Estimating winter wheat yield by assimilation of MODIS LAI into WOFOST model with Ensemble Kalman Filter

Abstract

Timely and accurate crop growth monitoring and yield estimation are vital for guaranteeing food security and agricultural sustainable development. As two main methods of crop yield estimation, crop growth simulation models and remotely sensed methods both have their respective advantages. Crop growth models are widely used in simulating crop development stage and yield in a small scale while satellite remote sensing has a great advantage in agriculture monitoring for its spatial continuity and temporal dynamic property. Data assimilation has become an effective way of estimating crop yield for it overcomes the limitations by combining satellite remote sensing data and crop growth models. Leaf Area Index (LAI) is an important vegetation biophysical parameter, which has been extensively applied in crop yield estimation. This study presents a method of assimilation of Moderate Resolution Imaging Spectrometer (MODIS) LAI data product into World Food Studies (WOFOST) model for winter wheat yield with Ensemble Kalman Filter (EnKF). We take winter wheat of Xinghua in Jiangsu province as the study object and chose WOFOST as the crop growth simulation model. Several winter wheat variety inheritance parameters and soil parameters are adjusted by the field measured data. Other sensitive parameters are adjusted by using FSEOPT optimization program to recalibrate the model and simulate the development stage and eco-physiological processes more accurately. The values of MODIS LAI (MCD15A3) are relatively low because of cloud contamination and mixed pixels. To solve this problem, this study firstly applies a Savitzky-Golay (S-G) filtering algorithm to MODIS LAI products to obtain filtered LAIs and then correct the filtered LAIs with field measurements data. This method can eliminate the anomalies and improve the accuracy of the MODIS LAI effectively. We take LAI as the assimilation state variable of EnKF algorithm and corrected MODIS LAI as observed data. Finally, the time-continuous LAI values are input WOFOST model to estimate winter wheat yield. We use statistics yield from Xinghua station to validate the accuracy of simulated yield. The root mean square error (RMSE) reduces from 587 kg/ha to 361 kg/ha compared to the official statistical yield data. Our results indicate that the accuracy of winter wheat yield is improved after the assimilation.