Page 45 - GIS for Science, Volume 3 Preview

MODELING HABITATS
With foundations in the mid-20th century, predictive approaches to developing habitat maps are not new. However, the science of habitat suitability modeling has matured, with advances in modeling algorithms and the increasing availability of fine-scale data characterizing the environmental factors to which species respond (Sofaer et al. 2019). The question today is not “Are good habitat models possible and useful?” but “How can habitat modeling efforts be brought to scale so that products are available and understandable to decision-makers?”
To fulfill its mission to provide biodiversity location information to decision- makers, NatureServe has brought habitat modeling to scale using its accumulated inventory data and the biological and modeling expertise of its scientists. This approach moves habitat modeling out of the academic sphere and into more practical applications. Initial predictions of species habitat can be readily generated through a dynamic process and then, through review and iteration, refined over time. Through this process, the information available to guide decision-making is constantly improving.
Habitat modeling process
2
1
3
64 5
1. Species occurrence data—Records from nearly 50 years of biological inventory con- ducted by state Natural Heritage Program field biologists provide the expert verified occurrence data needed to train models. These data are often supplemented with occurrence data from additional sources, including natural history museum and citizen science records.
2. Environmental predictors—NatureServe has compiled a national library with more than 200 terrestrial and freshwater spatial environmental predictors on soils, climate, landform, land cover, hydrology, and other environmental factors. These high-quality environmental data layers make high-resolution modeling possible.
3. Modeling engine—Using Microsoft’s cloud-based artificial intelligence (AI) modeling engine, researchers process terabytes of data to intersect species occurrence data with environmental predictors. The goal is to characterize the relationship between these variables, build predictive models, and generate maps of potential habitat.
4. Model products—Predictive modeling products include continuous habitat suitabil- ity predictions from 0 (low) to 1 (high) and classified maps distinguishing high, medium, and low probability habitat and areas unlikely to support the species. Metadata provide information on environmental factors influencing the prediction, model performance, and appropriate uses for end products.
5. Expert review—Each model is uploaded into a GIS-based web application that allows experts familiar with the species to view the predictions and assess how well the models performed. Reviewers’ comments help refine the models and communicate about confidence in the predictions.
6. Field validation—Model products are also used to guide sampling and inventory efforts, providing new data that can be fed back into an iterative modeling process and improve products.
Preventing Species Extinctions 33