat) may provide dynamic roughness at the land surface from knowledge of the
improved measurements of surface roughness, surface PBL height and satellite-derived skin temperatures. Agr. and
emissivity and long wave incoming radiation, which Forest Met. 52, 181–198.
would reduce the number of empirical relatioships Droogers, P., v.d. Abeele, G.D., Cobbaert, J., Kim, C.P., Rossler-
ova, R., Soet, M., Stricker, J.N.M., 1993. Basic data sets
currently used by SEBAL. description and preliminary results of EFEDA-Spain. Report
37. Wageningen Agricultural University Wageningen, Depart-
ment of Water Resources, Wageningen, The Netherlands.
Acknowledgements Feddes, R.A., Menenti, M., Kabat, P., Bastiaanssen, W.G.M., 1993.
Is large-scale inverse modelling of unsaturated flow with areal
The authors are indebted to Mrs Mieke van Dijk average evaporation and surface soil moisture as estimated from
remote sensing feasible? J. of Hydrology, 143, 125–152.
and Mr Bram ten Cate for their careful preparation of Gash, J.H.C., Kabat, P. et al., 1997.The variability of evaporation
the final manuscript while the first author was on during the HAPEX-Sahel Intensive Observation Period. Journal of
leave. The assistance of Dr. Jim Lenahan of the Hydrology, special issue on HAPEX-Sahel no.188/189: 385–399.
International Water Management Institute at Goutorbe, J.P. et al., 1994. HAPEX Sahel: A large scale study of
Colombo (Sri Lanka) is respectfully appreciated. land-atmosphere interactions in the semi-arid tropics. Annales
Geophysicae 12, 53–64.
Gurney, R.J. and Hall, D.J., 1983. Satellite-derived Surface Energy
Balance Estimates in the Alaskan Sub-Arctic. J. of Climate and
References Applied Met., 22, 115–125.
Hurk, B.J.J.M., Bastiaanssen, W.G.M., Pelgrum, H., van Meijgaard,
Abdel Gawad, S.T., Abdel Khalek, M.A.,Boels, D., El Quosy, D.E., E., 1997. A new methodology for assimilation of initial soil
Roest, C.W.J., Rijtema P.E., Smit, M.F.R., 1991. Analysis of moisture fields in weather prediction models using meteosat
water management in the Eastern Nile Delta. Reuse Report 30. and NOAA data. J. of Applied Met. 36, 1,271–1,283.
DLO Winand Staring Center, Wageningen, The Netherlands. Itier, B., Brunet, Y., McAnneney, K.J., Lagouarde, J.P., 1994.
245 pp. Downwind evolution of scalar fluxes and surface resistance
Bastiaanssen, W.G.M., Pelgrum, H., Droogers, P., de Bruin, H.A.R., under conditions of local advection. Part 1: A reappraisal of
M. Menenti, 1997. Area-average estimates of evaporation, boundary conditions, Agr. and Forest met. 71, 211–255.
wetness indicators and top soil moisture during two golden Jarvis, P.G., 1976. The interpretation of the variation in leaf water
days in EFEDA. Agr. and Forest Met. 87, 119–137. potential and stomatal conduct ance found in canopies in the
Bastiaanssen, W.G.M., Roest, C.W.J. Pelgrum, H., Abdel Khalek, field. Phil. Trans. R. Soc. London, Ser. B 273, 593–610.
M.A., 1992. Monitoring of the irrigation performance on the Jochum, A.M., Michels, B.I., Entstraser, N., 1993. Regional and
basis on actual evapotranspiration: Comparison of satellite local variations of evaporation fluxes during EFEDA. Conf. of
data and simulation model results. In: J. Feyen, E. Mwendera Hydroclimatology, 17–22 Jan. 1993, Anaheim. 4 pp.
 W.G.M. Bastiaanssen et al. / Journal of Hydrology 212–213 (1998) 213–229 229

Jochum, A.M., 1993. Evaporation and energy fluxes during for large area semi-arid land application with limited climate
EFEDA: horizontal variability are area averaging. In: H.J. data. J. of Hydrology 121, 119–132.
Bolle, R.A. Feddes and J. Kalma (eds). Exchange processes at Pelgrum, H.,1992. Mapping areal surface energy balances during
the land surface for a range of space and time scales, IAHS Publ. daytime using Meteosat data. Internal Note 195. DLO Winand
212, 373–380. Staring Centre, Wageningen, The Netherlands. 50 pp.
Kabat, P., Prince, S.D., Prihodko, L., 1997. Hydrologic Atmos- Pelgrum, H., Bastiaanssen, W.G.M., 1996. An intercomparison of
pheric Pilot Experiment in the Sahel (HAPEX Sahel), Methods, techniques to determine the area-averaged latent heat flux from
measurements and selected results from the West Central Super- individual in situ observations: a remote sensing approach using
site, Report 130, DLO Winand Staring Centre, Wageningen, EFEDA data. Water Res. Res. 32(9), 2,775–2,786.
The Netherlands, 215–221. Schuepp, P.H., Lecler, M.Y., Mackerson, J.I., Desjardins, R.I.,
Kondo, J, Saigusa, N., Sato, T., 1990. A parameterization of 1990. Footprint prediction of scalar fluxes from analytical solu-
evaporation from bare soil surfaces. J. Applied Met. 29: 385–389. tions to the diffusion equation. Boundary Layer Met. 50, 355–373.
Kustas, W.P., Moran, M.S., Jackson, R.D., Gay, L.W., Duell, Sellers, P.J., Hall, F.G., 1992. FIFE in 1992: Results, scientific gains
L.F.W., Kunkel, K.E., Matthais, A.D., 1990.Instantaneous and and future research directions, Journal of Geophysical Research
daily values of the surface energy balance over agricultural 97(D17): 19,091-19,109.
fields using remote sensing and reference field in an arid envir- Shuttleworth, W.J., R.J. Gurney, A.Y. Hsu and J.P. Ormsby, 1989.
onment. Rem. Sens. Env. 32, 125–141. FIFE: the variation in energy partitioning at surface flux sites.
Ma, Y., 1996. Regionalization of surface flux densities andseasonal IAHS Red Book Series no. 186: 67-74.
variation in the heterogeneous landscape of HEIFE. Internal Smith, R.C.G. and B.J. Choudhury, 1990. Relationship of multi-
Note 422. DLO Winand Staring Centre, Wageningen, The spectral satellite data to land surface evaporation from the
Netherlands. 24 pp. Australian continent. Int. J. of Remote Sensing 11-11, 2069–
Menenti, M., Bastiaanssen, W.G.M., Hefny, K., Abd El Karim, 2088.
M.H., 1991. Mapping of groundwater losses by evaporation in Taconet, O., Bernard, R. Vidal-Madjar, D., 1986. Evapotranspira-
the Western Desert of Egypt. Report 43. DLO Winand Staring tion over an agricultural region using a surface flux/temperature
Centre, Wageningen, The Netherlands. 116 pp. modelbased on NOAA-AVHRR 5 data. J. of Clim. and Applied
Mitsuta, Y.,1994 (ed.). Proc.Int. Symp. on HEIFE. Disaster Met. 25, 284–307.
Prevention Research Institute, Kyoto University, Kyoto, Wang, J., Ma, Y., Menenti, M, Bastiaanssen, W.G.M., Mitsuta, Y.,
Japan. 722 pp. 1995. The scaling-up of processes in the heterogeneous land-
Ottle, C., Vidal-Majar, D.D., Girard, R., 1989. Remote sensing scape of HEIFE with the aid of satellite remote sensing. J. Met.
applications to hydrological modeling. J. of Hydrology 105, Soc. of Japan 73(6), 1234–1244.
369–384. Xue, Y., Shukla, J., 1990. The influence of land surface properties
Owe, M., van de Griend, A.A., 1990. Daily surface moisture model on Sahel climate. J. Climate 6, 2232–2245.