Modeling individual tree mortality in the Sierra Nevada in response to the 2012-2016 California drought
Nicole M. Hemming-Schroeder, University of California, Irvine, hemmingn@uci.edu (Presenter)
Carl A. Norlen, University of California, Irvine, cnorlen@uci.edu
Steven D. Allison, University of California, allisons@uci.edu
James T. Randerson, University Of California, Irvine, jranders@uci.edu
The severe 2012-2016 California drought resulted in 50% tree mortality in some regions of the Sierra Nevada. However, tree mortality is challenging to predict at high spatial resolution. We used a new dataset of 1 million trees derived from lidar and multispectral data from the National Ecological Observatory Network in the Sierra Nevada to model tree mortality from biophysical feature variables using three machine learning methods: (1) random forests, (2) gradient-boosted decision trees, and (3) neural networks. We found that gradient-boosted decision trees performed the best of the three methods we explored and achieved 70% accuracy. The most important feature variables for predicting tree mortality were initial tree height, mean precipitation during the drought, and initial tree cover. These factors may account for the combination of beetle selection preference and water stress which contribute to tree mortality during severe drought. Predicting individual tree mortality during severe droughts may improve projections of forest carbon sequestration and support conservation efforts.
Poster: Poster_HemmingSchroeder_2-35_139_35.pdf
Associated Project(s):
Poster Location ID: 2-35
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: TE