Use of multitemporal satellite data for vegetation change detection in Namibia

As water supply strongly influences plant growth in semiarid areas, precipitation events in their spatio-temporal variability are reflected in NDVI time series (AVHRR, MODIS) of individual rainy seasons and complicate a phenological delimitation of vegetation units. To overcome this problem and to develop a "near-to-real-time"-vegetation monitoring, relationships between NDVI and previous rainfall events are investigated, as it is assumed, that the different vegetation units (e.g. grass savanna, tree savanna) show specific reactions on water availability. Following a decision tree approach, homogenous rainfall-vegetation classes are separated from seasonal AVHRR-NDVI data and correlations to rainfall data including recent and past multi-day-sums are computed for each class. A supervised classification and a change map from two sets of Landsat data are used for final validation and class labeling.