Modeling complex habitat use for threatened species decision-making in a landscape with competing ecological and economic values

Species distribution models (SDMs) are widely used in conservation; however, their uptake in decision-making can be limited when they fail to reflect species' ecological complexity and/or end- user needs. We developed SDMs for two threatened cave-roosting bat species, the ghost bat Macroderma gigas and Pilbara leaf-nosed bat Rhinonicteris aurantia, in the Pilbara region of Western Australia, an area of globally significant biodiversity and intensive mining activity. Using a novel neighborhood approach, we linked expert-validated roost records to the surrounding environment within the putative foraging area, enabling discrete roosting and foraging requirements to be incorporated into models. Habitat suitability was predicted using an ensemble of tuned-MaxEnt, Boosted Regression Trees and Random Forest models with outputs refined into a spatial tool for conservation decision-making via iterative co-design with end users. Ensemble models performed strongly (AUC ROC ≥ 0.96) and revealed distinct predictors of potential roosting and foraging habitat. Roosting habitat was predicted by rugged terrain and iron-rich strata, while foraging habitat was best explained by high vegetation diversity and moderate ruggedness. Over 70% of predicted habitat overlapped areas under mining tenure, with many roosts located close to operational mines, and only 5–6% of predicted habitat occurring within conservation estate. This highlights the complex challenges of conserving biodiversity in extractive landscapes. The resulting spatial tool incorporated decision-relevant information, making distinct habitat types visible and actionable for planners and regulators. Our neighborhood modeling framework and co-design process provide a transferable approach for developing fit-for-purpose biodiversity planning tools in other contested, multiuse landscapes.