| In our country’s irrigation districts,there is a widespread phenomenon of irrational irrigation and excessive fertilizer application,which causes serious waste of water resources and environmental pollution.Therefore,conducting research on optimizing irrigation and fertilization shedules at the regional scale is of great significance.Crop models can describe the response of crop growth and development to environmental and management factors from a mechanistic perspective and are widely used in water and fertilizer management decision-making.However,when crop model parameters determined at the point scale are extended to the regional scale,there is often a problem of insufficient representativeness,which results in poor adaptability of the determined irrigation and fertilization shedules at the regional scale.Remote sensing can quickly obtain information on crop growth over large areas.Combining crop models with remote sensing methods can improve the simulation accuracy of crop models at large scales and provide new ways to use crop models to develop irrigation and fertilization shedules at the regional scale.Therefore,this thesis takes summer maize in the Fenwei Plain as an example and comprehensively applies methods such as assimilation of remote sensing data and crop models,spatial clustering,and multi-objective optimization.The research was conducted on issues such as sensitivity analysis and calibration of crop model parameters under different water and fertilizer treatment conditions,remote sensing estimation of regional leaf area index(LAI),remote sensing inversion of crop model regional cultivar parameters based on data assimilation and spatial clustering,and optimization of irrigation and fertilization shedules.The main research content and conclusions are as follows:(1)In order to address the issue of unclear parameter sensitivity of the CERES-Maize model under different water and fertilizer treatment conditions,a global sensitivity analysis method was employed to analyze the sensitivity of parameters in the crop model.A set of sensitive parameters was determined,and field observation data were used to calibrate and validate the sensitive parameters.The simulation accuracy of the model based on the determined set of sensitive parameters was compared with that of the model using commonly chosen sensitive parameters,demonstrating the rationality of the determined set of sensitive parameters.The results showed the following:1)there are differences in the sensitivity of crop model parameters under fully irrigated and deficit irrigated conditions,manifested by a weakening of the influence of crop parameters on daily aboveground biomass,daily LAI,daily evapotranspiration,yield,and maximum aboveground biomass with decreasing irrigation water,while the influence of soil parameters increases.Nitrogen fertilizer factor has little effect on the sensitivity of parameters.2)The sensitive parameter set under water and fertilizer deficit conditions is determined as:photoperiod sensitivity parameters(P2),thermal time from silking to physiological maturity(P5),kernel filling rate during the linear grain filling stage and under optimum conditions(G3),soil fertility factor(SLPF),soil drainage rate(SLDR)and soil runoff curve number(SLRO).The model simulation accuracy of the sensitive parameter set for LAI,yield,and soil water content is compared with the parameter set commonly selected by users,which includes thermal time from emergence to the end of juvenile stage(P1),P2,P5,maximum possible number of kernels per plant(G2),G3,and the interval in thermal time between successive leaf tip appearances(PHINT).Compared with the parameter set commonly used by users,the normalized root mean square error(n RMSE)of simulating LAI,yield,and soil water content was reduced from 0.08 to 0.07,0.04 to 0.03,and 0.09 to0.07,respectively.Therefore,the sensitive parameter set determined in this thesis can more accurately simulate the growth of summer maize under different water and fertilizer conditions in the Fenwei Plain,providing a theoretical basis for the calibration and simplification of large-scale crop models.(2)In response to the problems of cloudy and rainy weather during the summer maize growing season and the fragmented land plots in the Fenwei Plain,which lack high spatiotemporal resolution LAI,high spatiotemporal resolution Sentinel-2 optical remote sensing data was used to extract the summer maize planting areas.Using Sentinel-1microwave remote sensing backscattering information and LAI obtained through regional multi-point sampling,the accuracy of two LAI estimation models,support vector regression(SVR)and random forest(RF),were analyzed and compared to determine the optimal model for regional remote sensing LAI estimation.The results show that:1)the combination of RF classifier and classification features extracted from Sentinel-2 optical satellite remote sensing data can extract fragmented summer maize planting areas in the Fenwei Plain,with an overall classification accuracy of 95.52%and a Kappa coefficient of 0.91.The extraction results can provide basic data for further estimating the LAI of summer maize in the region.2)The microwave backscatter coefficient and incident angle under the vertical emission-vertical reception and vertical emission-horizontal reception polarization modes have a good correlation with LAI.Considering the temporal characteristics of LAI,two LAI estimation models based on SVR and RF algorithms were constructed based on microwave backscatter coefficient,incident angle,and the date of microwave remote sensing data acquisition.3)The data of 2021 were divided into training and testing sets,and the corresponding n RMSE of the two estimation models based on SVR and RF algorithms were 0.17,0.20 and 0.20,0.25,respectively.Further testing of the trained models using data from 2020 showed that the n RMSE of the two LAI estimation models were 0.22 and 0.26,respectively,indicating that the LAI estimation model based on SVR algorithm is better than that based on RF algorithm,with higher accuracy and inter-annual stability.4)The LAI of summer maize in the Fenwei Plain was estimated using the SVR-based model,which showed that there is a large spatial difference in the LAI of summer corn in the region,suggesting that using single-point LAI calibration to parameterize crop models may result in large errors.Compared with commonly used MODIS and Landsat satellites,the Sentinel satellite has higher temporal and spatial resolution,and the microwave information can better avoid cloud and rain interference.Therefore,the remote sensing data used in this thesis can be well applied to the spatial distribution remote sensing measurement of crop LAI in fragmented planting areas and cloudy/rainy summer seasons in most regions of China.(3)Existing research has not considered the spatial differentiation of crop cultivars when formulating irrigation schedules at the regional scale.Taking the Xianyang area as an example,the LAI obtained from remote sensing was assimilated into a crop model to invert the representative parameters of different zones,thereby constructing a regional-scale crop model that considers the differentiation of cultivar parameters.Based on this,the irrigation shedule for summer maize was optimized using this regional-scale crop model.By comparing the water-saving and yield-increasing effects of the optimized irrigation shedule with the local current irrigation shedule,the scientificity of using remote sensing data assimilation to determine representative clutivar parameters for formulating regional-scale irrigation shedules was demonstrated.The results show that:1)the accuracy of simulating regional LAI and yield by the crop model calibrated based on plot-scale experimental data was poor.Compared with treating the Xianyang area as a whole,assimilating remote sensing LAI and inverting clutivar parameters of the CERES-Maize model by zones can significantly improve the ability of the crop model to simulate regional LAI and yield.The R~2 of LAI increased from 0.62 to 0.73,and the n RMSE decreased from 0.24 to 0.20.The R~2 of yield increased from 0.69 to 0.77,and the n RMSE decreased from 0.13 to 0.11.Therefore,using remote sensing data assimilation to invert cultivar parameters by zones can better reflect the growth characteristics of summer maize in the region.2)Compared with the local current irrigation shedule,the crop model considering the spatial differentiation of crop cultivar determined an irrigation schedule that saved 17.7,15.2,and 42.7 mm of water in dry,normal,and wet years,respectively,and increased the regional yield by an average of 153.52,228.00,and 165.10 kg/ha,respectively.The average water use efficiency of the region increased by 0.035,0.042,and 0.030 kg/m~3.This demonstrates the scientificity and rationality of using remote sensing data assimilation to determine representative cultivar parameters for formulating regional-scale irrigation shedules.(4)In response to the difficulty in considering the spatial differentiation of crop cultivars and sowing dates when formulating large-scale irrigation and fertilization shedules for crops,taking the Fenwei Plain as an example,the spatial clustering algorithm was integrated into the assimilation framework of microwave remote sensing data and crop models.Two parameter inversion methods were proposed based on the clustering features of cultivar parameters and sowing dates or cultivar parameters,sowing dates,and geographic locations.The corresponding parameter CAs and CAGCs were obtained,and a regional summer maize irrigation and fertilization optimization method that integrates remote sensing and crop models was constructed by combining a regional-scale crop model and a multi-objective optimization algorithm based on these two parameters.Meanwhile,the entire Fenwei Plain was taken as one region and divided into three sub-regions according to geographic location.Representative parameters for the entire study area(RRs)and for each sub-region(SRRs)were obtained through data assimilation,and corresponding irrigation and fertilization shedules were formulated.The water-saving,fertilizer-saving,and yield-increasing effects of four methods were compared.The results showed that:1)Based on RRs parameters,the mean n RMSE of the model simulation of regional LAI and yield in 2020 and 2021 were 0.47 and 0.26,respectively;based on SRRs parameters,the corresponding mean n RMSE of LAI and yield were 0.45 and 0.23,respectively;based on CAs parameters,the corresponding mean n RMSE of LAI and yield were 0.23 and 0.17,respectively;based on CAGCs parameters,the corresponding mean n RMSE of LAI and yield were 0.17 and 0.11,respectively.The CAGCs parameter produced the best results in simulating regional LAI and yield,indicating that the CAGCs parameter can not only reflect the spatial information of cultivars and sowing dates themselves,but also reflect the spatial differences in cultivars and sowing dates due to different geographic locations.Therefore,using this method to invert model parameters is scientific.2)In order to demonstrate the rationality of this method in determining regional irrigation and fertilization shedule,the Fenwei Plain was divided into 3×3 km~2 units,and the remote sensing LAI was assimilated into the crop model to determine the cultivar parameters and sowing dates of each unit.Based on this,combined with soil and meteorological data for each unit,the optimal irrigation and fertilization shedule for each unit in a typical hydrological year was determined using multi-objective optimization algorithm and computer parallel technology,and the corresponding irrigation and fertilization shedule were determined based on the grouped crop cultivar parameters and sowing dates determined by the four methods.3)Analyzed the differences between the irrigation and fertilization shedules optimized based on RRs,SRRs,CAs,and CAGCAs and the unit-by-unit optimization shedule.In dry,normal,and wet years,the differences in average irrigation and nitrogen application amounts between RRs-based optimization and unit-by-unit optimization were 7.81,15.08,-33.02 mm and-37.74,0.85,-9.10 kg/ha,respectively.The corresponding differences for SRRs were 18.06,12.22,59.82 mm and-33.64,14.51,-9.1 kg/ha,respectively.For CAs,the differences were 35.39,36.76,-49.48 mm and-6.08,3.16,7.38 kg/ha,respectively.For CAGCs,the differences were6.56,3.06,13.41 mm and 2.42,2.34,-3.54 kg/ha,respectively.In dry,normal,and wet years,the root mean square error(RMSE)of yield obtained based on RRs optimization and unit-by-unit optimization were 2570.63,1429.76,and 2373.51 kg/ha,respectively,the RMSEs of water use efficiency(WUE)were 0.40,0.24,and 0.50 kg/m~3,and the RMSEs of nitrogen use efficiency(NUE)were 4.54,2.78,and 3.96 kg/kg.For SRRs,the RMSEs of yield were2113.92,1355.02,and 2292.92 kg/ha,the RMSEs of WUE were 0.29,0.21,and 0.47 kg/m~3,and the RMSEs of NUE were 2.88,2.06,and 4.28 kg/kg.For CAs,the RMSEs of yield were921.78,1189.07,and 1433.65 kg/ha,the RMSEs of WUE were 0.13,0.21,and 0.29 kg/m~3,and the RMSEs of NUE were 1.71,2.09,and 2.62 kg/kg.For CAGCs,the RMSEs of yield were 736.52,756.99,and 882.19 kg/ha,the RMSEs of WUE were 0.14,0.15,and 0.24 kg/m~3,and the RMSEs of NUE were 1.44,1.75,and 3.58 kg/kg.4)Comparing the results,it was found that the irrigation and fertilization shedule based on CAGCs was the closest to the unit optimization irrigation and fertilization shedule,followed by CAs.Therefore,the parameter inversion method based on crop cultivar,sowing date,and geographic location clustering can improve the ability of crop model to develop regional-scale irrigation and fertilization shedules.This further confirms the scientific and rational nature of the method of determining representative cultivar parameters for each region through data assimilation fusion and spatial clustering algorithm and developing regional-scale irrigation and fertilization shedules.In summary,this thesis proposes a method for estimating regional LAI based on microwave remote sensing backscatter information and support vector regression model.The method has strong applicability and robustness,and can estimate regional LAI of summer maize under cloudy and fragmented field conditions,providing LAI data that meet the requirements of spatial resolution and accuracy for inverting crop model parameters in large regions.The framework of remote sensing inversion of crop model parameters and sowing dates and regional irrigation and fertilization shedule optimization developed in this thesis innovatively integrates the spatial clustering algorithm into the assimilation of remote sensing-acquired LAI and crop model,proposes a method to obtain crop model parameters and sowing dates in a large scale,and integrates the multi-objective optimization method of irrigation and fertilization shedule,which initially solves the problem of determining irrigation and fertilization shedule under the conditions of spatial variation of crop cultivars and sowing dates in a regional scale.However,the study did not consider the available water resources for irrigation,which may result in insufficient irrigation and fertilization.In future optimization systems,considering the constraint of available water resources may better guide agricultural production practices.The applicability of this framework to crops in other ecological regions should also be tested in the future. |
PDF Full Text Request