Combined use of NDVI Time Courses at Low and High Spatial Resolution to Estimate Land Cover and Crop Evapotranspiration in Semi-arid Areas

The general objective of this study is to establish the water balance over the semi-arid plain of Marrakech/Al Haouz, a 2000 km² intensively cropped area in center of Morocco. This requires firstly to map the land cover, secondly to monitor the vegetation dynamics and thirdly to set up evapotranspiration models. In this objective, we used NDVI time series derived from a 6-year archive of MODIS data at 250 m resolution (2000-2006). We applied a linear unmixing model to predict land cover fraction of the dominant land cover classes (orchard, bare soil and annual crop). Two methods for extracting endmembers (typical signatures of land classes) are evaluated; the first is supervised and uses prior information derived from high resolution data, while the second extracts endmembers directly from MODIS data. In both cases, the typical signatures are realistic and the two methods give comparable results, with errors around 0.12 in land cover fractions. The second method provides annual land cover maps during the 2000-2006 period along with the vegetation dynamics (typical signatures of NDVI time courses). This information is used to estimate the crop evapotranspiration using the FAO-56 algorithm, with cultural crops coefficients (Kc) adjusted for each land cover class. Finally, map of annual evapotranspiration are obtained, which are not yet fully validated. This study confirms the potential of using low spatial resolution data to manage water resources at regional scale.