Enhancing crop model parameter estimation across computing environments: Utilizing the GLUE method and parallel computing for determining genetic coefficients

Estimating genetic coefficients is essential to accurately simulate crop development and growth for modeling studies but has been challenging due to lack of robust and fast procedures. While there are several optimization techniques, the Generalized Likelihood Uncertainty Estimation (GLUE) is a Bayesian method that is popular among the modeling community due to its application for sensitivity and uncertainty analysis and capability to explore the global parameter space. However, the time required for its search method to estimate the optimal parameter set is a significant constraint and limitation. Parallel computing has emerged as a solution to boost the efficiency of genetic coefficient calibration using GLUE. In this study, we introduce a new system that leverages parallel computing for calibrating genetic inputs for crop growth models within the Cropping System Model (CSM) of the Decision Support System for Agrotechnology Transfer (DSSAT). Designed and tested for both conventional and High-Performance Computing (HPC) environments, the Generalized Likelihood Uncertainty Estimation Parallelized (GLUEP) is available for most crops that are simulated with DSSAT-CSM and provides a user-friendly graphical interface within the DSSAT software. It accelerates the genetic-specific parameter calibration process and adds new functionality that enables users to optimize intrinsic model parameters, which were previously unavailable for calibration purposes. Four case studies using cultivars for wheat, maize, soybean, and potato showcase the application of GLUEP. We also conducted a comparison with DSSAT-GLUE and evaluated the performance gains of GLUEP for multiple operational systems, including Windows, MacOS, and Linux, as well as under conventional and HPC environments. The multi-core processing results indicate performance improvements across all computer systems that were analyzed. The comparison between the sequential processing of DSSAT-GLUE and the parallel processing of GLUEP indicates a reduction in execution time ranging from 87.4% to 95.4%. These results highlight the GLUEP capabilities in streamlining the calibration process, enabling more efficient and accurate predictions for crop growth modeling studies.