|Title||Using machine learning to model complex landscapes: predicting the geographic range of Vesicular Stomatitis across the western United States|
|Publication Type||Conference Paper|
|Year of Publication||2019|
|Authors||N. Burruss D, Peters DC, Rodriguez LL, MCVEY DAVIDS, Elias E, PELZEL-MCCLUSKEY ANGELAM, Savoy H, Derner JD, PAUSZEK STEVENH|
|Conference Name||Ecological Society of America Abstracts|
|ARIS Log Number||361480|
Complex relationships among a virus, host, vector, and environment challenge prediction of disease spread through time and space. In addition, changing climate and land use regimes result in novel risk to biota and emphasize the need to improve understanding of mechanisms governing disease. Vesicular Stomatitis (VS) is the most common vesicular livestock disease in North America. Recent investigations have characterized how VS generally spreads from its endemic range in Central America northward through the western United States (US). Despite regional insights into VS, there is still a need to explore the broader environmental relationships limiting the spatial extent of VS to improve our understanding of the potential geographic range. The objective of this study was to model the potential range of VS in the US under current and alternative climate scenarios. The current model of VS occurrence was constructed using a transdisciplinary big data-model integration approach coupled with machine learning to predict the potential maxima of VS occurrence using disparate data sources representing soil, climate, land use, host, and vector properties. An information-theoretic approach was used to construct and select a model which was projected into potential future climate scenarios. The current extent of VS is largely confined to the western portion of the United States. The probability of occurrence was most associated with low summer surface runoff, low winter and summer precipitation, moderate summer and fall vegetation greenness, and increased maximum winter temperatures. Predictions using alternative climate data suggest that reductions in NDVI and increases in temperature can result in predicted occurrences that are spatially explicit and non-uniform in their direction of change. Our results suggest that the result of modifications to the environment can be complex and may either mitigate or increase occurrence risks. We suggest that these results can help define priority areas for research, monitoring, and mitigation efforts. Moreover, this analysis provides an illustration of the robustness of the big data-model integration approach in exploring complex problems at multiple scales.