Predicting Dynamics of Vegetative Drought Classes Using Fuzzy Markov Chains

Disclaimer This document describes work undertaken as part of a programme of study at the Faculty of Geo-Information Science and Earth Observation of the University of Twente. All views and opinions expressed therein remain the sole responsibility of the author, and do not necessarily represent those of the Faculty. ABSTRACT Drought is a naturally occurring event, causing temporary imbalance water availability and vegetation damage. It exists when the amount of received precipitation has been significantly below normal recorded levels. To reduce the devastating effects of drought and minimize the losses, early warning system can help decision and policy makers to implement policies timely. Satellite-based Normalized Difference Vegetation Index (NDVI) data are consistently available and continuous in space and time, applied to prediction drought in this research. As drought has many characteristics, varying region by region and may last for several months or even years, it represents a challenge to fully evaluate the characteristics for its prediction. In this research, the main objective is to predict the changes of vegetative drought classes, also called states. This is done by modelling these changes using Markov chains applied to predefined fuzzy vegetative drought states. Four regions of different agricultural patterns in Kenya are selected as study areas to apply this method. There is a strong relationship between NDVI and accumulated almost three-month precipitation data. The highest correlation value R can be larger than 0.9. It indicates that NDVI can be an indicator for vegetative drought prediction. The every dekadal NDVI data are acquired from FEWS NET from 2004 to 2008. Fuzzy membership functions are applied in this research as a description of drought classes. The vegetative drought classes are classified by fuzzy classification. Under the Markovian property tests, the NDVI anomaly data can be modelled in first-order Markov chain, but the time homogeneity is interrupted by the data in February and September. The validation data is the comparison of prediction result and pre-existing reference data. Half of the pixels in study area are well predicted by fuzzy Markov chain. And also, the changing of fuzzy membership function influences the result of prediction. In conclusion, the Markov chain with fuzzy membership function has the potential to be applied in vegetative drought prediction and provide benefit for early warning system. ii ACKNOWLEDGEMENTS I would like to express my earnest gratitude and thankfulness to my supervisors, Prof. Dr. Ir. A. Stein and Dr. Ir. Wietske Bijker …