Estimate Evapotranspiration Using Modis Data with a Modified Triangle Method for a Semi-arid Watershed

Three major types of ET estimation methods are commonly used on the basis of remote sensed data, that is, statistical analysis method, energy balance residual method, and evaporation fraction (EF) method. Evaporation-ratio method estimates ET by calculating the ratio between latent heat flux and surface available energy (i.e., net solar radiation Rn - soil surface heat flux G) so as to avoid the calculation of sensible heat flux which has high uncertainty due to its sensitivity to the estimation of air resistance. The critical part with the EF method is the calculation of EF, for which the triangle method is most widely used. The triangle method is developed based on vegetation index ~ surface temperature (VI~Ts) feature space. The triangle method is comparatively easy to apply and based on mostly remotely sensed data, but it does not effectively distinguish soil evaporation from vegetation transpiration, and it assumes that for fully vegetated area actual ET is equal to potential ET, without considering the effects of vegetation water stress on the transpiration of plants, which result in significant overestimation of actual ET. A more recently developed EF method is the MOD16 algorithm (Nishida, 2003) which considers EF as a mixture of EF for bare soil (EFsoil) and EF for fully vegetated area (EFveg) according to the vegetation coverage (fveg), so that the contribution of soil and vegetation to total ET are explicitly separated. However, while theoretically MOD16 algorithm takes into account more physical processes, more uncertainties are introduced due to much more complicated calculations. In addition, the MOD16 algorithm is calculated based on NDVI time series whereas many studies have shown that Enhanced Vegetation Index (EVI) is a better alternative to NDVI. In the present study, we estimated ET based on Moderate Resolution Imaging Spectroradiometer (MODIS) EVI and land surface temperature data with a modified triangle method which has the same framework as the triangle method, but explicitly includes the effects of vegetation water stress as in the MOD16 algorithm. The modified triangle method was applied to the Walnut Gulch Experimental Watershed, Arizona, United States, where flux observation data are freely available, and the ET estimates as well as middle results (i.e., Soil heat flux G, Solar Radiation Rs and Net Radiation Rn) are compared with the observations. The results show that, the estimates of G, rs and rn are fairly good compared with the observations, whereas the estimates for ET are acceptable although less good than G, Rs and Rn. In addition, our study results highlight the necessity of MODIS data pre-processing, including the de-noising, de-stripping and land surface temperature correction. Besides the modified triangle method, the conventional triangle method, MOD16 algorithm, and the FAO-50 algorithm are also applied to the Walnut Gulch Watershed. The result shows that the estimates given by the modified triangle method are most consistent to the observations.