Authors: Dongmei Chen*, University of Oregon, Patrick Bartlein, University of Oregon
Topics: Biogeography, Natural Resources, Human-Environment Geography
Keywords: Species distribution modeling, neural network, climate change, mountain pine beetle, North America
Session Type: Paper
Start / End Time: 8:00 AM / 9:40 AM
Room: Napoleon C2, Sheraton 3rd Floor
Presentation File: No File Uploaded
The recent range expansion of the native mountain pine beetle (MPB) in western North America is widely attributed to climate change in the literature. Climate directly affects beetle behavior and indirectly impacts beetles by affecting their interactions with other biotic factors. However, it remains unclear how changing climate has impacted the MPB outbreaks over time. In this study, we apply species distribution modeling to reconstruct the spatial pattern of MPB in the western US in the past century based on its current range and investigate the relative importance of various bioclimatic variables that are crucial for beetle development identified in the literature. To infer historical data, we examined a variety of models and the data were split into training, validation and test sets for all models. The test error represents predicting earlier beetle ranges using later data. The selection of models is based on using cross-validation accuracy on the test set as the criterion. The preliminary “best” model is a feed-forward neural network (FFNN) (accuracy: 94% training; 88% validation; 90% test). Although the FFNN had the highest prediction accuracy, interpretation of NN models is notoriously difficult. To better understand the significance of bioclimatic variables on large-scale beetle outbreaks, we examined the coefficients of the best logistic regression model.