Model Optimization And Application Potential Evaluation Of Process-based Ratoon Rice Growth Simulation Model

Ratoon rice is a green and high efficiency production system,has several advantages such as less labour requirement,lower production costs,higher resource use efficiency,and more environmental benefits.In modern agricultural production,large-scale,informationized and intelligentized management practices are replacing the traditional one.The exploration of ratoon rice growth model is very important for realizing the modernization of rice production.Recently,Dr.Tao Li had developed a ratoon rice growth model based on ORYZA v3,named ORYZA＿R.However,the performance of ORYZA＿R in simulating the rice growth and grain yield of main crop（MC）and ratoon crop（RC）is need to be improved and evaluated,and its potential application in assisting with optimizing management regimens of ratoon rice production is need to be revealed.This study made the first attempt to estimate the key parameter values of RC growth by using field data collected from the ratoon rice experiments of varieties,heights of stubbles left from the MC(SH_MLeft)and nitrogen（N）applications in 2015 and 2016.Then,the parameters of development rate,leaf and stem growth,dry matter partitioning,and the dry matter transported from the stubbles to the tillers were calibrated to improve the performance of ORYZA＿R in simulating the growth stage,grain yield,total above-ground biomass,organs biomass and leaf area index（LAI）of MC and RC.After evaluating the model performance,a sensitive analysis of ratoon crop parameters and a scenario analysis of critical management practices of ratoon rice were carried out.Furthermore,to investigate how the variation in the experimental data used for calibration influenced the prediction error of crop model,we organized two groups of scenarios for model calibration and cross prediction with experimental data from different replicates and sowing dates.The main results are as follows:1. The results indicated that the calibrated model could accurately simulate the growth stages of MC and RC with root mean square error ranged from 1. 6 d to 2.6 d and from 2.5 d to 3.4 d,respectively.In terms of yield and biomass simulation,the calibrated model could properly predict the grain yield,total above-ground biomass,stem biomass,panicle biomass and leaf area index of the MC with normalized root mean square error（NRMSE）ranged from 7%to 29%and R²ranged from 0.59 to 0.98.For the simulation of the RC,the best agreement between simulated and observed values was found in grain yield（NRMSE ranged from 9%to 12%,R² ranged from 0.70 to 0.92）,following by the total above-ground biomass,panicle biomass,and stem biomass（NRMSE ranged from 13%to 29%,R² ranged from 0.51 to 0.96）,and the worst agreement were found in green leaf biomass and leaf area index（NRMSE ranged from 36%to 49%,R² ranged from 0.78 to0.84）.After the rigorous calibration,ORYZA＿R model could be satisfactorily applied in simulating the growth stages,grain yield,and panicle and stem biomass for ratoon rice.2. To evaluate the effect of index,variety,and growth variable on the parameter sensitivity analysis,Kendall’s coefficient of concordance and Top-down coefficient of concordance were calculated as measures of agreement between sensitivity rankings derived from methods under consideration.The rankings of the least influential parameters were quite stable among indexes,varieties,and growth variables.However,the influential parameters had disagreement on the rankings at different indexes and varieties.The results seemed to indicate that multiple index,variety,and growth variable were need to identify most influential parameters instead of least influential parameters.Based on that,an optimization method of parameter sensitivity analysis was suggested.3. To distinguishing influential and uninfluential parameters in RC simulation,three indexes and two varieties were under consideration. Our results showed that the most influential parameters of total above-ground biomass,panicle biomass,and grain yield were identified as the initial development stage of RC,development rate,and those parameters related to the production and allocation of dry matter at the panicle initial stage.The least influential parameters were identified as shading tolerance,parameters related to the allocation of dry matter at grain-filling phase,and parameters related to the growth of LAI.These results provided a reference for the ORYZA＿R model calibration of ratoon season.4. Impacts of the key agronomic management practices to the growth of ratoon rice were simulated by three scenarios of SH_MLeft,nitrogen application for promoting the development of regenerated bud(N_RBud),and nitrogen application for promoting the growth of regenerated tiller(N_RTiller).（1）The simulated grain yield of RC across four varieties ranged from 1573.6～2683.3 kg hm^-2 to 3853.3～5402.5 kg hm^-2 when SH_MLeftincreased from 10 cm to 60 cm,and the mean value of the simulations was the highest when the SH_MLeft was 40 cm.（2）The simulated grain yield of RC responded more significantly to application rate than to application time of N_RBud.On average across varieties,the grain yield of RC increased by 1148.2 kg hm^-2 with the increase in N_RBudrates from 0 to 100 kg N hm^-2,with a large discrepancy in simulated value between rates of 0 and 50 kg N hm^-2.（3）The simulated grain yield of RC decreased by 68.1 kg hm^-2 for every 1 d delay of application of N_RTiller,and almost decreased to the same value as that without N_RTiller at 15 d after harvesting.The simulation results of ORYZA＿R model on the effect of agronomic management practices were reasonable,this model has certain application value in assisting with management regimens optimizing of rice ratooning.5. This study evaluated the impact of variation in the experimental data on prediction error. When the variation in observed values increased from 2.6%～5.6%of replicates to6.1%～20.4%of sowing dates,the NRMSE between simulated values and observed values of growth variables increased from 5.3%～14.4%to 9.2%～31.5%,and the contribution of the predictor variance due to variability of observation to the overall prediction error increased from 0.02～0.93 to 0.50～0.99,which even outweighed the contribution of model bias.Our results implied that the variability of the experimental data used for the model parameterization was a major source of model prediction error and its uncertainty.Therefore,controlling the quality of experimental data used for parameterization can reduce prediction error and its uncertainty of crop model.