GLUEOS: A high performance computing system based on the orchestration of containers for the GLUE parameter calibration of a crop growth model

Authors: HYUN, SHINWOO - Seoul National University; PARK, JIN YU - Seoul National University; KIM, JUNHWAN - National Institute Of Crop Science - Korea; Fleisher, David; KIM, KWANG SOO - Seoul National University

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/21/2022
Publication Date: 4/5/2022
Citation: Hyun, S., Park, J., Kim, J., Fleisher, D.H., Kim, K. 2022. GLUEOS: A high performance computing system based on the orchestration of containers for the GLUE parameter calibration of a crop growth model. Computers and Electronics in Agriculture. https://doi.org/10.1016/j.compag.2022.106906.
DOI: https://doi.org/10.1016/j.compag.2022.106906

Interpretive Summary: Computer based crop models are useful tools to evaluate agricultural management options. However, a calibration process is required to ensure the models provide the most accurate predictions. This process is time-consuming and does not always produce calibration values that give the best model fit. To address this issue, a computer platform was developed that consisted of multiple processors and software. This parallel processing framework shortens the time required to obtain calibration values and generates a global set of results that are typically more accurate than existing approaches. This was demonstrated with calibration of rice cultivar in which the new results were shown to be more accurate. This research can be adapted at low cost for most existing crop models which can benefit nearly all crop modeling studies that require calibration, including those involved in resource management, crop sustainability, and other food security related issues. Crop modelers, policy planners, and growers can benefit directly from this resulting work.

Technical Abstract: Cultivar parameters for crop models are typically estimated through calibration procedures based on global optimization methods such as Generalized Likelihood Uncertainty Estimation (GLUE). These approaches require significant computational resources and would benefit from the use of parallel processing infrastructure. A framework for a distributed computing system, referred to as the Generalized Likelihood Uncertainty Estimation Orchestration System (GLUEOS), was developed to facilitate this calibration process. GLUEOS was designed to manage multiple virtual containers within a distributed computing system using a simple web-based interface. In a case study, cultivar parameters for the cultivar Shindongjin were obtained for the CERES-Rice model using GLUEOS. The procedure was repeated 100 times to take into account and quantify the uncertainty associated with parameter estimates. The elapsed wall time was compared between two computing systems including low end single board (LESB) and high end desktop (HEDT) computers with average difference of just 22 seconds when 32 containers were used for a single round of calibration. The computing efficiency of GLUEOS was affected by framework configuration, including the number of containers assigned per node. The mean value from the 100 calibrated parameter sets resulted in reliable estimate of rice heading dates for both calibration and validation datasets. These results suggested that GLUEOS would allow for robust and accurate calibration of cultivar parameters with minimum expertise required for operating a distributed computing system. Utilizing GLUEOS with other crop models would facilitate a wider range of studies for any research that requires parameter calibration.