This study is conducted by a research team led by Professor Wenping He from the School of Atmospheric Sciences at Sun Yat-sen University. The team develops a new climate prediction method based on slow feature analysis, termed the SFA-based Climate Prediction (SCP) method. It directly extracts slowly varying, potentially predictable signals from precipitation data to predict the spatiotemporal evolution of summer precipitation anomalies over South China. Evaluation results demonstrate that the SCP method has higher prediction skill than two dynamical models and a hybrid dynamical–statistical prediction system, which provides a new approach and practical tool for regional seasonal precipitation prediction.
Unlike conventional statistical models that rely on a limited set of preselected predictors, the SCP method does not prescribe specific climate factors in advance. Rolling independent forecasts for 2011–2022 (lead time is about three months), show that the SCP method achieves a regional mean temporal correlation coefficient of 0.36, which exceeds that of the hybrid dynamical–statistical prediction system FODAS (0.19) and the two dynamical models BCC-CSM1.1 (0.13) and CFSv2 (0.08). It can also accurately predict whether regional mean precipitation is higher or lower than the normal level in 10 of the 12 summers. The correlation between predicted and observed regional mean precipitation anomalies reaches 0.62, while FODAS is 0.45, BCC-CSM is 0.28, and CFSv2 is 0.17. A comprehensive evaluation with various metrics proves that SCP always outperforms the three prediction systems.
By reducing the dependence on a set of preselected fixed predictors, the SCP method provides a new pathway for improving seasonal prediction skill of summer precipitation. Its flexible framework may also be extended to other regions, providing potential technical support for flood and drought risk management and climate-related decision-making.
See the article:
Wang S, Guo J, He, W. 2026. A new method for predicting summer precipitation over South China based on slow feature analysis. Science China Earth Sciences. 69(7), 2654–2662, https://doi.org/10.1007/s11430-025-1970-0
