Bayesian Estimates of Changes in Russian River Runoff in the 21st Century Based on the CMIP6 Ensemble Model Simulations

Abstract

Based on simulations with an ensemble of CMIP6 (Coupled Models Intercomparison Project, phase 6) climate models using Bayesian averaging, an analysis of changes in the runoff of a number of Russian rivers—the Volga, Ob, Yenisei, Lena, Amur, and Selenga—has been carried out this century. Bayesian weights take into account the skill of runoff reproduction by models (long-term average runoff, linear runoff trend over a time interval with available runoff observations, and interannual and interdecadal variability). The skill of reproduction of runoff characteristics by individual CMIP6 ensemble models varies most widely for long-term average runoff; runoff trend; and, to a lesser extent, interannual variability. In the 21st century, the ensemble average runoff increases for most of the analyzed rivers, with the exception of the Volga. This increase is more pronounced in scenarios with large anthropogenic impacts. It is especially significant for the SSP5-8.5 scenario (Shared Socioeconomic Pathways, 14 5-8.5), in which the trend of increase in runoff in 2015–2100 relative to its modern long-term average value is (10 ± 4)% for the Ob, (16 ± 3)% for the Yenisei, (39 ± 7)% for the Lena, (36 ± 7)% for the Amur, and (18 ± 6)% for the Selenga. The main reason for changes in ensemble average runoff in the 21st century in models under all SSP scenarios is the changes in precipitation. Accounting for differences in model skill when reproducing river runoff on average for 2015–2100 reduces intermodel deviations relative to the corresponding values when uniformly weighting the model calculation results by 6–26%, depending on the SSP scenario and river catchment.