Considerations for selecting a machine learning technique for predicting deforestation. (September 2020)

• Record Type: Journal Article
• Authors: Mayfield, Helen J.; Smith, Carl; Gallagher, Marcus; Hockings, Marc
• Is Part Of: Environmental modelling & software. Volume 131(2020)
• Publication Date: 2020-09
• Subjects: Machine learning -- Deforestation -- Generalized linear mixed models -- Bayesian networks -- Artificial neural networks; 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.2020.104741
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: There are a many reasons for creating quantitative models of deforestation, supported by a variety of modelling techniques for doing so. We examine the suitability of four different modelling techniques for predicting deforestation; Bayesian networks (BNs), artificial neural networks (ANNs), Gaussian processes (GPs), and generalised linear mixed models (GLMMs). The analysis is provided in the context of the Verified Carbon Standard Approved Methodology for Avoided Unplanned Deforestation to illustrate specific examples of scenarios where each technique would be suitable. ANNs potentially improve predictions over GLMMs, while GPs were better able to generalise to other geographic areas. ANNs were sensitive to design decisions, which may limit their applications where results are required within short time frames. BNs easily facilitate scenario analysis, but are also sensitive to design decisions. With machine learning becoming easier each day, we encourage researchers to consider carefully which technique is suitable for their situation prior to beginning implementation. Highlights: Linear regression, Bayesian networks, neural networks and Gaussian processes are evaluated for predicting deforestation. ANNs and BNs potentially improve predictions over GLMMs but are more sensitive to design decisions. The Verified Carbon Standard methodology for avoided unplanned deforestation shows problem characteristics affect model choice. Researchers are urged to consider the constraints and objectives of their study before selecting a modelling technique.