Winter Canola Growth Monitoring And Yield Simulation Based On Low-altitude UAV Remote Sensing

Accurate acquisition of winter rapeseed phenology information and yield prediction are crucial for the formulation of field management practices such as irrigation,fertilization and harvesting.Remote sensing technology has the characteristics of high timeliness and wide monitoring range in obtaining crop phenology dynamics and yield estimation,but it lacks the mechanism description of crop growth process and has poor universality.The crop model,as a process-oriented dynamic model,accurately predict crop growth dynamics and final yield.However,when the model is extended from single point to regional scale,with the increase of spatial scale,the inhomogeneity of spatial distribution of surface environment and meteorological factors makes the initial parameters and crop state variables of the model change,and the basic parameters are difficult to obtain,which reduces the applicability of the model.Therefore,the assimilation yield estimation model is constructed by coupling the crop model with the remote sensing data to realize the complementary advantages of the two.In this study,winter rapeseed was used as the research object,and two consecutive field experiments in the growing season were carried out.The construction and implementation of the winter rapeseed regional yield estimation system were systematically studied from the aspects of exploring the dynamic response mechanism of winter rapeseed growth under different sowing and density scenarios,establishing the phenological dynamic remote sensing monitoring model,simulating the growth mechanism and process law of winter rapeseed,and the collaborative application of data assimilation algorithms.The main research contents and conclusions are as follows:(1)Based on the three vegetation indexes,the growth dynamic curves of winter rapeseed under different sowing and density conditions were established.Combined with local meteorological data,the response mechanism of winter rapeseed phenology to sowing date adjustment was clarified.With the adjustment of sowing date,the growth days of winter rapeseed fluctuated within 185-230 days;An improved shape model method was proposed to monitor the four phenological stages of winter rapeseed sowing-flowering period,and the dynamic interpretation ability of shape models established by different VIs and mathematical models was evaluated.The results showed that the model combination of AGF-CIred-edge had the highest estimation ability for phenological stages(seeding stage:RMSE=2.8 days,overwintering stage:RMSE=3.7 days,bolting stage:RMSE=5.4 days,flowering stage:RMSE=3.2 days,overall estimation:RMSE=3 days).(2)Combined with field experiment data and phenological remote sensing monitoring data,the localization parameter calibration of AquaCrop model was completed,and the ability of the model to simulate the phenological dynamics,CC,biomass and yield of winter rapeseed in multiple scenarios was evaluated.The simulation results showed that CC simulation achieved high accuracy under different sowing dates(RMSE:7-22%,EF:0.54-0.95,R2:0.700.97).However,with the decrease of sowing density,the accuracy of CC simulation began to decrease.The simulated value of crop transpiration was close to the field measured value,with R2 and RMSE of 0.79 and 26 mm,respectively.In the simulation of biomass and yield,the medium density and high density treatments achieved higher simulation accuracy.However,the ability to explain the growth of winter rapeseed in low-density scenarios is insufficient,and the model overestimates the production potential of winter rapeseed in this scenario by about 40%.Studies have shown that the AquaCrop model has great potential in the formulation of rape cultivation measures and sowing schemes.However,the ability of phenological simulation,the underestimation of early canopy and the overestimation of production potential under low density conditions need further study.(3)Based on the validation results of the AquaCrop model,the canopy composition parameters(CGC,CDC,Mcc,Mat)and yield composition parameters(STbio,WP,HI,Kcb,FLO),two sensitive parameters related to crop yield,were established as assimilation parameters.The winter rapeseed CC data obtained from remote sensing data was used as the assimilation variable,and the assimilation yield estimation model was established by embedding the SCE-UA algorithm.The verification results based on the assimilation model show that:for the simulation of phenological days,after introducing Mat and FLO two phenological parameters,the simulation accuracy of the two stages of Time to flowering’ and Time to maturity’ is improved by 6 days and 9 days respectively;After assimilation,the model effectively improved the simulation ability of the model for the early canopy of rapeseed.The CC simulation accuracy of the model for winter rapeseed increased from RMSE:7-22%to RMSE:7-14%.In the simulation of biomass,the assimilation model corrected the overestimation of biomass under low density treatment,but the accuracy of biomass estimation in the early stage of rape growth decreased.Due to the calibration of phenological days by the assimilation model,the accuracy of yield simulation has been significantly improved.Compared with Pe:0-35.3%before assimilation,the yield estimation accuracy was improved to Pe:2.7-15.7%after model assimilation.Compared with the estimation results of the AquaCrop model before assimilation(RMSE:26 mm,EF:0.69,R2:0.79),the assimilation model achieved higher estimation accuracy(RMSE:19 mm,EF:0.82,R2:0.84).