Spatial modeling of drought events using max-stable processes

With their severe environmental and socioeconomic impact, drought events belong to the most far-reaching natural disasters. Effects are tremendous in rain-fed agricultural areas as in Africa. We analyzed and modeled the spatio-temporal statistical behavior of the Normalized Difference Vegetation Index as a risk indicator for drought, reflecting its stochastic effects on vegetation. The study used a data set for Rwanda obtained from multitemporal satellite remote sensor measurements during a 14-year period and divided into season-specific spatial random fields. Maximal deviations from average conditions were modeled with max-stable Brown-Resnick processes taking methodological and computational challenges into account. Those challenges are caused by the large spatial extent and the relatively short time span covered by the data. Extensive simulations enabled us to go beyond the observations and, thus, to estimate several important characteristics of extreme drought events, such as their expected return period.