基于APSIM模型旱地小麦叶面积指数相关参数的优化

【目的】 模型的有效应用依赖于参数的快速、准确估算。本研究拟解决作物生长模型参数本土化率定过程中运算量大、耗时长、精度低、效率低的问题。【方法】 依据甘肃省定西市安定区2个试验点（李家堡镇麻子川村和凤翔镇安家沟村）,多年（2002—2005年和2015—2017年）大田试验数据以及定西市安定区1971—2017年气象资料,利用混合蛙跳算法智能的迭代搜索原理,对APSIM模型旱地小麦叶面积指数相关参数进行了优化,并采用相关性分析方法对模型校正结果进行检验。【结果】 利用青蛙群体即相对独立又合作协调的子群内局部深度搜索与子群间全局信息交流生物进化学习策略,有效提高了运算的速度,实现了对APSIM模型中与旱地小麦叶面积指数相关参数的快速、准确估算。相关参数主要包括：主茎上节出现所需的热时间间隔、小麦出苗后初始化的节数、小麦出苗后初始化的叶片数、小麦出苗后初始化的叶面积指数、某日正在生长的节数和最大比叶面积。分别使用穷举试错法所得参数值和混合蛙跳算法所得参数值模拟叶面积指数,参数优化后,叶面积指数模拟值和实测值之间的RMSE（root mean square error）平均值由0.069降低到0.027,NRMSE（normalized root mean square error）平均值由8.09%降低到4.56%,M_E（model effective index）平均值由0.979提高到0.993。【结论】 相对于参数率定常用穷举试错法,混合蛙跳算法具有自发学习特征的智能迭代行为,实现了参数的自动率定,提高了效率。基于该算法进行APSIM模型旱地小麦叶面积指数相关参数的优化,使得模型对叶面积指数的模拟精度显著提高,证明该算法的使用对作物生理生态系统复杂模型的校正效果良好,为改善模型参数率定过程存在的运算量大、耗时长、精度低、效率低的缺点提供了一种行之有效的方法。

Parameter Optimization for the Simulation of Leaf Area Index of Dryland Wheat with the APSIM Model

【Objective】The effective application of the model depends on the fast and accurate estimation of parameters. The current problems in the calibration of crop growth model parameters include large data volume, long time consumption, lack of precision, and low efficiency. The study tried to solve the problems. 【Method】Based on the field experimental data of two experimental sites (Mazichuan village, Lijiabaotown and Anjiagou village, Fengxiangtown) in Andingdistrict, Dingxi city in multiple years (2002-2005 and 2015-2017) and the meteorological data in Andingdistrict, Dingxicity from 1971 to 2017, the parameters related to dryland wheat leaf area index (LAI) in the APSIM (agricultural production systems simulator) model were optimized with the intelligent iteration search principle of shuffled frog leaping algorithm (SFLA) and tested by the correlation analysis method. 【Result】The biological evolution learning strategy of local depth search within sub-group and global information communication between sub-group in frog population, which was relatively independent and coordinated, was used to effectively improve the speed of calculation and realize the fast and accurate estimation of the parameters related to dryland wheat LAI in the APSIM model. The related parameters mainly included: The required thermal time interval for node appearance on the main stem, the initial node number at emergence, the initial leaf number at emergence, the initial leaf area index at emergence, the growing node number, and the maximum specific leaf area. LAI was respectively simulated by using the parameters based on the trial and error method and based on SFLA. After parameter optimization, the root mean square error (RMSE) between simulated and measured wheat LAI reduced from 0.069 to 0.027, the normalized root mean square error (NRMSE) decreased from 8.09% to 4.56%, and the model effective index (M_E) increased from 0.979 to 0.993. 【Conclusion】Compared with the trial and error method, which was usually used in the calibration of APSIM model, the intelligent iterative behavior with spontaneous learning characteristics based on the SFLA could realize automatic calibration of the parameters and improve the efficiency. The parameters estimated based on the SFLA could remarkably improve the simulation accuracy of wheat LAI. The application of SFLA was effective in calibrating crop models involving complex eco-physiological processes, and it could provide an effective parameter optimization method for improving the disadvantages in the model parameter calibration process include large data volume, long time consumption, lack of precision, and low efficiency.