Evaluating modelling decisions and spatial predictions in ecosystem mapping

This is a Preprint and has not been peer reviewed. This is version 3 of this Preprint. You must log in to post a comment. There are no comments or no comments have been made public for this article. This is a Preprint and has not been peer reviewed. This is version 3 of this Preprint. Add a Comment You must log in to post a comment. Comments There are no comments or no comments have been made public for this article. Ecosystem maps support a vast array of applications in conservation, land management, and policy. The capacity of such an ecosystem map to support these applications is determined by its accuracy and in turn by the decisions made the modelling procedure. We evaluated the influence of select modelling decisions for a pixel-based, random forest classification model used to map ecosystems for the remote Tiwi Islands, Australia. Across the three modelling decisions of classification scheme using the Global Ecosystem Typology, satellite sensor selection between Landsat-9 and Sentinel-2, and covariate set composition by including additional data sources, we evaluated model performance using multiple metrics and produced spatially explicit uncertainty maps to communicate map limitations. Covariates additional to those from satellite images consistently improved model performance, representing the most impactful pathway to accuracy gains without sacrificing ecological detail unlike class aggregation. The choice of satellite and sensor provided smaller accuracy gains, with Landsat-9 acquisitions generally outperforming Sentinel-2, potentially because the spatial heterogeneity of ecosystems is modulated by the coarser resolution. Uncertainty maps are practical and accessible tools for producers to communicate limitations which is critical information for decision-making and achieving conservation goals. https://doi.org/10.32942/X2VK7W Ecology and Evolutionary Biology, Other Ecology and Evolutionary Biology remote sensing, Earth Observation, Vegetation mapping, land cover, island ecology, tropical savanna, machine learning, biogeography Published: 2025-02-05 10:29 Last Updated: 2026-05-15 10:01 CC BY Attribution 4.0 International Conflict of interest statement: None Data and Code Availability Statement: Data and code will be made publicly available upon acceptance of the manuscript. Language: English