Study On Winter Wheat Yield Prediction Based On Assimilating Remote Sensing Data And Crop Growth Model

The remote sensing yield estimation and crop growth simulation modeling have both advantages and limitations used in crop monitoring and yield forecasting. Remotely sensed technology is very useful in regional agriculture monitoring because of its Spatial-continuous and time-dynamic characteristics, it’s hard to reflect the intrinsic physiological and biochemical processes of crop growth. Crop growth models was succeed used in simulating of crop growth and development process of small-scale case and single-species crops the internal. But it is powerless in regional scale crop growth monitoring because of the parameters on a regional scale of the crop are difficult to get. As a technology which can combine the remote sensing data and crop growth model, data assimilation become an effective way of regional yield prediction.This paper took winter wheat of Hengshui district as the study object, WOFOST as the crop growth model, MODIS leaf area index(LAI) data as observation data, used Ensemble Kalman Filter(EnKF) algorithm to assimilate the crop growth model and remote sensing data to forecast the winter wheat production. To solve the problem of MODIS-LAI (MOD15A2)’low accuracy caused by the pixel heterogeneity, this study firstly removed the abnormal information with Savitzky-Golay(SG) filter and then corrected it with field measurements LAI data. The results show that this method can effectively eliminate the data anomalies of the MODIS-LAI. This paper took LAI as a connection point for remote sensing data and crop growth model, used EnKF to obtain a time-continuous of estimation LAI values. These values were input WOFOST model and estimate the yield of winter wheat. We used official statistics yield of2008winter wheat to validate the accuracy of simulated yield. This paper analyzes the impact of ensemble size, assimilation step, and run mode crop of model to the result. The results showed that assimilation corrected MODIS-LAI can reflect the spatial distribution of yield and the simulated yield value is more approximate to the statistical value (compared with the yield simulated by WOFOST model). Studies showed that the way of assimilating remote sensing data and crop growth model to predict crop yields has broad application potential in the regional-scale yield forecasting.