Monitoring And Modeling Winter Wheat Growth By Integrating The Hyperspectral Data And SAFY-FAO Crop Model

Accurately obtains crop growth states,grain yield,and water demand information are important for field management,food storage and trading,and agricultural irrigation decisions.Assimilating remote sensing data and crop model can get the advantages of both method and eliminate some drawbacks.One the one hand,it can provide the“true values”of crop states to correct the simulation deviation,and can effectively reduce the difficulty of obtaining initial growth conditions and model parameters at the regional scale.Besides,the assimilation system contains the intrinsic mechanism in the process of crop growth and development.To date,it is an important development trend for monitoring crop growth state.Based on the different irrigation scenario experiments during two consecutive growing seasons of winter wheat,our study comprehensively analyzed and evaluated the retrieval capability of the different hyperspectral data process methods on the wheat leaf area of winter index,such as hyperspectral characteristic parameters,vegetation index,partial least squares regression and two machine learning methods.The optical leaf area index estimation model was established to provide basis for assimilation with SAFY-FAO model.The water stress response process of original model was improved,and a global sensitivity analysis of model key parameters on the variation of leaf area index was carried out using the extend Fourier amplitude method?EFAST?.The parameters were selected to be optimized based on the sensitivity analysis results.The parameters are optimized and adjusted by comparing crop model simulated and canopy reflectance spectral retrieved using SCE-UA?Shuffled Complex Evolution?optimization algorithm.Moreover,the assimilation system was established when all parameters determined.Finally,the accuracy,reliability and robustness of assimilation system for simulating leaf area index,dry aboveground mass,grain yield,evapotranspiration and soil moisture were validated and evaluated under different irrigation scenarios using in situ measurement data over two winter wheat growing seasons.The main achievements of our study were as follows:?1?The performances of 23 kinds of hyperspectral characteristic parameters,12 kinds of vegetation indices,partial least squares regression,artificial neural network and support vector machine for estimating leaf area index were systematically evaluated.Results indicated that the red band reflectance minimization(?_rb)has the highest accuracy for estimating and predicting leaf area index among hyperspectral characteristic parameters,with the RMSE of0.744;the higher prediction accuracy and lower sensitivity at high leaf area index was observed on the vegetation index of MTVI2,with the RMSE of 0.683;before using partial least squares regression,the numbers of principal components should be determined.The ten principal components were settled by cross-validation,and then the PLS was used to estimate LAI.The performance of LAI estimation showed good adequacy,with the RMSE of 0.450,respectively.A agreeable estimation accuracy?RMSE=0.476?was obtained when using a three layer neural network with nine neurons in hidden layer.The support vector regression machine based on Gaussian kernel function has a consistent result with neural network method,with the RMSE of 0.499.In summary,after comparing the performance of different methods,the partial least square regression was the optimal method for estimating LAI of winter wheat.?2?The study developed a new non-linear water stress process equation to replace the simple linear stress equation in the original SAFY-FAO model,and introduced three water stress parameters to enable the water stress response sensitivity could be adjusted under different crops and cultivars.The global sensitivity analysis of eight parameters of Pl_a?Pl_b?S_TT?R_s?LUE?p_u?p_l and f,related to the crop growth and development,for five leaf area index and dry matter variables were carried out.The parameters of Pl_a and Pl_b had higher total sensitivity index for the variables of maximum LAI,date of maximum LAI,means of LAI during the growth period,and the maximum accumulated dry matter.After evaluating the sensitivity and analyzing the function in model,five parameters of Pl_a?Pl_b?S_TT?R_s and LUE were decided to be optimized in assimilation process.?3?The LAI was selected as the target in model assimilation.Five model parameters were adjusted and by the SCE-UA optimization algorithm based on global minimization of the cost function.The results indicated that model parameters could be optimized,but solutions were not unique.Therefore,the uncertainty of the simulation due to the local optimal solution and non-unique could be effectively avoided by analyzing the distributions and frequencies of lots of repetition.?4?The dynamic of winter wheat leaf area index,dry aboveground mass,grain yield,evapotranspiration and soil moisture of nine irrigation scenarios during two winter wheat growing seasons were successfully simulated based on assimilation coupling system.The accuracy,reliability and robustness for each variable were validated and were evaluated using in situ measurement data.The results showed that excellent accuracy were obtained for simulating the dynamic change of leaf area index under the middle to later growth period water deficit scenarios?RRMSE<10%?.The dry matter simulation accuracy in the severe water deficit scenario were inadequate,with the RRMSE value was greater than 30%;For the winter wheat grain yield simulation,excellent results were obtained in all irrigation scenarios for both growing seasons,with a good regression between in situ measured and estimated yield(RMSE=0.48 t?ha^-1,RRMSE=9.5%,MRE=8.4%).The simulated soil moisture of the surface 20 cm soil layer was in good agreement with measured during the two growing seasons.?RRMSE<20%?;The simulation results for soil water storage in 1-m soil layer depth have good overall accuracy?RRMSE<20%?.The difference between measured and simulated soil moisture was emerged after jointing;The simulated accumulated evapotranspiration in two growing season were higher than measured values,and the simulation accuracy was adequacy?RMSE=43.4 mm,RRMSE=17.1%,MRE=17.9%?,which could be applied to the estimation and prediction of regional winter wheat water demand.