Geostatistical characterization of higher-order statistics in simulated data of precipitation under global climate models

Abstract

Global warming and climate change have increased the frequency and severity of hydroclimatic extremes, with drought remaining one of the most persistent threats to environmental sustainability and water security. Since precipitation largely controls the onset, persistence, and intensity of drought conditions, improving its statistical characterization is essential for understanding hydroclimatic variability and evaluating climate model performance. Although Global Climate Models (GCMs) are widely used to simulate precipitation under changing climatic conditions, their evaluation has predominantly relied on lower-order statistical measures, whereas higher-order moments that characterize the distribution of precipitation, particularly kurtosis, have received comparatively little attention. This study investigates the spatial variability of precipitation kurtosis derived from 22 CMIP6 GCMs over the Tibetan Plateau using monthly precipitation records from 1961 to 2014 at 95 locations. An observation-based precipitation dataset was incorporated as a reference for evaluating the spatial characteristics of GCM-derived kurtosis. Spatial dependence was characterized using the robust Cressie–Hawkins variogram estimator, and six theoretical variogram models were evaluated through leave-one-out cross-validation (LOOCV) before being incorporated into Simple, Ordinary, and Universal Kriging for spatial interpolation. The kriging approaches were comparatively evaluated LOOCV. The results reveal substantial inter-model differences in the spatial dependence of precipitation kurtosis, indicating contrasting representations of precipitation distributional characteristics among the selected GCMs. Among the evaluated kriging approaches, Ordinary Kriging generally provided comparatively lower prediction errors for the assessed GCMs. The optimized variogram models successfully captured the underlying spatial structure and produced coherent interpolation surfaces with spatially varying interpolation uncertainty. These findings provide an additional statistical perspective for climate model evaluation and may support future investigations of precipitation-based drought assessment, climate risk analysis, and ensemble weighting strategies.