Application of Extreme Value Theory in Predicting Floods in Region 3, Zimbabwe

Abstract

The use of Extreme Value Theory to predict the possibility of floods has advanced our understanding of the occurrence of this rare phenomenon. The main goal of this research is to use the Block Maxima technique of the Extreme Value Theory on annual maximum rainfall to predict the occurrence of floods by estimating the probability and intensity of future flooding events. The Extreme Value Theory is more suitable for predicting floods because it focuses on extreme events, providing better accuracy for rare and high-magnitude occurrences. Unlike traditional methods, the Extreme Value Theory handles outliers effectively and models tail behaviour which is crucial for understanding flood risks. It also adapts well to non-stationary conditions like climate change, while traditional methods assume past conditions will repeat. The annual maximum rainfall data used in this research came from the Meteorological Services Department of Zimbabwe in Harare, Zimbabwe. It covered five "convenience sampled" provinces in the purposefully sampled natural Region 3, Zimbabwe. The regions include the Midlands, Mashonaland East, Manicaland, Masvingo and Matabeleland South. We applied the Maximum Likelihood Estimation Method on the annual maximum rainfall data to estimate the Generalised Extreme Value parameters which were used to estimate the return levels and return period. The return level plots and chi-square distribution were used to assess the model fit. The results showed that the return level for a 200-year flood event is 1685.13mm, with a return period of 200years.The forecast results showed that the magnitude of annual maximum rainfall increases with time, suggesting a high risk of flooding. Hence, early warnings and preparedness, risk assessment and management can help control or lessen flooding in the future while saving the environment and providing humanitarian aid in times of need.