Application of multivariate recursive nesting bias correction, multiscale wavelet entropy and AI-based models to improve future precipitation projection in upstream of the Heihe River, Northwest China

Abstract

Accurate projection of future precipitation is a major challenge due to the uncertainties arising from the atmospheric predictors and the inherent biases that exist in the global circulation models. In this study, we employed multivariate recursive nesting bias correction (MRNBC) and multiscale wavelet entropy (MWE) to reduce the bias and improve the projection of future (i.e., 2006–2100) precipitation with artificial intelligence (AI)-based data-driven models. Application of the developed method and the subsequent analyses are performed based on representative concentration pathway (RCP) scenarios: RCP4.5 and RCP8.5 of eight Coupled Model Intercomparison Project Phase-5 (CMIP5) Earth system models for the upstream of the Heihe River. The results confirmed the MRNBC and MWE were important statistical approaches prudent in simulation performance improvement and projection uncertainty reduction. The AI-based methods were superior to linear regression method in precipitation projection. The selected CMIP5 outputs showed agreement in the projection of future precipitation under two scenarios. The future precipitation under RCP8.5 exhibited a significantly increasing trend in relative to RCP4.5. In the future, the precipitation will experience an increase by 15–19% from 2020 to 2050 and by 21–33% from 2060 to 2090.