CLDAS降水产品的适用性分析及机器学习订正应用

Applicability analysis and machine learning correction of CLDAS precipitation product

为获得中国气象局陆面数据同化系统(CLDAS)多源降水融合实况分析产品在甘肃省的精细化评估结论并提高该产品的精度,采用甘肃省2215个地面站的降水数据作为检验源,评估CLDAS快速融合和实时融合降水产品的适用性,并基于评估结论采用机器学习方法及优化策略,订正CLDAS降水产品。结果表明:CLDAS降水产品能较好地反映甘肃省降水时空分布,日值的精度整体不如小时值;XGBoost算法较其他模型有显著优势,且在降水量大时订正效果更显著;通过提升训练数据集质量、特征工程和精细调参优化后的XGBoost模型使3级以上降水量的均方根误差减少近50%,1~2级的误判降水得到明显改善,该模型和算法能推广应用于具有相同气候特征的领域。

In order to obtain the refined evaluation conclusion of the multi source precipitation fusion real time analysis product of the China Meteorological Administration Land Data Assimilation System(CLDAS)in Gansu Province and improve the accuracy of the product,precipitation data from 2215 ground stations in Gansu Province were used as the test source to evaluate the applicability of CLDAS rapid fusion and real time fusion precipitation products.Based on the evaluation conclusion,machine learning methods and optimization strategies were used to correct the CLDAS precipitation product.The results showed that the CLDAS precipitation product can better reflect the spatiotemporal distribution of precipitation in Gansu Province.The accuracy of daily values is not as good as hourly values.The XGBoost algorithm has significant advantages over other models,and its correction effect is more significant when precipitation is high.By improving the quality of the training dataset,feature engineering,and fine-tuning optimization,the XGBoost model reduces the root mean square error of precipitation above level 3 by nearly 50%,significantly improves the misjudgment of precipitation between level 1 and level 2.This model and algorithm can be applied to fields with the same climate characteristics.