B102-12
The predictability of near-term changes in forest biomass: a case study in boreal North America
Tuesday, 15 December 2020: 12:03
Virtual
Arden Llewellyn Burrell1, Scott J Goetz2, Sol Cooperdock3, Richard Massey4, Xanthe J Walker5, Michelle C Mack6,7, Robbie A Hember8, Logan T. Berner2, Adrianna Foster9, Stefano Potter10 and Brendan M Rogers10, (1)Woods Hole Research Center, Woods Hole, MA, United States, (2)Northern Arizona University, SICCS, Flagstaff, AZ, United States, (3)University of California Los Angeles, Department of Atmospheric and Oceanic Sciences, Los Angeles, CA, United States, (4)Northern Arizona University, School of Informatics, Computing, and Cybersystems, Flagstaff, AZ, United States, (5)Northern Arizona University, Flagstaff, United States, (6)Northern Arizona University, Flagstaff, AZ, United States, (7)Natural Resources Canada, Victoria, BC, Canada, (8)University of Virginia Main Campus, Charlottesville, VA, United States, (9)Woods Hole Research Center, Falmouth, MA, United States
Abstract:
Climate change is impacting forest ecosystems around the globe. Over the last 70 years, anthropogenic climate change caused the boreal forest to warm at a much faster rate than most other terrestrial biomes. Continued warming will likely drive
significant changes in forest productivity, mortality and recruitment. Some of these changes will occur rapidly while others are expected to gradually unfold. For example, tree-level productivity can decline decades before mortality and associated occurrence of landscape-level changes. The ability to predict these changes would be invaluable to both researchers and land managers, as it would allow time to address or mitigate impending changes to forests and their ecosystem services. There is evidence that specific satellite-based vegetation indices can be used to predict both changes in site-level productivity and to provide early warning of tree mortality. These remotely sensed data can potentially be used to develop tools to predict future changes in biomass when used in combination with site and climate characteristics, as well as an understanding of how productivity dynamics relate to tree mortality.
In this study, we present a modeling framework to predict 10-year changes in forest biomass at forest inventory plots across Canada and Alaska. We integrate several long-term satellite vegetation indices along with soil characteristics, long-term climate variables and species composition. A range of machine learning methods including Random forest and extreme gradient boosted models were trained on over 10,000 forest inventory plots with repeat measurements covering a wide range of forest types across boreal North America. We present results of our predictive models, cross-validated using a withheld subset of the forest inventory plots. Using these models, we investigate the potential to future biomass change. These models have the potential to improve our understanding of long-term and large-scale drivers and changes in boreal forest biomass as well as to provide scientific and management communities with a novel monitoring tool.
