基于EnKF和随机森林回归的玉米单产估测

Estimation of Maize Yield Based on Ensemble Kalman Filter and Random Forest for Regression

为了提高玉米的估产精度,以河北省中部平原为研究区域,采用CERES-Maize模型模拟2013—2018年8个典型样点玉米整个生育期的叶面积指数(LAI),将遥感反演的LAI与CERES-Maize模型模拟的LAI相结合,通过集合卡尔曼滤波(EnKF)同化算法实现2013—2018年玉米主要生育时期旬尺度LAI的同化,运用随机森林回归法计算同化和未同化的LAI权重,进而建立玉米单产估测模型,对2015年53个县(区)的玉米进行单产估测和精度评价,并分析2013—2018年玉米的单产时空分布特征。结果表明,采用EnKF算法对8个研究样点进行单点同化,同化LAI更符合玉米实际生长情况;将样点LAI同化值从单点尺度扩展到区域尺度,同化LAI图像减少了相邻像素间LAI陡升陡降的现象,其效果优于遥感反演的LAI;与未同化LAI构建的估测模型相比,应用同化的LAI所建的估测模型精度明显提高,R2提高了0.0245;在2015年河北中部平原53个县(区)估产结果中,总体平均相对误差为12.11%,RMSE为371 kg/hm^2,NRMSE为6.18%;河北中部平原玉米单产估测结果呈现个别年份波动、总体呈先减少后增加的年际变化特点,并呈现西部地区最高、北部和南部地区次之、东部地区最低的空间分布特征。

In order to improve the estimation accuracy of maize,the central plain of Hebei Province was chosen as research area,and the remote sensed LAI and simulated LAI by CERES-Maize model was combined in eight typical samples from 2013 to 2018 by using the ensemble Kalman filter(EnKF)data assimilation approach.The random forest regression was used to estimate maize yield by using monitored LAI and the assimilated ones respectively.The optimal model for estimating maize yields in study area from 2013 to 2018 was selected,and the measured maize yield of the year 2015 was used to validate the accuracy of the optimal model.The results showed that the single point assimilation of eight samples using the EnKF algorithm was more consistent with the actual growth of maize.The assimilated LAIs were extended from the sampling sites to the regional scale,the phenomenon of LAIs rising and falling between adjacent pixels was reduced and the effect was better than the remote sensing inversion LAIs.Compared the yield estimation models with the monitored LAIs,the accuracy of the yield estimation models with the assimilated LAIs was improved,and the R2 was increased by 0.0245.The yield estimation model was applied to estimate maize yield in 53 counties(districts),in general,the average relative error of the estimated yield was 12.11%,and the root mean square error was 371 kg/hm^2,the normalized root mean square error was 6.18%.The yearly estimated yield from 2013 to 2018 in the central plain of Hebei Province was fluctuated in individual years,and the overall distribution in time was characterized by a tendency to decrease first and then increase,and the spatial distribution of maize yield was the highest in the western region of the plain,following by the north and south regions,and the lowest was in the eastern region.