Applicability analysis and machine learning correction of CLDAS precipitation product
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 2 215 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.
CLDAS precipitation productassessmentmachine learningXGBoost model
万方数据
维普
