The importance of land use for accurate species distribution models
Species distribution models are relied upon for wise land use planning in the urban-wildlife interface where conservation and development efforts can conflict. Although the built environment influences species’ distributions, these models rarely include land use. Further, existing species distribution models can over predict suitable habitat for individual species, communities, or potential refugia in the face of climate change.
Maxent is one model that predicts species distributions using animal occurrence and environmental datasets. To assess the importance of land use in a species distribution model, I constructed and compared models with and without land use variables for gray foxes in the San Francisco Bay Area, CA. I created two species distribution models using gray fox presence-only (n=210) and presence/absence (n=64) records, and 9 environmental and land use variables.
Including land use resulted in a model that was more sensitive and more specific. This analysis showed that including land use improved the accuracy of a species distribution model. Additionally, the land use variables explained more than half of the model’s importance.
This examination of the importance of land use in a species distribution model for gray foxes demonstrates that including land use improves the accuracy of a species distribution model. Further, land use variables can be more important than environmental variables in describing a species’ distribution. Omitting human land use variables predicts a more dispersed distribution of suitable habitat, which could lead to inaccurate prioritization of land for conservation.
Evaluating connectivity map predictions
Habitat fragmentation in human-dominated landscapes is seen as a major threat to biodiversity persistence. Nearly all corridor conservation plans designed to restore habitat connectivity are based on modeled data, and are rarely tested with empirical field data. For this project, I compared a simple, biologically informed model that predicts landscape permeability for wildlife movement using occurrence data from a large-bodied predator across a gradient of land use in the Santa Cruz Mountains, CA.
In collaboration with Dr. Adina Merenlender and Dr. Chris Wilmers, I compared three general landscape feature-based permeability models with species occurrence data for pumas (Puma concolor) in California’s Santa Cruz Mountains. Each model was derived from a estimated linear relationship between a specific landscape feature related to the built environment (distance to roads, mean parcel size, and median patch size) and observed responses by bird and meso-carnivore assemblages from the literature. Our results showed that observed puma movement in the Santa Cruz Mountains corresponded with the modeled permeability gradients. Specifically, pumas were observed to readily use moderately disturbed habitats, and rarely were detected in the most heavily disturbed areas.
This comparison of a more generic connectivity model estimate with animal field observations shows that while generic models can be useful for corridor designs in highly disturbed environments they may be less useful in moderately impacted rural to semi-natural landscapes, where more detailed studies of species behavior may be required to delineate functional corridors. Mapping the level of landscape permeability that surrounds the built environment, as measured by distance to roads and housing density, offers a spatially explicit way to identify areas important wildlife movement. This approach provides a tool to help managers and land-use planners quickly and affordably prioritize habitat corridors for biodiversity conservation across fragmented landscapes, even when species data are unavailable.
A PDF of the full article in Landscape and Urban Planning is available here.