Quantifying Carbon Fluxes over the Alaska North Slope Using Eddy Covariance Flux Tower Observations and Machine Learning

Stephen Shirley, University of Montana, stephen.shirley@ntsg.umt.edu (Presenter)
Jennifer Dawn Watts, Woods Hole Research Center, jwatts@whrc.org
John S Kimball, University of Montana, john.kimball@mso.umt.edu
Donatella Zona, San Diego State University (USA), dzona@mail.sdsu.edu
Eugenie Euskirchen, University of Alaska, Fairbanks, seeuskirchen@alaska.edu
Walter Oechel, San Diego State University, woechel@mail.sdsu.edu
Susan M. Natali, Woods Hole Research Center, snatali@whrc.org

Arctic permafrost regions contain over half of the global soil organic carbon pool. Due to climate warming in Arctic-boreal regions, this large store of carbon is vulnerable to microbial decomposition and loss to the atmosphere as greenhouse gases. While a warmer and wetter Arctic may increase vegetation growth and productivity, a drier climate could contribute to greater magnitudes of ecosystem respiration and methane oxidation, heightening the release of stored soil carbon to the atmosphere as carbon dioxide (CO2) and methane (CH4), and shifting the Arctic from a carbon sink to carbon source. Empirical and satellite data-driven models provide a useful method to quantify and monitor Arctic-boreal net ecosystem carbon budgets at coarse spatial scales, and are complimented by in situ eddy covariance flux tower measurements of ecosystem CO2 and CH4 flux at a less than 1 km scale. This presentation provides an update of ongoing efforts to model carbon dioxide and methane fluxes over the Alaska North Slope using machine learning random forest models trained with observations from eight eddy covariance flux towers located across northern Alaskan tundra. Random forest model inputs were determined based on their importance factor and include MERRA2 reanalysis soil moisture and temperature, snow mass, photosynthetic active radiation, and wind speed, as well as MODIS leaf area index (MCD15A2H), normalized difference vegetation index (NDVI; MOD13Q1/MYD13Q1), and daytime land surface temperature (LST; MOD11A2/MYD11A2). Eddy covariance tower sites have been categorized by wetland type using a fine resolution synthetic-aperture radar derived classification of Alaska wetlands. Here we present a site comparison of estimated daily CO2 and CH4 fluxes from the empirical random forest carbon flux model to eight in situ eddy covariance flux tower observations, along with Soil Moisture Active Passive Level 4 Carbon (SMAP L4C) and Terrestrial Carbon Flux (TCF) satellite data driven model outputs. Our results reveal shifts in the importance of different environmental controls influencing CO2 and CH4 fluxes both seasonally and between lowland and upland tundra sites. These results are being used to derive local scale (100-m resolution) carbon flux maps over northern Alaska for linking field, airborne and satellite based assessments and understanding of the regional carbon budget.

Associated Project(s): 

• Kimball-04
• Oechel-01

Poster Location ID: 90

Session Assigned: Carbon Dynamics