基于双参数和粒子滤波同化算法的夏玉米单产估测

Estimation of Summer Maize Yield Based on Bi-variables and Particle Filter Assimilation Algorithm

摘要:

为了提高河北省中部平原夏玉米的估产精度和进一步验证粒子滤波同化算法对农业作物估产的适用性，采用粒子滤波算法同化CERES-Maize模型模拟和MODIS数据反演的叶面积指数（Leaf area index，LAI）、条件植被温度指数（Vegetation temperature condition index， VTCI），应用随机森林回归算法确定夏玉米不同生育时期LAI和VTCI的权重，构建单产估测模型。结果表明，无论是单点尺度还是区域尺度，同化的LAI和VTCI均能较好地响应外部观测数据，同化LAI可减缓CERES-Maize模型模拟LAI的剧烈变化；同化VTCI结合模型模拟和遥感观测，更能反映夏玉米对水分胁迫的敏感性。利用2015年河北省中部平原各县（区）夏玉米产量对较优估产模型进行精度验证，结果表明，同化前后夏玉米产量模拟结果与统计产量间的归一化均方根误差由12.71%下降到10.50%，平均相对误差由12.57%下降到8.43%，说明基于同化LAI和VTCI构建的双参数单产估产模型可用于区域夏玉米单产估测。

Abstract:

To validate the feasibility of particle filter assimilation algorithm for crop yield estimation, and improve accuracy of summer maize yield estimation in the central plain of Hebei Province, the leaf area index (LAI) and vegetation temperature condition index(VTCI) simulated by the CERES-Maize model were assimilated with the state variables retrieved from MODIS data. The random forest method was used for determining the weights of different variables at the growth stages of summer maize. The maize yield estimation model was established based on the weights of variables and measured yield. The results showed that no matter at the sampling sites or at regional scale, the assimilated LAI and VTCI were better able to respond the monitored LAI and VTCI, the assimilated LAI decreased the sharp changing points which LAIs were simulated by CERES-Maize, the assimilated VTCI was in good agreement with those of both the remotely sensed VTCI and the simulated VTCI, and the assimilated VTCI was a good index for indicating crop water stress of summer maize. The optimal model was selected for summer maize yield and estimation accuracy of the year 2015 in the central plain of Hebei Province, normalized root mean square error (NRMSE) between simulated and observed summer maize yields before and after performing PF assimilation scheme was decreased from 12.71% to 10.50%, and relative error (RE) was decreased from 12.57% to 8.43%. Therefore, the established yield model based on the assimilated LAI and VTCI fully integrated the advantages of remote sensing information and crop model, and can be used for estimating summer maize yield accurately.