Estimating Climate-Driven Non-Stationary Flood Return Periods Using the GAMLSS model and EWT approach

Hydrological regimes are increasingly altered by the combined influences of climate variability, climate change, and anthropogenic interventions, challenging the traditional assumption of stationarity in flood frequency analysis (FFA). For sustainable water resources planning and flood risk management, it is crucial to consider non-stationary nature of flood behaviour. In this study, we examine the spatiotemporal aspects of flood pattern in the Upper Narmada Basin, utilizing a non-stationary flood frequency framework that incorporates the non- stationarity behaviour due to climate variability, climate change and reservoir influence. We develop single-covariate (SC) and multi-covariate (MC) non-stationary models based on Generalized Additive Models for Location, Scale, and Shape (GAMLSS), incorporating climate indices, reservoir metrics, and time as predictors. This study prioritizes the estimation of non-stationary return periods in scenarios driven by climate variability, applying the Expected Waiting Time (EWT) method. The findings show significant non-stationarity in flood return period caused by both climate variability and reservoir influence. This leads in significant variations from return levels calculated under stationary assumptions. The results underscore the risk of underestimating or overestimating flood risks when depending on conventional stationary FFA in a dynamic climate. Thus, it is vital to refine flood return levels with non-stationarity measures for effective and sustainable hydrological planning in climate-sensitive and regulated river basins.