Validation of SEBAL
Accuracy assessments of SEBAL-based ET fluxes under various climatic conditions
The accuracy of the aerial patterns of actual evapotranspiration determined from remote sensing can be assessed from independently collected ground measurements. Several field methods exist to measure the evaporative fluxes and the partitioning of available radiant energy into sensible and latent heat fluxes. All these methods have their own scale (area influencing the measurement signal) and accuracies. The Surface Energy Balance Algorithm for Land (SEBAL) has been tested against the methods at the places described in table 1.
Table 1: Validation of SEBAL-based ET fluxes at field scale using different measurement techniques. Eddy-correlation measurements are worked out separately in table 2
Field instrument | Country | Location | Landscape | No. of days compared |
---|---|---|---|---|
Weighing Lysimeter1 | Idaho-USA | Kimberly, 1987 | Irrigated crops | 4 |
Scintillometer2 | Turkey | Gediz basin, 1998 | Irrigated crops | 2 |
Scintillometer3 | Sri Lanka | Horana, 1999 | Palm trees and rice | 25 |
Bowen ratio4 | Egypt | Qattara Depression, 1986 | Desert | 3 |
Bowen ratio5 | Spain | Tomelloso, 1991 | Rainfed crops | 4 |
Bowen ratio6 | Brazil | Petrolina, 1998 | Irrigated bananas | 1 |
Bowen ratio7 | Botswana | Serowe, 1997 | Dry pastures | 4 |
Bowen ratio8 | Pakistan | Rechna Doab, 2000 | Irrigated crops | 20 |
Bowen ratio9 | Kenya | Naivasha, 1998 | Savannah | 4 |
Eddy-correlation | Several | see table 2 | see table 2 | 30 |
There is a general consensus among meteorologists that eddy-correlation systems provide the most reliable means to measure evaporation fluxes in situ. The eddy-correlation method uses high-frequency measurements of vapour density, air temperature and vertical wind speed to compute the sensible and latent heat fluxes directly without involvement of other parameters. Eddy-correlation data is also representative, because the footprint of fluxes derived from eddy-correlation data covers usually a distance of approximately 1km in the upwind direction (the fetch varies with height, wind speed, surface roughness and observation height). The source area is therefore approximately 1km2. An accuracy of 10 to 20% in the measurements of individual sensible and latent heat fluxes at a confidence interval of 95% is generally assumed for measurements made over a uniform surface with an adequate fetch. SEBAL has been validated against eddy-correlation instruments as specified in table 2.
Table 2: Data from eddy-correlation experiments to validate SEBAL-based ET fluxes
Country | Landscape | Climate | Source |
Spain (1991) | Rainfed and irrigated crops | Semi-arid | Bastiaanssen et al., 1998 10 |
China (1991) | Desert and oases | Arid | Wang et al. 1995 11 |
Niger (1992) | Savannah | Warm and humid | Roerink, 1995 12 |
Netherlands (1995) | Forest and pastures | Cold and humid | Soeterik, 1995 13 |
Italy (1997) | Rainfed and irrigated crops | Semi-arid | Su and Menenti, 1999 14 |
New Mexico (2000) | Cottonwood | Semi-arid | Conrood and McDonnel, 2001 15 |
Most of the accessible eddy-correlation data contained information on the daytime evaporative fraction. Evaporative fraction is defined as latent heat flux/(latent + sensible heat flux). The evaporative fraction lies essentially between zero and one - if advective conditions don't prevail. This helps in understanding the soil wetness conditions of the landscape (EF ~ 0 in oven dry soils and EF ~ 1 in saturated soils). The uncertainty of evaporative fraction interpreted from the eddy-correlation data is with 25% less than for the accuracy of the individual heat fluxes (10 to 15 %).
Table 3: Deviations of evaporative fraction (EF) derived from remote sensing and from in situ measurements for single day events at field scale (~ 1km2)
Country | n | University or Research Centre | RMSE-EF absolute | Inside confidence band (%) |
Spain, 1991 | 6 | Karlsruhe, Berlin, Copenhagen | 0.07 | 67 |
China, 1991 | 2 | Lanzhou Institute of Plateau and Atmospheric Physics | 0.03 | 100 |
Niger, 1992 | 3 | Copenhagen, DLO-Winand Staring Centre | 0.14 | 100 |
The Netherlands, 1995 | 11 | KNMI, DLO-Winand Staring Centre | 0.12 | 70 |
Italy, 1997 | 2 | Basel | 0.06 | 100 |
New Mexico, 2001 | 6 | New Mexico | 0.18 | 67 |
The average Root-Mean-Square-Error (RMSE) of the absolute evaporative fractions for a single day event is 0.1. The percentage of data points located inside the confidence band is 84%, which implies than on average 16% of the SEBAL evaporative fraction assessments lie outside the confidence envelope. Hence, the overall error of evaporative fraction from
SEBAL for single day events and for scales in the order of 1km2 is 16%.
Figure 1 shows that there is no bias in the evaporative fraction data towards the drier or wetter side of the fields used in the validation study; the performance of the lower and higher regions of evaporative fraction are not significantly different. This may be regarded as a non-sensitivity factor of SEBAL to soil wetness conditions. Occasions with negative and positive deviations are present, although the largest deviation occurs when the deviation (measured-estimated) is positive, which suggests that SEBAL more often underestimates the energy partitioning between sensible and latent heat fluxes, than it overestimates. The data points of figure 1 have a correlation of R2=0.80 (n=30)
Figure 2 demonstrates that in 50% of the cases (the average situation), the deviation is small (DEF=0.07). The error at one time the standard deviation (68% of the cases) is DEF=0.12, and it increases to DEF=0.30 at two times the standard deviation (95% of the cases). Figure 1 has shown that deviations are not necessarily errors and that only 16% of
the cases should be marked as erroneous and that 84% falls within the error bounds. The resulting errors are therefore small (see table 4)
Table 4: Errors at different levels of probability for single day events at field scale (~1km2)
Probability | Standard Deviation | Deviation in evaporative fraction | Error in evaporative fraction |
50% | 0s | 0.07 | 0.011 |
68% | 1s | 0.12 | 0.019 |
95% | 2s | 0.30 | 0.048 |
99% | 3s | 0.33 | 0.053 |
Figure 1: Relationship between evaporative fraction estimated from satellites (EF-SEBAL) and measured with eddy-correlation devices (EF eddy-correlation) for single day events. The 25% uncertainty in measurement error at a confidence interval of 95% (2s) is plotted also.
Figure 2: Distribution of deviation between SEBAL and eddy-correlation for single day events.
Literature
1 Allen, R.G., W.G.M. Bastiaanssen, M. Tasumi and A. Morse, 2001 Evapotranspiration on the watershed scale using the SEBAL model and Landsat images, ASAE Meeting Presentation, paper number 01-2224, Sacramento, California, USA, July 30-August 1, 2001
2 Meijninger, W.M.L. and H.A.R. de Bruin, 2000 The sensible heat fluxes over irrigated areas in western Turkey determined with a large aperture scintillometer, Journal of Hydrology vol. 229: 42-49
3 Hemakumara, H.M., L. Chandrapala and A. Moene, 2001 Evapotranspiration fluxes over mixed vegetation areas measured from large aperture scintillometer, Agricultural Water Management (in press)
4 Bastiaanssen, W.G.M. and M. Menenti, 1989 Mapping groundwater losses in the Western Desert of Egypt with satellite measurements of surface reflectance and surface temperature, in (ed.) J.C. Hooghart, Water Management and Remote Sensing, TNO Committee on Hydrological Research, proceedings and information no. 42: 61-89
5 Pelgrum, H. and W.G.M. Bastiaanssen, 1996 An intercomparison of techniques to determine the area-averaged latent heat flux from individual in situ observations: a remote sensing approach using EFEDA data. Water Resources Research vol. 32(9): 2775-2786
6 Castro, A.H., P.V. de Azvedo, B.B. da Silva and J.M. Soares, 1999 Water consumption and crop coefficient of grape vine in the region of Petrolina, Pernambuco State, Brazil, Revista Brasileira de Engenharia Agricola e Ambiental 3(3): 413-416 (in Portugese)
7 Timmermans, W.J. and A.J.M. Meijerink, 2001 Satellite derived actual evapotranspiration and groundwater modelling, Botswana, in (eds.) M. Owe, K. Brubaker, J. Ritchie and A. Rango, Remote Sensing and Hydrology 2000, IAHS Red Book publication no. 267
8 Bastiaanssen, W.G.M., M. Ud-din-Ahmed and Y. Chemin, 2001 Satellite surveillance of water use across the Indus Basin, Water Resources Research (accepted)
9 Farah, H.O., 2001 Estimation of regional evaporation under different weather conditions under different weather conditions from satellite and meteorological data, Ph.D. Thesis, Wageningen University, Department of Water Resources: 170 pp.
10 Bastiaanssen, W.G.M., H. Pelgrum, J. Wang, Y. Ma, J. Moreno, G.J. Roerink and T. van der Wal, 1998
The Surface Energy Balance Algorithm for Land (SEBAL): Part 2 validation, Journal of Hydrology vol. 212-213: 213-229
11 Wang, J., Y. Ma, M. Menenti, W.G.M. Bastiaanssen and Y. Mitsuta, 1995 The scaling up of land surface processes over a heterogeneous landscape with satellite observations, J. of Met. Soc. of Japan, vol. 73(6): 1235-1244
12 Roerink, G.J., 1995 SEBAL estimates of the areal patterns of sensible and latent heat fluxes over the HAPEX-Sahel grid, a case study on 18 September 1992, DLO-Staring Centre Interne Mededeling 364, Alterra, Wageningen: 81 pp.
13 Soeterink, K.J., 1995
Estimating of seasonal evaporation and water stress in the Netherlands using Landsat images, M.Sc. Thesis, Technical University of Delft, Section Land and Water Management: 52 pp + annexes
14 Su, Z. and M. Menenti, 1999
Mesoscale climate hydrology: the contribution of the new observing systems, pre-execution phase, final report, BCRS USP-2, 99-05, Netherlands Remote Sensing Board, Delft, The Netherlands: 141 pp.
15 Conrood and McDonnel, 2001