NDVI time series and Markov chains to model the change of fuzzy vegetative drought classes

Abstract

The objective of this study is to explore the potential of using Markov chains to model the changes of vegetative drought classes. NOAA-AVHRR dekadal NDVI images and fuzzy functions are used to characterize the drought classes while capturing the gradual transition between them. The transition probabilities are estimated using the maximum class membership values at a location. The Markov transition probability matrix is then used to model the changes of vegetative drought classes at selected locations. Future vegetative drought classes are predicted using the estimated transition matrix, then compared with actual data. Twenty pixel locations clustered in four regions of the two main agricultural type in Kenya are selected to implement this approach. Half of the pixels are predicted correctly. 5 of them are predicted either one class higher or lower and 2 of them, two classes higher. We can conclude that Markov chains applied to fuzzy numbers have the potential to model the changes of of vegetative drought classes at a pixel, hence provide a benefit for early warning systems.