Abstract:
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 26 652 mm
2∙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.