Application of Ensemble Kalman Filter and Random Forest Algorithm in Assimilation and Fusion of Heterogeneous Remote Sensing Precipitation Data
To minimize the non-homogeneous error of heterology remote sensing precipitation products,this paper develops the data assimilation and fusion algorithm of Ensemble Kalman Filter(EnKF)combined with spatial Random Forest(RF).Five representative satellite remote sensing precipitation products in the Yangtze River Basin(ERA5,Ter-raClimate,GPM,TRMM,and PERSIANN-CDR)were chosen to carry out assimilation and fusion process using EnKF-RF on the basis of analyzing the consistency of star-ground precipitation data.And then the accuracy was evaluated by u-sing independent meteorological station.The results show that the accuracy of the heterogeneous products in capturing precipitation in the Yangtze River Basin ranks as TRMM＞GPM＞TerraClimate＞PERSIANN-CDR＞ERA5;The NNSE of the five precipitation products after EnKF assimilation increases to 0.93-0.96,and the root mean square error(RRMSE)decreases to 89.48-176.03 mm,which is higher 11.46%-22.34%and lower 96.35%-122.60%than that before assimila-tion;The RF fusion improved the accuracy of single-source precipitation products,and the NNSE of the fusion-generated precipitation products was 0.99,and the RRMSE was 43.56 mm.The EnKF-RF assimilation and fusion strategy for heter-ogeneous precipitation data reduced the error of single-source precipitation products,which has a large potential for appli-cation in the Yangtze River Basin and even at the global scale.
data assimilationdata fusionmulti-source heterogeneityremote sensing precipitation products
NETL
NSTL
万方数据
