Effect of pixel scale on evapotranspiration estimation by remote sensing over oasis areas in north-western China

One of the major uncertainties in the remote sensing estimates of regional evapotranspiration (ET) over heterogeneous landscapes is that significant estimation differences between different data scales always exist. It is necessary to predetermine the optimal remote sensing scale under such conditions. In this study, the effect of pixel scale on ET estimation over a typical oasis in north-western China was investigated. By the area-averaging and wavelet multi-resolution analysis aggregation techniques, four “input” up-scaling and “output” up-scaling procedures were implemented, respectively, to determine the optimal data scale. Validation suggested that high-resolution Landsat-based ET estimates could be used as the ground reference data when in situ measurements were not available. The differences between Landsat-based and MODIS-based ET estimates (i.e. the so-called Regional Error Distribution of the latter) showed that 1-km resolution MODIS data overestimated ET over the landscapes where NDVI was less than or equal to ~0.10; the data underestimated ET over the landscapes where NDVI was greater than or equal to ~0.40; ET could only be relatively accurately predicted by the data on the surfaces where NDVI ranged from 0.10 to 0.40. However, MODIS-based 1-km ET estimates could effectively reveal the predominant spatial distribution trend of regional ET. The results confirmed that the optimal remote sensing scale for modeling ET was approximately 480 m over the study area; land surface heterogeneity could cause significant errors in ET estimation once data scales exceeded this threshold; in addition, MODIS data at 250-m resolution were adequate for correcting the 1-km resolution ET estimates over heterogeneous landscapes.