冬小麦遥感估产回归尺度分析

Regression scale analysis of winter wheat yield estimation by remote sensing

将统计业务和遥感估产结合起来,以北京市统计局提供的实割实测产量数据作为野外样方,利用抽样村和地块两种尺度的实测数据,用抽样村整体回归、地块整体回归和地块分层回归3种方法进行遥感估产,将所得结果与北京市统计局发布的统计单产从不同级别进行比较分析。结果表明,利用抽样村和地块两种尺度的实测数据进行回归估产都可以得到高精度的市级单产;在区县级别上利用地块尺度的实测数据进行估产得到的区县级单产精度高于抽样村尺度;在村级上利用地块实测数据进行单产预测能够较抽样村尺度更好的反映实际单产,模型更加稳定。因此,利用地块尺度的实测产量数据建立整体回归和分层回归模型都是可行,有效的,可以得到小区域尺度高精度的单产结果。

In the paper, statistic survey and yield estimation by remote sensing were combined by using two kind of ground survey data provided by Beijing Statistical Bureau, including sample village scale and plot scale to predict winter wheat yield. Sample villages, plots and plots stratification were used to estimate yield, and the correlations between predicted yield and statistic data were analyzed on different scales. The results showed that using ground survey data of sample village scale and plot scale both could get high-precision yield in Beijing area; on district-level, using ground survey data of plot scale could get more accurate yield than sample village scale; on the village level, predicting yield with ground survey data of plot scale could fit statistical yield better than sample village scale, and the yield model was more stable. Therefore, using ground survey data of plot scale to build entirety regression and stratified regression model are feasible and effective, both of them can get estimated yield of high-precision on small region level.