Enhancing non-stationary analysis of extreme precipitation through a precise extreme event extraction approach.

The increasing frequency of global extreme hydrological events has highlighted the critical need for reevaluating hydraulic structures’ safety design considerations, mainly through non-stationary hydrological time series analysis. This study, conducted in the Krishna River Basin of India, aims to develop a robust methodological framework for non-stationary analysis of extreme precipitation events, emphasizing the importance of accurate extreme event extraction. Accurate extraction of extreme events is crucial for non-stationary analysis, as it ensures that the events analyzed are truly extreme. This precision is vital for reliable predictions and effective safety design in the face of changing climatic conditions. The study is divided into two major parts. First, the block maxima and peaks over threshold (POT) methods for extracting extreme events were compared. In the block maxima approach, a block size of one year was considered, whereas, in the POT approach, three threshold selection methods were considered: percentile-based (90th, 95th to 99th percentiles), top 'n' values and graphical method. The graphical method was identified as the most effective, based on parameter stabilization, return value matching from two extreme value distributions, and Akaike information criterion (AIC), confirming its superiority in model fitting. With accurate extreme events extracted, the study proceeded to non-stationary analysis (NSA) using nine covariates, categorized into climate change, global warming, local temperature anomalies, and trends. A total of 23 stations were analyzed, identifying significant covariate combinations for each station through the lowest AIC values. NSA indicated that the selected covariates significantly influenced the non-stationary behaviour of extreme precipitation events. This study emphasizes the critical need for precise extreme event extraction in non-stationary analysis. The graphical method for threshold selection and identifying significant covariates offers a reliable approach to understanding and predicting extreme precipitation events.