基于CFSv2产品和机器学习的东江流域月降水预报

Monthly precipitation forecast in the Dongjiang Basin based on CFSv2 products and machine learning

中长期降水预报一直以来是研究水文气象变化的热点,其精度与可靠性不高的问题亟待解决。以东江流域为研究对象,通过距平相关系数ACC、标准化均方根误差NRMSE、平均绝对误差MAE和多模型稳定性指数MSI评估CFSv2模式产品在月尺度的预测精度与稳定性,采用CFSv2模式降水预报、CFSv2模式预报因子结合机器学习模型预报2种方法预测未来降水。结果表明,不同预见期下,CFSv2模式降水预测与实测降水量具有较高的相关性,对于枯水期的预测效果好于汛期,但随着起报时间发生改变,降水预测的差异性较大,模型稳定性较差;CFSv2模式预报因子结合机器学习模型提高了预测的稳定性,相较于CFSv2模式降水预测,MSI从0.45降低到0.25,在很大程度上减小了由于起报时间改变产生的预报随机性。研究成果可为中长期降水预测提供一种新的思路,并为中长期水文预报和水资源管理提供决策依据。

Mid to long-term precipitation forecasting has always been a hot topic in hydrometeorological research,with the issue of low accuracy and reliability needing urgent solutions.This study focuses on the Dongjiang Basin and evaluates the prediction accuracy and stability of CFSv2 model products at the monthly scale using the anomaly coefficient of correlation(ACC),normalized root mean square error(NRMSE),mean absolute error(MAE),and the multi-model stability index(MSI).Two methods,namely the CFSv2 model precipitation forecast and the machine learning model forecast combined with CFSv2 model predictors,are employed to predict future precipitation.The results show that under different lead times,the CFSv2 model precipitation forecast exhibits a high correlation with observed precipitation,performing better during the dry season compared to the flood season.However,there is significant variability in precipitation forecasts and poor model stability with changes in the initial time.Combining CFSv2 model predictors with machine learning models improves the forecast stability,reducing the MSI from 0.45 to 0.25 and effectively reducing the randomness in forecasts caused by changes in the initial time.The findings contribute to providing a new approach for mid to long-term precipitation forecasting and offer decision-making support for mid to long-term hydrological forecasting and water resource management.