A Bayesian Network-Based Multimodal Blended Multiscalar Standardized Index for Improved Assessment of Extreme Hydrological Events

Abstract

Extreme occurrences are recognized as serious risks for all organisms worldwide. The frequency and intensity of extreme events are gradually rising across the world. Therefore, accurate and precise assessment of extreme events is crucial for mitigation strategies and sustainable policies. Recent research has commonly used ensemble global climate models (GCMs) to simulate precipitation data. However, the development of different methodologies often underperforms due to errors and spatio-temporal variation in simulated data from various climate models. To address these challenges, the objective of this study is to enhance the framework of a multimodal ensemble to improve the future characterizations of extreme events. This paper introduces a new framework, the Multimodal Blended Multiscalar Standardized Index (MBMSI), based on Bayesian Networks (BNs), for integrating and blending multiple GCM outputs to more accurately assess the frequency and severity of extreme events. BNs offer a powerful probabilistic approach for capturing nonlinear dependencies, quantifying uncertainty, and effectively combining information from diverse models, making them particularly suitable for addressing the variability and structural uncertainty present in GCMs. MBMSI integrates K-Component Gaussian Mixture Distributions (K-CGMDs) for probabilistic quantification. The MBMSI framework is applied to 50 randomly selected grid locations across the Tibetan Plateau, where extreme drought events were observed in October 1933 and extreme wet events were observed in September 1933 at a one-month time scale. There are variations in the regional patterns of extreme drought and extreme wet category throughout a year, therefore six-time scales -i.e., one-, three-, six-, nine-, twelve- and twenty-four-month time scales, are used. The analysis reveals comparable counts of extreme wet and extreme dry events across the six, nine, and twenty-four-month time scales, indicating consistent variability in precipitation patterns. However, spatial distributions of extreme events vary across all the time scales, highlighting the dynamic nature of drought behavior over different time scales. These findings underscore the importance of considering climate variation and uneven precipitation distribution when developing drought monitoring systems, such as the continuous assessment of drought onset, duration, intensity, and spatial extent using observational data and model outputs. Consequently, the implications of this research are far-reaching. The MBMSI can be a valuable tool for climate scientists, hydrologists, and policymakers involved in drought risk management, climate adaptation planning, and water resource governance.