Evaluating Climate Change Impacts on Streamflow Changes in the Source Region of Yellow River: A Bayesian Vine Copula Machine Learning (BVC-ML) Approach

Abstract

In this study, we proposed a Bayesian Vine Copula Machine Learning (BVC-ML) method to predict streamflow changes in the Yellow River source area based on projections from three GCMs under various climate change scenarios. The BVC-ML method was to (i) use the vine copula method to reflect the interdependence between the predicted variable (i.e., streamflow) and predictions from different machine learning (ML) techniques, (ii) derive deterministic and probabilistic predictions from the vine copula model conditional on corresponding ML predictions and (iii) integrate predictions from different vine copula models to generate the final results. The proposed BVC-ML method was then applied for future streamflow projections based on outputs from CMIP6. The results from the BVC-ML method show that the studied area would generally experience more streamflow increases in most months, and the increases would become more significant as the climate change shifts from SSP126 to SSP585. The outputs from different GCM models also lead to various streamflow changes in the studied area, with the projections from ACCESS-CM2 leading to the highest streamflow increases. Furthermore, the BVC-ML method is capable of deriving both deterministic and probabilistic predictions from the conditional distributions, and the 10% and 90% quantiles can reflect predictive uncertainties. The results from the quantile predictions show that May, July and October would have the highest increases in streamflow, which are consistent with the mean streamflow increases. Overall, the proposed BVC-ML method is demonstrated to be a promising tool for predicting streamflow changes under different climate change scenarios. The findings would have significant implications for water resource management and climate adaptation over the studied region.