Deciphering the Variations in the Generalized Extreme Value Distribution Parameters in the Non-stationary Flood Frequency Analysis

A changing climate intensifies extreme precipitation events, increasing the likelihood of flooding. The floods resulting from climatic changes are often modelled by non-stationary flood frequency analysis by introducing covariates in the location and scale parameters of the Generalised Extreme Value distribution keeping the shape parameter a constant. This study focuses on conducting non-stationary flood frequency analysis using annual maximum discharge data from four stations in the Muvattupuzha river basin, Kerala, India, incorporating the shape parameter variation. Non-stationary flood frequency curves were developed by introducing indices that account for the climate change and variability. Two stations of the basin portrayed the non-stationary model with shape parameter quadratically varying with El Niño Southern Oscillation as the best-fit model. The study also evaluates the effectiveness of incorporating climate indices into the shape parameter of the distribution function within a non-stationary framework. Findings reveal that non-stationary models incorporating shape parameter variations outperformed the stationary model by over 29% and the non-stationary model with a constant shape parameter by over 26% for the 100-year return period. Uncertainty in flood quantile estimates was assessed using confidence intervals generated via the parametric bootstrap method. The study highlights that assuming a stationary climate or a constant shape parameter can result in underestimating extreme floods, increasing the risk of flooding and infrastructure failures. The findings highlight the necessity of incorporating climate-informed shape parameter variations into non-stationary flood frequency analysis to enhance flood mitigation and adaptation strategies.