Applying Stationary and Nonstationary Generalized Extreme Value Distributions in Modeling Annual Extreme Temperature Patterns

Abstract

This study applies both stationary and nonstationary generalized extreme value (GEV) models to analyze annual extreme temperature patterns in four stations of Southern Highlands region of Tanzania: Iringa, Mbeya, Rukwa, and Ruvuma over a 30-year period. Parameter estimates reveal varied distribution characteristics, with the location parameter μ ranging from 28.98 to 33.44, and shape parameter ξ indicating both bounded and heavy-tailed distributions. These results highlight the potential for extreme temperature conditions, such as heatwaves and droughts, particularly in regions with heavy-tailed distributions. Return level estimates show increasing temperature extremes, with 100-year return levels reaching 33.95 °C in Ruvuma. Nonstationary models that incorporate time-varying location and scale parameters significantly improve model fit, particularly in Mbeya, where such a model outperforms the stationary model (p value = 0.0092). Trend analyses identify significant temperature trends in Mbeya (p value = 0.0123) and Ruvuma (p value = 0.0015), emphasizing the need for adaptive climate strategies. These findings underscore the importance of accounting for nonstationarity in climate models to better understand and predict temperature extremes.