C013-0014
Identification of Regions Susceptible to Thermokarst Initiation on the Alaska Arctic Coastal Plain Using Random Forest Models
Tuesday, 8 December 2020
Poster
Bob Bolton, University of Alaska Fairbanks, International Arctic Research Center, Fairbanks, AK, United States, Rawser Spicer, International Arctic Research Center, Fairbanks, United States, Helene Genet, University of Alaska Fairbanks, Institute of Arctic Biology, Fairbanks, AK, United States and Amy Lynn Breen, University of Alaska, Fairbanks, International Arctic Research Center, Fairbanks, AK, United States
Abstract:
Landscape change in permafrost regions, caused by thermokarst, can result in profound impacts on the energy and water balance; carbon fluxes; wildlife habitat; and infrastructure. The Alaska Thermokarst Model (ATM) is an intermediate-scale, state-and-transition model designed to simulate landscape evolution in polygonal tundra due to thermokarst. A number of studies have shown the following sequence of events regarding polygonal tundra evolution: 1) a relatively stable landscape; 2) a "pulse" or extreme climate event that initates the thermokarst process; 3) rapid landscape evolution; and 4) stabilization of the landscape. Jorgenson (2006) reported that initial and advanced degradation of these ice-wedge dominated landscapes can occur a 20-year period (and quickly as 10-years) and advanced stabilization within a 20-30 year period. Stabilization of the landscape occurs when the seasonal thaw layer depth is unable to penetrate the protective layer (surface soil layer that buffers surface processes from underlying ice-rich permafrost).
This study focuses on the second step of the landscape evolution process - initiation of the thermokarst process using random forest models. Random forests are an ensemble learning technique that combines the results of many independent decision trees to create results that avoid the overfitting in regular decision trees. The random forests we present were trained against our “climate priming” thermokarst initiation model on the Alaska Arctic Coastal Plain. We examined 13 features to train the random forest model. Of those 13 features, the 5 most important features in predicting thermokarst initiation are: late summer precipitation, next summer precipitation, winter temperature, and early winter precipitation. The goal of this work is develop an computationally efficient module for ESMs to identify the timing and locations of thermokarst initiation.
