Estimating the spatial and model uncertainties in yielding extreme rainfall return levels across India

Continental parts of India. Extreme rainfall events, shaped by anthropogenic climate change and natural variability, pose a threat to ecosystems and infrastructure. Though Generalized Extreme Value distributions are standard to model extreme rainfall, spatial and model uncertainties remain inadequately explored. This study investigates whether an optimal Non-stationary Generalized Extreme Value (NSGEV) model for extreme rainfall varies regionally and how external covariates affect model structural and spatial uncertainties in the estimates. Using several NSGEV models to Annual Maximum Rainfall (AMR), the study associates distribution parameters with Time and global climate drivers. The study found that although AMR across India exhibits substantial non-stationarity, statistically significant trends are limited. Non-linear models with Time and Pacific Decadal Oscillation and a hybrid model incorporating lagged climate drivers for the location parameter best explain AMR temporal variance. Rare rainfall events once expected every 71 years now may occur within a decade. Uncertainty analysis indicates higher spatial uncertainty in extreme rainfall estimates over northeastern, central, and coastal regions. While the median spatial variance remains stable in some zones as return periods increase, outliers suggest a strong localized influence from climate drivers. Model uncertainty is minimal for low quantiles and arid zones but is higher in tropical and temperate zones, highlighting the importance of model selection in areas with high spatial rainfall variability. • 64 NSGEV models show return periods are shorter than those from stationary models. • We quantify uncertainty from stationary to hybrid non-stationary AMR models. • Study shows high spatial uncertainty in return levels in the same climate zone.