Impact of Spatial Scale on Remotely Sensed Evapotranspiration Estimates from Heterogeneous Land Surfaces

Remote sensing can provide estimates of spatially distributed actual evapotranspiration (ET) at different spatial scales. Methods that combine remotely sensed observations with other ground-based information on land and atmospheric properties usually allow improved estimates of spatially distributed evapotranspiration (ET) than when remotely sensed data are used alone. Spatial resolution of remotely sensed ET estimates depends on the specifications of the satellite instruments (e. g., 60 m for Landsat, 1 km for MODIS and 10 km for AVHRR). Since satellite sensors have different spatial, spectral and radiometric resolutions, it is essential to understand the consistency of ET estimates from different sensors. The main objective of this paper is to understand the spatial scaling effects of remotely sensed ET estimates from coarse resolution AVHRR data, through the analysis of MODIS data in a highly heterogeneous catchment. This study was undertaken in the Musi catchment in the Krishna River Basin, India. The average land parcel size in this catchment is 0.5 ha with highly variable crops and land management practices. In this study, the Surface Energy Balance Algorithm for Land (SEBAL) was used to estimate spatially distributed evapotranspiration by combining ground-based metrological data and remotely sensed data from MODIS sensor. Observed pan evaporation data from weather station were used for validation of the 250m resolution MODIS ET output, before up scaling to course resolution products. In order to quantify the difference in ET estimates originating solely from the observation scale, ET estimates were aggregated to the AVHRR scale using two different approaches. First, high-resolution ET was estimated at the original pixel resolution of MODIS and then aggregated to the coarser resolution scale (output up scaling). Second, ET estimated directly at the coarser resolution by aggregating the fine resolution input data to the coarse resolution scale (input up-scaling). The relationship between the sub-pixel-scale heterogeneity and scaling effects on ET estimate is investigated and other factors affecting the observed discrepancies between ET estimates from MODIS and AVHRR are discussed. Overall a 0.12mm difference observed in daily ET between the two up scaled process at catchment scale analysis.