基于同化叶面积指数和条件植被温度指数的冬小麦单产估测

应用数据同化方法将遥感信息与作物生长模型融合,是估测区域作物产量的重要方法之一。以2008—2014年越冬后的冬小麦为研究对象,选择与作物长势、产量及水分胁迫信息密切相关的叶面积指数(LAI)和条件植被温度指数(VTCI),采用粒子滤波算法对CERES-Wheat模型模拟和遥感数据观测的LAI和VTCI实施同化,分别基于观测LAI和VTCI、同化LAI和VTCI构建冬小麦单产估测模型。结果表明,同化LAI变化趋势更加符合关中平原冬小麦的实际生长状况,同化VTCI能更好地反映冬小麦的水分胁迫程度。应用观测LAI和VTCI构建的估产模型决定系数为0.402,而单独应用LAI或VTCI单变量构建的估产模型决定系数分别为0.279和0.339,说明应用LAI和VTCI双变量构建的估产模型的精度优于单独应用LAI或VTCI单变量的精度。相比于观测LAI和VTCI构建的估产模型,基于同化LAI和VTCI构建的估产模型的决定系数从0.402提高到0.547。表明基于同化LAI和VTCI构建的估产模型的精度明显提高。

Winter wheat yield estimation based on the assimilated leaf area index and vegetation temperature condition index

Recently, the combination of remote sensing and crop growth models using data assimilation methods becomes an important approach for estimating regional crop yields. A wheat yield estimation study was carried out in the Guanzhong Plain of Shaanxi Province, China, in the years from 2008 to 2014 after the reviving of winter wheat, and the leaf area index (LAI) and vegetation temperature condition index (VTCI) were chosen as the key parameters. A particle filter algorithm with the sequential important sampling procedure was applied to assimilate the LAI and VTCI retrieved from MODIS data and those simulated by using the CERES-Wheat model. Winter wheat yield estimation models were developed using the observed LAI and VTCI or the assimilated LAI and VTCI, respectively. The results showed that the assimilated LAIs had good temporal and spatial continuity, and the peak and seasonal trend of the assimilated LAIs were more in line with the winter wheat actual grow and development in the Plain. The assimilated VTCIs were better correlated with precipitation, suggesting the assimilated VTCI was a better indictor for crop water stress of winter wheat. Compared with the yield estimation models using the LAI or VTCI a single variable (the coefficient of determination, R²=0.279 or 0.339), the models based on the double variables had better accuracy (R²=0.402). Compared with the estimation models developed by using the observed LAI and VTCI (R²=0.402), the models developed by using the assimilated ones had better accuracy (R²=0.547), indicating that the yield estimation accuracy was further improved by using the assimilated LAI and VTCI.