Comparison of machine learning methods emulating process driven crop models. (April 2023)

• Record Type: Journal Article
• Authors: Johnston, David B.; Pembleton, Keith G.; Huth, Neil I.; Deo, Ravinesh C.
• Is Part Of: Environmental modelling & software. Volume 162(2023)
• Publication Date: 2023-04
• Subjects: Metamodels -- Surrogates; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2023.105634
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Performing large scale simulation analyses using complex process-driven models can be very time consuming and incur significant computational expense. These analyses involve generating synthetic datasets and include processes such as impacts analysis (IA) and variance-based sensitivity analysis (SA). Machine learning (ML) provides a potential alternative path to reduce computational costs incurred when generating output from large simulation experiments. We assessed the accuracy and computational efficiency of three ML-based emulators (MLEs): artificial neural networks, multivariate adaptive regression splines, and random forest algorithms, to replicate the outputs of the APSIM-NextGen chickpea crop model. The MLEs were trained to predict seven outputs of the process-driven model. All the MLEs performed well (R 2 > 0.95) for predicting outputs for the training data set locations but did not perform well for previously unseen test locations. These findings indicate that modellers using process-driven models can benefit from using MLEs for efficient data generation, provided suitable training data is provided. Highlights: Machine learning emulators of process-driven models produce accurate predictions. Some emulator types produce more consistently accurate predictions than others. Comprehensive training data is essential to ensure emulators can predict accurately.