Improving winter wheat biomass and evapotranspiration simulation by assimilating leaf area index from spectral information into a crop growth model

Data assimilation, a state-of-the-art method that merges remote sensing data with a dynamic model to improve model performance, has been widely used in land surface process modeling. Application of data assimilation under various water conditions can provide insight of crop response to different water supply rates, which is useful for agricultural water management in arid and semi-arid regions. For this purpose, we developed a generic data assimilation methodology by integrating both the Shuffled Complex Evolution (SCE) and the Ensemble Kalman Filter (EnKF) algorithms into the Simple Algorithm For Yield and Evapotranspiration (SAFYE) model to provide improved simulation of winter wheat biomass and yield, and simulation of evapotranspiration (ET) under different water-supply scenarios. An experiment with nine irrigation scenarios was conducted during the 2013—2015 growing cycles. Field spectral data were employed to retrieve the leaf area index (LAI), which was then used as a single state variable to determine other parameters in the SAFYE model using a global optimization algorithm. Time-series LAI was eventually assimilated in the SAFYE model based on the EnKF algorithm to improve overall model simulation. The results showed that the simulated crop growth dynamics followed the measurements well in most cases when the estimated LAI was assimilated. The accuracy of simulated biomass at the daily step was improved, with the maximum RMSE decreased from 199.4 to 123.8 g m−2 and from 466.6 to 393.4 g m−2 for the 2013–2014 and 2014–2015 growing seasons respectively. A good agreement was also achieved between the estimated and field measured grain yield (R2 = 0.901, RMSE= 31.9 g m−2, RRMSE=6.55%) for both growing seasons. The simulation of soil water content in the top 0—20 cm soil layer was better (RMSE: 3.3—5.0 mm) than that of 0—100 cm layer (RMSE: 11.7—29.6 mm). Accuracy of the simulated ET under early-stage water deficit scenarios was lower than that under other scenarios, with a positive mean relative error of 14% (3.4—24.3%) during two growing seasons. This study demonstrates the great potential of coupling remote sensing data to improve the performance of SAFYE model in modeling winter wheat growth.