Pan-Arctic Representativeness for Site Selection and Model Evaluation
Authors
Jitendra Kumar1, Forrest M. Hoffman1* (hoffmanfm@ornl.gov), W. Robert Bolton2, Stan D. Wullschleger1, Colleen Iversen1
Institutions
1Oak Ridge National Laboratory, Oak Ridge, TN; 2University of Alaska–Fairbanks, AK
URLs
Abstract
Characterizing the complex interactions among climate, landforms, permafrost, hydrology, biogeochemistry, vegetation, and snow is important to understanding how the Arctic will evolve under a rapidly changing climate. The NGEE-Arctic project is designed to advance such a predictive understanding and to deliver a process-rich ecosystem model suitable for modeling the evolution of Arctic ecosystems in a high-resolution Earth system model. In situ measurements and observational data are required to inform model development, and an optimal sampling strategy is needed to characterize vegetation and soil states and their responses to climate. Using a collection of pan-Arctic climatic, topographic, edaphic, and vegetation data, researchers applied a quantitative multivariate machine learning methodology to stratify environmental gradients into ecoregions, which serve as potential sampling domains, and to determine the representativeness of eddy covariance measurement and soil and vegetation sampling sites. Data from tundra and boreal regions were included in the analyses, and about a dozen unique domains were identified across the eight countries encompassing Arctic tundra. Maps of ecoregions were produced, and the relative representativeness of available sampling sites in providing pan-Arctic coverage of carbon, water, and energy data was estimated. These analyses provide quantitative information about the spatial and temporal coverage provided by each sampling location and will be used to inform site selection for phase four of NGEE-Arctic. Moreover, the multivariate approach offers a quantitative method for up-scaling measurements, downscaling models, and identifying locations where additional measurements might greatly enhance model process representations.