Abstract: 【Objective】 It is a key research and application field in agricultural remote sensing to estimate crop yield through remote sensing technology, and provide timely and reliable yield information for regional field production. Accurate estimation of crop yield has important significance to ensure national food security, to constitute the social development planning, to guide and regulate the macro adjustment of planting structure, and to improve the management skills of agricultural enterprises and farmers. In order to further improve the accuracy of estimating winter wheat yields by remote sensing, and display application effect of domestic imaging products in agricultural production, this study constructs theoretical yield estimation model based on the domestic remote sensing image through screening sensitive remote sensing variables of estimating theory yield of winter wheat, so as to achieve theory yield estimation of regional winter wheat by remote sensing and provide a reference to timely understand yield tendency of winter wheat at the different ecological regions. 【Method】 The research used 2010-4-26, 2011-4-28, 2012-4-28 and 2013-5-2 HJ-1A/1B images at winter wheat anthesis stage as remote sensing data and extracted 13 remote sensing variables. In Jiangsu Province, the 5 counties of Taixing, Jiangyan, Yizheng, Xinghua and Dafeng were selected as the experimental sampling area, and representative samples were selected samples in the experimental sampling area, and were measured indoor. A total of 335 measured samples of winter wheat theoretical yield were divided into modeling datasets and testing datasets according to the ratio of 3﹕2. Based on the minimum value of predictive residual error sum of square (PRESS), the number of required principal component model was determined. The yield estimation model was assessed through determination coefficient (R²), root mean square error (RMSE) and determination coefficient (R²). This research was undertaken to make a systematic analysis on the quantitative relationships of satellite remote sensing variables to winter wheat theoretical yield. Depending on the partial least squares regression (PLS), the multivariable remote sensing estimation models and the space level distribution maps of winter wheat theoretical yield were constructed and verified through modeling and testing datasets, and the estimation effect of the PLS model was compared to linear regression (LR) and principal components analysis (PCA) algorithm models, respectively. 【Result】 The results of this research indicated that the majority of remote sensing variables were significantly related to theoretical yield, and there were significant multiple relationships among the majority of remote sensing variables. For the theoretical yield model based on PLS, the number of the best principal components was 4. Structure intensive pigment index, Normalized difference vegetation index, Green normalized difference vegetation index and Plant senescence reflectance index were identified as the sensitive remote sensing variables for estimating winter wheat theoretical yield. Through testing the theoretical yield model based on PLS algorithm with modeling and testing datasets, for the theoretical yield model, the R² were 0.79 and 0.76, respectively, the RMSE were 720.45 kg·hm^-2 and 928.05 kg·hm^-2, respectively, the RE were 11.45% and 13.92%, respectively. The PLS models with selected sensitive variables performed better to estimate winter wheat theoretical yield. PLS algorithm models to estimate winter wheat theoretical yield obtained the higher accuracy by above 25% and above 27% than the LR algorithm models, by above 15% and above 16% than the PCA algorithm models, respectively. The results of applying the PLS model were correspondent with the actual distribution of winter wheat theoretical yield on regional scale and had strong application ability. 【Conclusion】It was concluded that PLS algorithm could provide an effective way to improve the accuracy of estimating winter wheat theoretical yield on regional scale based on aerospace remote sensing, and contribute to large-scale application and promotion of the research results.

Key words: remote sensing, winter wheat, HJ-1A/1B images, partial least squares regression (PLS), theoretical yield estimation model