Location: Agroecosystems Management ResearchTitle: Estimating crop genetic parameters of the DSSAT model with modified PEST software
|MA, HAIJIAO - NORTHWEST A&F UNIVERSITY|
|Malone, Robert - Rob|
|JIANG, TENGCONG - NORTHWEST A&F UNIVERSITY|
|YAO, NING - NORTHWEST A&F UNIVERSITY|
|CHEN, SHANG - NORTHWEST A&F UNIVERSITY|
|SONG, LIBING - NORTHWEST A&F UNIVERSITY|
|FENG, HAO - NORTHWEST A&F UNIVERSITY|
|YU, QIANG - NORTHWEST A&F UNIVERSITY|
|HE, JIANQIANG - NORTHWEST A&F UNIVERSITY|
Submitted to: European Journal of Agronomy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/4/2020
Publication Date: 2/27/2020
Citation: Ma, H., Malone, R.W., Jiang, T., Yao, N., Chen, S., Song, L., Feng, H., Yu, Q., He, J. 2020. Estimating crop genetic parameters of the DSSAT model with modified PEST software. European Journal of Agronomy. 115:126017. https://doi.org/10.1016/j.eja.2020.126017.
Interpretive Summary: The DSSAT (Decision Support System for Agrotechnology Transfer) model is one of the most widely used crop models. Because crop-related input variables (parameters) are difficult to obtain through measurement and have a range of acceptable values, accurate estimation is a prerequisite for crop model application. Methods to estimate model parameters include manual trial and error methods and automatic optimization algorithms. Automatic optimization provides an objective means of automatically estimating these parameters and is increasingly used due to its high efficiency and the resulting accuracy of the optimized model. For example, the GLUE (General Likelihood Uncertainty Estimation) algorithm was previously linked to DSSAT resulting in DSSAT-GLUE. Although the DSSAT-GLUE software package automatically estimates model parameters, it requires many model runs and long run-times to achieve the most accurate crop model simulations. The PEST (Parameter ESTimation) software has been widely used in the automatic estimation of model parameters. A disadvantage of PEST, however, is that it can easily fall into local optima meaning that it can't always identify the set of parameters that results in the best simulations when compared to field observations such as crop yield. The objectives of this research were to (1) develop the DSSAT-PEST software package for automatically estimating parameters in the DSSAT model; (2) reduce the problem of the original PEST optimization algorithm falling into local optima; and (3) compare the performance of the traditional trial-and-error method, DSSAT-GLUE and DSSAT-PEST. Overall when comparing the model output to field observations, the DSSAT-PEST method produced simulations similar to or better than trial and error and DSSAT-GLUE, and DSSAT-PEST average run-times were 65% faster than DSSAT-GLUE. This research will help model developers, model users, and agricultural scientists more efficiently, accurately, and objectively optimize agricultural crop models.
Technical Abstract: Obtaining crop genetic parameters for crop model application quickly and accurately is difficult. We coupled the independent automatic parameter optimization tool PEST (Parameter ESTimation) with the crop growth model DSSAT (Decision Support System for Agrotechnology Transfer) using R language. The DSSAT-PEST package was developed to perform automatic optimization of crop genetic parameters used in the DSSAT model. In addition, the PEST tool was improved to reduce problems associated with local optima and runtime. Finally, the DSSAT-PEST package was used to estimate the genetic coefficients of five crops (maize, soybean, wheat, rice, and cotton) based on the experiments in the DSSAT database. Three methods were compared based on their efficiency and accuracy in crop genetic parameter estimation: the traditional trial-and-error method (default crop genetic parameters in the DSSAT database), DSSAT-general likelihood uncertainty estimation (GLUE) (an existing parameter estimation package in DSSAT), and DSSAT-PEST. Based on statistical analysis (relative root mean square error, fitness, and average absolute relative error (ARE)), the DSSAT-PEST optimization method provided better accuracy and efficiency than the other two methods. For example, the average AREs obtained with the DSSAT-PEST method for the five crops were 12.08%, 6.64%, 18.23%, 4.15% and 19.25%, respectively, which were similar to or better than the results of the DSSAT-GLUE and default methods. Additionally, the DSSAT-PEST average runtime was 65% faster than the DSSAT-GLUE average runtime. In general, the DSSAT-PEST method performed similarly to or better than the traditional trial-and-error method and DSSAT-GLUE in terms of both efficiency and accuracy, which may result in wider use of the DSSAT model.