Refining Extreme Event Selection in Hydrological Analysis Using the Comparative Distribution Fitting Method

Climate change has intensified extreme events across the globe, highlighting the need to study their nature and underlying drivers. However, the accuracy of such analyses depends on precise extraction of extreme events from hydrological time series. Identifying these events is challenging due to the spatial variability in hydrological patterns. Some regions experience frequent extremes, while others witness rare occurrences, making it difficult to define a universally applicable criterion for extreme event selection. Among the most commonly used approaches for extreme event extraction, the block maxima (BM) method selects the highest value within a fixed time block, while the peaks over threshold (POT) method identifies events exceeding a predefined threshold. Despite their widespread adoption, these methods lack a strong theoretical basis for determining an appropriate block size or threshold value, leading to inconsistencies in extreme event analysis. This study evaluates these traditional methods alongside the proposed comparative distribution fitting (CDF) approach, which determines the threshold based on statistical analysis of the time series data, ensuring a more robust and data-driven selection process. Results demonstrate that selecting extreme events using the CDF method significantly improved model performance, such that even stationary models under the CDF-based POT approach outperformed nonstationary models fitted to BM extractions. Extreme value modeling further confirms that CDF-based nonstationary models consistently yield superior results across all stations, reinforcing the importance of precise threshold selection. The findings provide a framework for enhancing extreme event analysis, underscoring that improper extreme event selection can undermine model reliability.