Skip to main content
ARS Home » Northeast Area » University Park, Pennsylvania » Pasture Systems & Watershed Management Research » Research » Publications at this Location » Publication #302141

Research Project: Multifunctional Farms and Landscapes to Enhance Ecosystem Services

Location: Pasture Systems & Watershed Management Research

Title: How well can we predict forage species occurrence and abundance?

item WANG, AUDREY - Pennsylvania State University
item Goslee, Sarah
item MILLER, DOUGLAS - Pennsylvania State University

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 1/26/2014
Publication Date: 2/4/2014
Citation: Wang, A., Goslee, S.C., Miller, D. 2014. How well can we predict forage species occurrence and abundance[Abstract]?. Northeast Pasture Consortium. p 3.

Interpretive Summary:

Technical Abstract: As part of a larger effort focused on forage species production and management, we have been developing a statistical modeling approach to predict the probability of species occurrence and the abundance for Orchard Grass over the Northeast region of the United States using two selected statistical methods: a Generalized Linear Model (GLM) and a Generalized Additive Model (GAM). The predicted maps describe the suitability of the species for the region and the role of environmental factors on species growth. Field observations and ecologically meaningful gridded environmental variables (climate, topography and soils) were used to create the models. The predictors involved in the models were statistically significant with the r-squared of the model accuracy being between 0.3 and 0.4. Our results indicate that the predicted spatial patterns of species distribution appear to be ecologically realistic for the most of areas and are consistent among the different statistical methods. From an ecological point of view, the distributions for species abundance were more accurately predicted than that of species occurrence. We estimated the predictive performance based on cross-validation using the quantitative criteria for comparing the predictive ability of the statistical models. Our dataset was divided into “training” for calibration with 70% of data and “testing” for validation with the remaining 30% of data, because we currently don’t have another independent dataset with suitable forage species available. These optimal models could be applied to explore scenarios based on historical, current and future land use change and climate change cross the landscape for the region, addressing potential species response to changing environment. The approach we have followed in this project could be used to establish species distribution models that may be useful for other herbaceous species prediction, given the existing data of about 100 forage species available in our database.