Winter wheat yield estimation based on support vector machine regression and multi-temporal remote sensing data

Developing and establishing high accurate models for crop yield estimation using remote sensing is of great significance in decision making for national food import/export and food security. A machine learning methodology called support vector machine regression (SVR) was introduced to construct remote sensing estimation model. Firstly, NDVIs from multi-temporal Landsat TM for main growing stage of winter wheat in 2004 and 2007 in Beijing suburb were used to construct yield estimation model by remote sensing through SVR. Secondly, cross validation was made on the model's stability and forecasting ability, and then the performance of SVR methodology was compared with traditional multivariate linear regression (MLR) methodology. The results showed that yield estimation model by remote sensing based on SVR could increase the precision of yield prediction. The determination coefficients were increased by 0.2162 and 0.2158, respectively, while the root mean squared errors were decreased by 0.1682 and 0.2912 in 2004 and 2007 compared with the multivariate regression methodology. Therefore, it is feasible and effective to estimate winter wheat yield by constructing estimation model based on SVR and multi-temporal remote sensing data, which provides the method to estimate the winter wheat yield via multi-temporal remote sensing data.