Regional Wheat Growth Monitor And Yield Predict Based On Sequential Assimilation Technique

The real-time and regional characteristics of remote sensing(RS)are complementary to the mechanism and prediction of crop model,and the combination of RS information and crop model is an effective approach to realize the accurate prediction of regional crop yield.The sequential assimilation strategy based on the suppose that the optimization of the simulation value of the model at a certain time can improve the simulation accuracy,which takes errors of the RS observations and crop model simulation values into account,is one of the hot topics in the study of coupling RS data with crop model.The sequential assimilation technique can avoid the errors resulted from inverting agricultural parameters by RS data if directly using RS data as assimilation parameters,which has particular advantages theoretically.In addition,the simulation process of coupling RS with crop model is mainly based on pixel by pixel,which brings large amount of calculations.Therefor,the problem of simulation efficiency also has become one of hotspots in the study of coupling RS data with crop model.In this study,the PROSAIL model is added on the basis of the coupling between RS data and wheat growth model(WheatGrow),and the vegetation index(VIs)in wheat different growth stages were used as coupling parameters.A sequential assimilation technique was constructed through assimilating the temporal VIs from RS inversion and the WheatGrow-PROSAIL model simulation,which was used to obtain the optimal LAIs sequence,and then to motivate WheatGrow to predict wheat growth indicators and grain yield more precisely.The result would provide significant theoretical basis and technical support for growth monitor and yield prediction of regional winter wheat.Based on the management zone method,wheat canopy soil adjusted vegetation index(SAVI)at different growth stages and soil nutrient indices,including alkaline nitrogen,organic matter and available potassium content,were selected as data sources to delineate simulation zones.At the same time,using the fuzziness performance index(FPI)and normalized difference vegetation index(NCE)to identify the best partition number.Results show that the best partition number is 10,and the values of variation coefficients of each zone,including alkaline nitrogen,organic matter and available potassium content,were between 1.56-5.62%,1.09-3.75%and 2.20-6.62%,which were less than their values in the whole study area with range of 5.63%,10.30%and 10.64%respectively;The variation coefficients values of wheat canopy SAVI at different growth stages of each zone,were between 3.52-16.47%,2.56-10.21%and 2.17-9.07%respectively,which were less than their variation coefficient values range of 24.25%,15.98%and 12.29%in the whole study area.The correlation coefficient(R2)and root mean square error(RMSE)between measured yield values and simulated values by running coupling model on each zone reached 0.683,717.21 kg·ha-1 respectively,which indicts that the partition achieved good effects.This study used the time series images with high spatial resolution,fused by the high spatial resolution and high temporal resolution images,as the information fusion point of RS data and crop model,and using the sequential assimilation technique to predict wheat growth indicators and grain yield more precisely.Furthermore,experimental data from different ecological sites and different years were used to acquire the optimal coupling VI and growth stage.Results indicated that:(1)SAVI and EVI were superior to NDVI and RVI as coupling parameters;(2)best prediction accuracy was achieved when SAVI from jointing to booting stage and EVI after booting stage were assimilated;(3)booting to heading was the optimal coupling window if only one phenological stage RS data was available;(4)the sequence RS data from jointing to filling stage was the optimal RS data of multiple stages for assimilation,obtaining the highest prediction accuracy.Model test results demonstrated that the new constructed sequential assimilation technique improved the prediction accuracy of WheatGrow model with RMSE of 0.843,1.202 g.m-2 and 510.68 kg.ha-1 for LAI,LNA and grain yield,respectively.These results would provide significant technical support for growth monitor and yield prediction of regional winter wheat.