Mediation of Fire-Climate Linkages by Vegetation Types in Alaskan Arctic Tundra Ecosystems: Impacts of Model Uncertainty on GCM-Based Forecasts of Future Fire Activity

Friday, 19 December 2014
Paul Duffy1, Philip E Higuera2, Adam M Young2, Fengsheng Hu3 and Michael Dietze4, (1)Neptune and Company, Los Alamos, NM, United States, (2)University of Idaho, Moscow, ID, United States, (3)University of Illinois at Urbana Champaign, Program in Ecology, Evolution and Conservation Biology, Urbana, IL, United States, (4)Boston University, Boston, MA, United States
Fire is a powerful landscape scale disturbance agent in tundra ecosystems. Impacts on biophysical properties (e.g. albedo) and biogeochemical function (e.g. carbon flux) underscore the need to better quantify fire-climate linkages in tundra ecosystems as climate change accelerates at northern high latitudes. In this context, a critical question is “How does the functional linkage between climate and fire vary across spatial domains dominated by different vegetation types?”

We address this question with BLM-Alaska Fire Service area burned data (http://fire.ak.blm.gov/predsvcs/maps.php) used in conjunction with downscaled historical climate data from the Scenarios Network for Alaska Planning (http://www.snap.uaf.edu/data.php) to develop gradient boosting models of annual area burned in Alaska tundra ecosystems. The sparse historical fire records in the Arctic necessitate explicit quantification of model uncertainty associated with the development of statistical analyses. In this work, model uncertainty is depicted through the construction of separate models depicting fire-climate relationships for regions defined by the graminoid, shrub, and wetland tundra vegetation classes (Circumpolar Arctic Vegetation Map: http://www.geobotany.uaf.edu/cavm/). Non-linear relationships between annual area burned and climate variables are depicted with partial dependence functions.

Our results show that vegetation-specific models result in different non-linear relationships between climate and fire. Precipitation variables generally had higher relative influence scores than temperature; however, differences between the magnitude of the scores were greater when models were built with monthly (versus seasonal) explanatory variables. Key threshold values for climate variables are identified. The impact of model uncertainty on forecasts of future fire activity was quantified using output from five different AR5/CMIP5 General Circulation Models. Model uncertainty corresponding to vegetation-specific functional linkages between climate and fire has a magnitude similar to that of GCM-based model uncertainty. Results of this work strongly suggest that future forecasts of fire activity must account for and quantify sources of uncertainty in order to provide useful information.