Abstract: 【Objective】With the advantages of wide coverage, high speed, large amount of information and strong dynamics, satellite remote sensing technology can obtain crop yield timely and accurately, reflect the spatial change trend of field crop yield. The remote sensing technology has become a hot research topic in agricultural production to estimate crop yield. Through improving the method of establishing remote sensing estimation yield models, this research aims to make further efforts to improve the accuracy of predicting crop yield and provide an intuitive and reliable reference for the macro understanding of crop yield formation and changes in different regions.【Method】In this paper, based on experimental data obtained from 2011-2012 in the fixed-point observation experiment in 5 counties of Jiangsu province (Dafeng, Xinghua, Jiangyan, Taixing, Yizheng), remote sensing data of HJ-1A/1B satellite images were used to analyze the quantitative correlations between the remote sensing vegetation index, key growth index and wheat yield per unit area at anthesis in order to further enhance the mechanism and reproducibility of remote sensing inversion models. The direct model building method was used to analyze the correlation between satellite remote sensing variables and wheat yield directly. While the indirect model building methods needed to choose remote sensing variables which closely related with yield, and choose growth indices which closely related with the remote sensing variables. Firstly, the corresponding wheat growth indices were monitored by using the sensitive remote sensing variables. Then, the indirect estimation model could be established and worked for the indirect remote sensing estimation in wheat yield. Based on the remote sensing vegetation index and the highest relationship, sensitive remote sensing variables were selected to estimate wheat yield, and the wheat yield estimation model, which was built and analyzed with ground measuring results in 2012, was analyzed with the linear regression analysis method and established by using direct and indirect model building methods. Based on the evaluation indexes: R² and RMSE, the accuracy of the two models was validated and compared using the observed data in 2011 in order to increase the quantitative level and reliability of remote sensing inversion models.【Result】Single factor models based on difference vegetation index (DVI) or ratio vegetation index (RVI) extracted from HJ-1A/1B image could predict the yield directly with root mean square error (RMSE) of 918 kg·hm^-2 and 1 399.5 kg·hm^-2. A two-factor model based on DVI and RVI could predict the yield directly with RMSE of 1 036.5 kg·hm^-2. The RMSE of the indirect yield model based on normalized difference vegetation index (NDVI) and leaf nitrogen accumulation was 805.5 kg·hm^-2. It was concluded that the HJ-1A/1B image at flowering stage was feasible and the precision was high. The accuracy of the indirect yield estimation model based on NDVI and leaf nitrogen accumulation was significantly higher than that of the direct estimation model. The RMSE decreased by 112.5 kg·hm^-2, 594 kg·hm^-2 and 231 kg·hm^-2 with the comparison of DVI direct estimation model, RVI direct estimation model and two-factor model, respectively. 【Conclusion】It was confirmed that HJ-1A/B, the satellite made in China, can meet the requirement of wheat yield estimation. Compared to the direct method, it is more feasible to predict field crop yield through remote sensing model based on the indirect method. The results provide a new way to accurately estimate field wheat yield using remote sensing technology.

Key words: wheat, HJ-1A/1B, anthesis, yield, prediction models