| Crop model is a powerful and effective tool of agro-environmental science,and it has become the basis and core of digital agriculture and modern agriculture research since the rising of agricultural information technology.Also,crop model is a complement for study of water-saving experiments,in which the model can be used to simulate the crop growth responding to water,nutrients,and mineral components in soil cross multi-environments and multi-scenarios,without the limitations encountered in field experiments.However,crop model has a complex structure and a large number of parameters,and the effects of each parameter on model outputs are usually not known to us.Furthermore,the values of some parameters have great uncertainty due to variability of climate,environment,and crop variety.Therefore,model calibration is always a difficult problem and model application is limited.In this study,parameter sensitivity and uncertainty analysis was fully conducted for ORYZA_V3 model to pursue following objectives:(?)to distinguish influential and uninfluential parameters in the model,(?)to explore spatial-temporal variability of parameter sensitivity and influencing factors,(?)to explore effective method to quantify uncertainty and to estimate parameter values for crop model.The main contents and results were summarized as follows:(1)Latin Hypercube Sampling(LHS)method was used to obtain parameter sets for five kinds of parameter variation ranges.In order to analyze the effects of parameter variation ranges on model output uncertainty,daily coefficients of variation(CV)and 90%confidence intervals were calculated for four output variables in ORYZA_V3 model.Then parameter sensitivity analysis was conducted by using regression-based method,and TDCC coefficient was used to evaluate the consistency of parameter sensitivity among different parameter variation ranges.The results indicated that,when a small or big parameter variance range was set for parameter sensitivity and uncertainty analysis,some parameter lost its influence on specific model output,which may mislead sensitivity analysis results.Based on that,an appropriate parameter variation range was suggested for sensitivity and uncertainty analysis of crop model.(2)Sensitivity analysis was conducted by using Extended FAST method in three years of contrasting weather types,which were chosen based on the annual accumulated temperature in late-season rice growing duration with 95%,50%,and 5%probability of exceedance.The parameter sensitivity for different model outputs of each stage in rice growing period was quantified,and the CV values and mean values of model outputs of each stage were also calculated.We found that influential parameters had unequal effects on different outputs,and they also had different effects in four stages,which presented significant temporal characteristic.Additionally,the sensitivity results were slightly different among three typical years,and the sensitivity indices of those yield-related parameters,the CV values and mean values of model outputs were strongly impacted by cold or hot weather types.(3)Eighteen typical experiment sites were chosen from sixteen rice cropping sub-regions in China.For each site,parameter sensitivity analysis of ORYZA_V3 model was conducted for 30a(1986-2015)by using Extended FAST method.Sensitivity indices of each parameter for model outputs in four stages were calculated,and standard variance of which was also analyzed.Furthermore,the variation of sensitivity indices were analyzed cross over different rice cropping sub-region and rice cropping types,as well as influencing factors of that.It indicated that the sensitivity indices of model parameters varied significantly in spatial,which were mainly affected by climate type and rice cropping type.However,the parameter sensitivity was not changed with altitude or longitude and latitude monotonously.There was no obvious difference between the sensitivity results of two sites in the same rice cropping sub-region.For WST2.0 and WSO2.0,the sensitivity indices varied significantly among years in all sites.Additionally,parameter sensitivity was significantly related to daily maximum temperature,daily minimum temperature,and accumulated temperature,while the sensitivity of some parameters for LAI2.0 were related to accumulated sun-hours too.(4)GLUE method was applied in model uncertainty analysis.Eight kinds of likelihood functions were presented by considering main factors that result in model uncertainty.Then effective functions were chosen by comparing their effects on parameter post distribution.And daily 90%confidence interval of model outputs in 2012-2013,which were simulated with the post parameter distribution obtained by the two kinds of likelihood functions.By evaluating the effectiveness of 90%confidence intervals,an appropriate likelihood function was suggested for crop model uncertainty analysis using GLUE method.(5)Trial and error,GA,GLUE,and PEST method were used to estimate the influential parameters against to biomass measurements of late-season rice in 2012 obtained in Jiangxi Irrigation Experiment Station.Then,the results were validated by measurements obtained in 2013,and the effectiveness of these methods was compared.We found that the parameter values estimated by the four methods were obviously different,which revealed 'equifinality' in crop model.GLUE method was presented perfectly in parameter estimation,and it calculated the post distribution of parameter,which had a special advantage in parameter estimation.PEST was the most efficient one,and the results based on which was similar to GLUE.While the simulations based on trial and error method was the worst due to effects of model user's subjectivity.GA method had low efficiency and the iteration result was vulnerable to initial values of parameter,thus it was not good as GLUE and PEST method for parameter estimation.In addition,some parameters still had strong uncertainty after calibration,which suggested that crop model should be calibrated in several process or step by step. |
PDF Full Text Request