The Application of the SVD Method to Reduce Coupled Model Biases in Seasonal Predictions of Rainfall

Abstract

Abstract The large systematic biases in coupled models impact seasonal prediction results. With a motivation to reduce the influence of coupled‐model biases on seasonal predictions, the singular value decomposition method was applied in our study to improve the ability to predict flood season precipitation. Based on the coupled climate model, CAS‐ESM‐C, we conducted ensemble seasonal prediction experiments from 1982 to 2018, with initial conditions provided by the assimilation system. The prediction system was integrated from March to August of each year with a focus on the June to August precipitation in China. The results showed that the prediction skills for anomalous summer precipitation were very low without bias corrections. However, the system effectively predicted the interannual variabilities in large‐scale atmospheric circulation systems that were associated with anomalous summer precipitation. We used the singular value decomposition method to reduce pattern‐dependent precipitation errors by replacing prediction patterns with observation patterns, and the predictive skill for precipitation dramatically improved. The results demonstrated that this correction method is a viable tool to reduce systematic biases in coupled model predictions.