Modelling Extreme Rainfall Patterns in Peninsular Malaysia Using Spatial Extreme Value Approaches

Abstract

Extreme rainfall poses significant challenges to communities, infrastructure, and ecosystems, particularly in tropical regions like Peninsular Malaysia. Recent decades have seen increasing rainfall variability and intensity due to climate change and monsoonal influences. This study models extreme rainfall using the generalized extreme value (GEV) distribution and extends the analysis spatially through max-stable processes (MSP), specifically the Smith and Schlather models. Using block maxima GEV marginals and MSP spatial models Smith and Schlather, this study quantifies spatial dependence in annual maxima from 11 stations (19902023) selected for data completeness and geographic coverage (coastal/inland). Missing data were imputed using inversedistance weighting (IDW). The return levels for 5-, 10-, 20-, and 30-year periods are assessed. The MSP approach enables modeling of spatial extremes by transforming return levels into Frchet margins. Parameters are estimated using maximum pairwise likelihood estimation (MPLE), and model performance is evaluated using the Takeuchi information criterion (TIC) and root mean square error (RMSE). Allowing a timevarying range parameter λ(t) within the Schlather model improves returnlevel estimates relative to stationary alternatives. These results support floodresilient design such as drainage capacity, dam safety and zoning. This study acknowledge nonstationarity due to climate change as a key extension and note that the framework can incorporate climate covariates.