|WANG, AUDREY - Pennsylvania State University
|MILLER, DOUG - Pennsylvania State University
Submitted to: Applied Vegetation Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/19/2016
Publication Date: 1/1/2017
Citation: Wang, A., Goslee, S.C., Miller, D., Sanderson, M.A., Gonet, J.M. 2017. Topographic variables improve climate models of forage species abundance in the northeastern United States. Applied Vegetation Science. 20:84-93. https://doi.org/10.1111/avsc.12284.
Interpretive Summary: Mapping potential abundance of major forage species based on climate, topography, and soils would enhance management capabilities and aid in understanding the effects of climate change on grazing agriculture. These models have mostly been developed for woody species and presence data, but to be useful for agriculture they must predict abundance of herbaceous species. We used field data on six common pasture species collected across the northeastern United States to model potential abundance. Random Forest models were the best predictors, and including topography as well as climate greatly improved the model. Adding soils had little additional effect. This model will be used to map forage species abundance under current and potential future climates.
Technical Abstract: Species distribution modeling has most commonly been applied to presence-only data and to woody species, but detailed predicted abundance maps for forage species would be of great value for agricultural management and land use planning. We used field data from 107 farms across the northeastern United States to model abundances for six forage species using Generalized Linear Models, Generalized Additive Models, and Random Forests. A hierarchical ecological framework encompassing climatic, edaphic, and topographic variables related to the plant species requirements of water, light and temperature was developed. Although many species distribution modeling studies have used only climatic variables, the inclusion of topography greatly improved explanatory power. Soils variables contributed little more beyond the information contained in climate and topography. Because GLM and GAM models are prone to overfitting, cross-validation was used to select the statistically-appropriate number of variables for each model. Random Forest models had greater overall predictive capability, and were used to produce the final potential abundance maps for the six forage species. This approach shows great promise for agricultural management in the northeastern United States.