A COMPARISON OF STATIONARY AND NON-STATIONARY GENERALIZED EXTREME VALUE MODELS WITH CLIMATIC COVARIATES IN MODELING RAINFALL EXTREMES

Extreme rainfall in Peninsular Malaysia is closely linked to seasonality, primarily driven by the southwest monsoon (SWM) period, starting from May to September and northeast monsoon (NEM) period, starting from November to March. Extreme rainfall events may lead to secondary disasters such as floods, landslides and crop damage. The west part of Malaysia also known as Peninsular Malaysia is significantly influenced by large-scale global climate phenomena such as El Niño-Southern Oscillation (ENSO) that could affect the rainfall pattern across this region. Understanding the changing behavior of extreme rainfall and its relationship with ENSO will increase planners’ ability to plan for, manage and respond to related flood events. This study investigates the trend and stationarity of extreme rainfall in Peninsular Malaysia by using the Mann-Kendall (MK) trend test and Augmented Dickey-Fuller (ADF) test. Rising trends in extreme rainfall were identified in most part of Peninsular Malaysia. This study also analyzes the suitability of stationary and non-stationary Generalized Extreme Value (GEV) models for modeling rainfall extremes. The non-stationary model integrates the Southern Oscillation Index (SOI) and a linear trend as covariates to capture potential climatic influences on extreme rainfall events. The model performance is assessed using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), enabling quantitative comparison. To assess the uncertainty of each model’s parameter, bootstrap method will be applied in this study. Likelihood ratio tests are employed to evaluate the robustness and significance of the models, ensuring the best model is selected. Preliminary results suggest that incorporating SOI and trend improves the model's ability to explain variability in rainfall extremes, offering insights into climatic drivers of extreme events. Extreme rainfall return levels are used to quantify potential flooding risk. This research has practical implications for understanding and predicting rainfall extremes under changing climate conditions.