基于APSIM模型的旱地小麦叶面积指数相关参数敏感性分析及优化

Sensitivity analysis and optimization of leaf area index related parameters of dryland wheat based on APSIM model

为解决作物模型参数率定过程中参数众多导致的敏感参数定位迟缓和调参效率低的问题,本研究运用敏感性分析和智能优化算法相结合的方法对作物模型参数进行调整,以甘肃省定西市安定区李家堡镇麻子川村(2002—2004年)和凤翔镇安家沟村(2015—2017年)大田旱地小麦试验数据(叶面积指数)为参照,利用扩展傅里叶幅度检验法(EFAST),对APSIM-Wheat旱地小麦叶片生长子模型的23个参数进行敏感性分析,得到对模型结果较敏感的部分参数,然后利用粒子群优化算法对部分敏感参数进行优化。结果表明:1)影响旱地小麦叶片生长最敏感的参数依次为叶面积指数为0时最大比叶面积、叶片生长的氮限制因子、出苗到拔节积温、消光系数、拔节到开花积温、蒸腾效率系数;2)旱地小麦叶片生长子模型的参数优化结果:叶面积指数为0时最大比叶面积为26652 mm^(2)∙g^(−1),叶片生长的氮限制因子为0.96,出苗到拔节积温为382℃·d,消光系数为0.44,拔节到开花积温为542℃·d,蒸腾效率系数为0.0056;3)上述参数优化后的叶面积指数实测值与模拟值之间的均方根误差平均值从参数优化前的0.080减小到0.042,归一化均方根误差平均值从11.54%减小到6.11%,模型有效性指数平均值从0.962增加到0.988,优化后叶面积指数的模拟更好。该方法相对于传统的手工试错法,避免了优化参数的不确定性,实现参数自动率定,提高模型参数的率定效率,有利于模型快速地本地化应用,并指导农业生产。本研究方法也对APSIMWheat模型中其他作物模块的参数调整优化具有指导意义。

Crop growth model parameterization is characterized by a large number of parameters and the low efficiency of parameterization.To determine the rate of crop model parameters quickly and efficiently,the promotion of rapid application of crop models in localization is required.In this study,we used a combination of sensitivity analysis and intelligent optimization algorithm to adjust the parameters of the crop model.We used the experimental data(leaf area index)of dryland wheat in large fields in Mazichuan Village,Lijiabao Town from 2002 to 2004,and Anjiagou Village,Fengxiang Town from 2015 to 2017 in Anding District,Dingxi City,Gansu Province as references.Using the extended Fourier amplitude sensitivity test method,a sensitivity analysis of 23 parameters of the APSIM-Wheat dryland wheat leaf growth sub-model was performed using SimLab software,and the sensitivity coefficients of each parameter to the model results were obtained.On this basis,the parameters with a larger first-order sensitivity index and global sensitivity index were selected as the optimization parameters,and R programming was used to construct the algorithmic fitness function,implement the particle swarm optimization algorithm,and run the APSIM-Wheat model to optimize the parameters automatically.We performed this to ensure fast and effective determination of the model parameters.The results showed that:1)the six parameters most sensitive to the leaf growth model of dryland wheat were,in descending order,maximum specific leaf area at a leaf area index of 0,nitrogen limiting factors in leaf growth,accumulated temperature from seedling to jointing,extinction coefficient,accumulated temperature from jointing to flowering,and transpiration efficiency coefficient;2)optimization of the parameters in the leaf growth submodel for dryland wheat resulted in a maximum specific at a leaf area index of 0 was 26652 mm2∙g−1,a nitrogen limiting factor in leaf growth was 0.96,an accumulated temperature from seedling to jointing was 382℃·d,an extinction coefficient was 0.44,an accumulated temperature from jointing to flowering was 542℃·d,and a transpiration efficiency coefficient was 0.0056;3)after the optimization of the aforementioned parameters,the mean value of the root mean square error between the measured and simulated values of the leaf area index decreased from 0.080 to 0.042.The mean value of the normalized root mean square error decreased from 11.54%to 6.11%,and the mean value of the model validity index increased from 0.962 to 0.988,indicating that the simulation of the leaf area index was better after the optimization.When compared with the traditional manual trial-and-error method,this method avoids the uncertainty of the optimization parameters,quickly and efficiently identifies the important parameters of the model,realizes automatic parameter rate fixing,improves the efficiency of model parameter rate fixing,alleviates the problem of many parameters and low efficiency in the process of model rate fixing,and finally,enables the model to be applied locally faster so that it can better guide the agricultural production.The methodology of this study is also instructive for the parameter tuning optimization of other crop modules in the APSIM-Wheat model.