Time-Varying Evaluation of Compound Drought and Hot Extremes in Machine Learning–Predicted Ensemble CMIP5 Future Climate: A Multivariate Multi-Index Approach

• Published in: Journal of hydrologic engineering 2024-04, Vol.29 (2)
• Main Authors: Swain, Sushree Swagatika, Mishra, Ashok, Chatterjee, Chandranath
• Format: Article
• Language: English
• Subjects: Agricultural industry; Climate; Climate and hydrology; Climate models; Climatic extremes; Climatic indexes; Drought; Evaluation; Extreme weather; Future climates; Gaussian distribution; Global climate; Global climate models; Hydrology; Learning algorithms; Machine learning; Modelling; Multivariate analysis; Normal distribution; Precipitation; Regional climate models; Regional climates; Resource management; Rivers; Runoff; Scaling; Standardized precipitation index; Statistical analysis; Statistical methods; Support vector machines; Temperature index; Water resources; Water resources management
• ISSN: 1084-0699; 1943-5584
• DOI: 10.1061/JHYEFF.HEENG-6026

Compound extremes can be expressed as the joint distribution or dynamic interaction of multiple variables and the interdependence of several extremes that have major effects on the agricultural sector. Analysis of these compound extremes in space-time varying domains is a challenging task for climate experts. It is even difficult to forecast such compound extremes in future climate when several widely accepted global climate models (GCMs) and regional climate models (RCMs) are available. Thus, an attempt is made to develop a compound extreme index considering precipitation, temperature, and runoff, with a specific focus on promoting sustainable water resource management, particularly for the benefit of the agricultural sector. This work presents a new outlook on the modeling framework for compound drought and hot extremes (CDHE) by developing the Standardized Compound Extreme Event Index (SCEEI), and highlights their time-varying evaluation (trend and prediction) along with severity assessments. The SCEEI is derived using three individual indices: Standardized Precipitation Index (SPI), Standardized Runoff Index (SRI), and Standardized Temperature Index (STI), which are obtained from three physical variables (precipitation, runoff, and temperature) followed by multivariate Gaussian distribution. The applicability of the SCEEI is explained with a case study on the Brahmani and Baitarani rivers of eastern India, aiming at the modeling of dry and hot extremes at a regional scale. Further, the time-varying evaluation of CDHE events—the projection of the extremes—is studied in machine learning (ML)-based ensemble future. Random forest (RF) and support vector regression (SVR) MLs are employed to derive ensemble future from seven RCM models of the Coordinated Regional Downscaling Experiment (CORDEX). The linear scaling approach (LSA)-based bias-corrected data were used as input for the ensembling process. The RF-predicted ensembles performed statistically better for the study area. The extreme events and their severities are further assessed in RF-predicted future ensembles. The two-tail-based Mann–Kendall test, followed by Sen’s slope estimator, is used to investigate the trend-based time-varying evaluation of CDHE events. The notable conclusion drawn from this study is that the severity and frequency of CDHE events are increasing in the ensemble future climate, particularly in representative concentration pathway (RCP) 8.5. The developed multivariate framework