Estimation of Actual Evapotranspiration in Fragmented Mediterranean Areas by the Spatio-Temporal Fusion of NDVI Data

Abstract

Actual evapotranspiration $(\text{ET}_{{\rm{A}}})$ is a fundamental component of the land water cycle that can be predicted by the combination of meteorological data and remotely sensed normalized difference vegetation index (NDVI) observations. The proficient application of this approach to the retrospective study of fragmented areas, however, depends on the preliminary use of spatio-temporal fusion (STF) methods capable of integrating different satellite datasets. One of these methods is the Spatial Enhancer of Vegetation Index image Series (SEVIS), which has been recently developed to improve the annual NDVI datasets based on one or a few high spatial resolution images. This STF method is currently applied to moderate resolution imaging spectroradiometer (MODIS) and TM/ETM+/OLI imagery taken over three fragmented areas in Tuscany (Central Italy), representative of different Mediterranean ecosystems, i.e., an urban grassland, a tomato field, and an olive grove. The performance of SEVIS is evaluated by comparing the $\text{ET}_{{\rm{A}}}$ estimates obtained from the original (MODIS) and synthetic (MODIS plus TM/ETM+/OLI) NDVI datasets to ground $\text{ET}_{{\rm{A}}}$ observations. The experimental results indicate that the original MODIS NDVI data cannot properly characterize the seasonal vegetation evolutions of the three study sites, which negatively affects the performance of $\text{ET}_{{\rm{A}}}$ simulation. In contrast, such evolutions are reasonably reproduced by the synthetic NDVI datasets, which improves the accuracy of the $\text{ET}_{{\rm{A}}}$ estimates both in terms of correlation and errors. The improvements are particularly evident during the summer dry period when the MODIS images are incapable of characterizing the actual vegetation response to water stress.