基于粒子滤波和多变量权重的冬小麦估产研究

为了构建能够反映作物长势的综合性指标以及准确估测作物产量,采用粒子滤波算法同化CERES-Wheat模型模拟和基于Landsat数据反演的叶面积指数(Leaf area index,LAI)、地上生物量和0~20 cm土壤含水率,获取冬小麦主要生育期以天为尺度的变量同化值,分析不同生育时期的LAI、地上生物量和土壤含水率同化值与实测单产的相关性,并应用熵值的组合预测方法确定不同状态变量影响籽粒产量的权重,进而生成综合性指数,并分析其与实测单产的相关性。结果表明,LAI、地上生物量和土壤含水率同化值和田间实测值间的均方根误差(Root mean square error,RMSE)以及平均相对误差(Mean relative error,MRE)均低于这些变量模拟值和实测值间的RMSE和MRE,说明数据同化方法提高了时间序列LAI、地上生物量和土壤含水率的模拟精度。基于不同状态变量的权重生成的综合性指数与实测单产间的相关性大于单个变量与实测单产间的相关性;基于综合性指数构建小麦单产估测模型,其估产精度(R2=0.78,RMSE为330 kg/hm2)分别比基于LAI、地上生物量和土壤含水率建立模型的估产精度显著提高,表明构建的综合性指数充分结合了不同变量在作物估产方面的优势,可用于高精度的冬小麦单产估测。

Winter Wheat Yield Estimation Based on Particle Filter Algorithm and Weights of Multi-variables

To establish a comprehensive index for monitoring the crop growth and estimating the crop yields accurately, the leaf area index (LAI), aboveground biomass and soil moisture (0~20cm) simulated by the CERES-Wheat model were assimilated with the state variables retrieved from Landsat data using the particle filter algorithm, for obtaining daily assimilated LAI, aboveground biomass and soil moisture values. Then linear regression analyses were performed to examine the relationships between the assimilated LAI, aboveground biomass or soil moisture and field-measured yields respectively, which were combined with the combination forecasting of entropy method, for determining the weights of different variables at the main growth stages of winter wheat. The comprehensive index was established based on the weights of variables, and the linear correlations between comprehensive index and measured yields were used for establishing wheat yield estimation model. The results showed that the root mean square errors (RMSEs) and mean relative errors (MREs) between the assimilated state variables and the field-measured ones were lower than the RMSEs and MREs between the simulations and the field-measurements, respectively. Thus the accuracies of the assimilated LAI, aboveground biomass and soil moisture time series were improved through the assimilation process. In addition, the correlation coefficients between the comprehensive index and the yields were higher than those between the individual variables and the yields at each wheat growth stage. And the accuracy of the yield estimation model established based on the comprehensive index (R2 was 0.78 and RMSE was 330kg/hm2) was significantly higher than those of the models established based on the LAI (R2 was 0.62 and RMSE was 448kg/hm2), aboveground biomass (R2 was 0.64 and RMSE was 431kg/hm2) and soil moisture (R2 was 0.67 and RMSE was 442kg/hm2) respectively. Therefore, the established comprehensive index fully integrated the advantages of the different variables in estimating crop yields, which can be used for estimating wheat yields accurately.