Winter Wheat Growth Simulation Based On Multiple Crop Models And Remote Sensing Data Assimilation

Timely and accurate information on crop growth and yield is critical to the guarantee for agricultural production,food security and sustainable development of agriculture in a country or region.Data assimilation approach that integrate remote sensing data with crop models have been proposed and developed as a potential approach to improve the accuracy of regional crop growth monitoring and yield estimation.However,the crop growth system is a highly nonlinear,and the crop model is a simplified tool for this complex nonlinear system.Different crop models have a different emphasis.At present,most of the crop models and remote sensing data assimilation systems were based on a single crop model,which model uncertainty could be underestimated by under sampling of the relevant model space.Therefore,in order to improve the accuracy and robustness of the crop data assimilation system,a multiple crop models data assimilation approach was constructed based on the data assimilation method and multi-model ensemble forecasting method in this research.This research focused on the construction and implementation of winter wheat growth simulation and yield estimation based on the multiple crop models data assimilation scheme.Such as the assimilation variable LAI remote sensing inversion,the parameters sensitivity and uncertainty analysis of different crop models,the development of multiple models’data assimilation algorithms,the construction of multiple crop models and remote sensing data assimilation schemes and their application in winter wheat growth simulation and yield estimation.The main research conclusions include:(1)Using Landsat-7 ETM+,Landsat-8 OLI and Sentinel-2 MSI data,the enhanced vegetation index(EVI)time series data were computed.The recurrent neural network method LSTM was used to mapping the winter wheat area based on the EVI data and winter wheat phenology characteristics.The overall accuracy of winter wheat mapping was 93.67%,and the Kappa coefficient was 0.82.A winter wheat LAI artificial neural network inversion model was generated based on PROSAIL radiation transfer model and multispectral remote sensing data.The regional winter wheat LAI were retrieved using this model.The retrieval LAI results showed that the R~2,RMSE and RRMSE between the inversion and the measured values were 0.95,0.45 and 10.54%.Therefore,the LAI inversion results accorded with the winter wheat actual growth situation,which meet the needs of regional winter wheat assimilation studies.(2)In this study,three crop models with different structures and complexity were selected,which were SAFY-WB,WOFOST,and CERES-Wheat models.The EFAST sensitivity analysis method was used to analyze the sensitivity of all crop model parameters.The sensitivity analysis results showed that some parameters that were sensitive to yield but not sensitive to growth process variables(LAI).Therefore,when parameter calibration,it is necessary to comprehensively consider the parameters which both sensitive to production process variables and yield.Combined with the measured data,the GLUE method was used to obtain the uncertainty(posterior distribution)of the sensitive parameters of each crop model.The average of each parameter posterior distribution was used as the optimized value.The obtained posterior distribution of model parameters was used as the input of model parameters to compare the uncertainty of LAI and yield for different crop models.The results showed that during the flowering stage to the mature stage,the uncertainty of LAI simulation of CERES-Wheat model was relatively large.There was a significant difference between the three crop models in simulating the peak of LAI.The uncertainty of CERES-Wheat and WOFOST was lower than the SAFY-WB model on yield simulation results.After parameters calibration,all models have accurately simulated the LAI and yield.Therefore,these three crop models were feasible and effective for winter wheat growth monitoring and yield prediction in the study area.(3)Two multi-model data assimilation methods 4DVar+BMA and EnKF+BMA were proposed,based on the data assimilation method(4DVar and EnKF)and the multi-model ensemble prediction method(BMA).In these methods,the LAI assimilation result of each model was averaged using model weight,and then obtain LAI assimilation result of multiple models.The multiple crop model data assimilation scheme was designed and applied in the winter wheat growth simulation and yield estimation.The field winter wheat LAI and yield data were used to discuss and analyze the multiple crop model assimilation scheme method feasibility and the parameter setting.The results showed that the winter wheat LAI and yield assimilation based on the multi-model data assimilation method were superior to the single model assimilation results.Between the winter wheat LAI simulated and measured based on 4DVar+BMA and EnKF+BMA,the R~2 both were 0.97,the RMSE were 0.35 and 0.33,the RRMSE were 9.94%and 9.36%Between winter wheat yield simulated and measured based on 4DVar+BMA and EnKF+BMA,the RMSE were 332 kg/ha and 301kg/ha,RRMSE were 4.22%and 3.93%.The field test showed that the4DVar+BMA and EnKF+BMA multi-model data assimilation scheme were feasible and effective.(4)A series of regional winter wheat LAI in Hengshui were assimilated into the multiple crop growth models ensemble based on the EnKF+BMA-based strategy.On the 1km scale,between the LAI simulated and the measured values,the acceptable estimates of winter wheat LAI were obtained with R~2 were 0.95,the RMSE were 0.22,the RRMSE were 4.88%.Between winter wheat yield simulated and statistic values,the RE of each county and city was less than 12%,the hole RE were 5%.The study also found that the only assimilated the remote sensing data of the key growth stages could significantly improve the simulation accuracy.The regional winter wheat yield estimates would be decreased with the spatial resolution of LAI data decreased,and the computational efficiency were greatly improved.Therefore,it is necessary to select the reasonable remote sensing data with and assimilation scheme to obtain accuracy result,especially for the regional crop growth simulation and yield estimation.