Representativeness assessment of the pan-Arctic eddy-covariance site network, and optimized future enhancements

Abstract

Abstract. Large changes in the Arctic carbon balance are expected as warming linked to climate change threatens to destabilize ancient permafrost carbon stocks. The eddy covariance (EC) method is an established technique to quantify net losses and gains of carbon between the biosphere and atmosphere at high spatio-temporal resolution. Over the past decades, a growing network of terrestrial EC tower sites has been established across the Arctic, but a comprehensive assessment of the network’s representativeness within the heterogeneous Arctic region is still lacking. This creates additional uncertainties when integrating flux data across sites, for example when upscaling fluxes to constrain pan-Arctic carbon budgets, and changes therein. This study provides an inventory of Arctic (here >= 60° N) EC sites, which has also been made available online (https://cosima.nceas.ucsb.edu/carbon-flux-sites/). Our database currently comprises 120 EC sites, but only 83 are listed as active, and just 25 of these active sites remain operational throughout the winter. To map the representativeness of this EC network, based on 18 bioclimatic and edaphic variables, we evaluated the similarity between environmental conditions observed at the tower locations and those within the larger Arctic study domain. With the majority of sites located in Fennoscandia and Alaska, these regions were assigned the highest level of network representativeness, while large parts of Siberia and patches of Canada were classified as under-represented. This division between regions is further emphasized for wintertime and methane flux data coverage. Across the Arctic, particularly mountainous regions were poorly represented by the current EC observation network. We tested three different strategies to identify new site locations, or upgrades of existing sites, that optimally enhance the representativeness of the current EC network. While 15 new sites can improve the representativeness of the pan-Arctic network by 20 percent, upgrading as few as 10 existing sites to capture methane fluxes, or remain active during wintertime, can improve their respective network coverage by 28 to 33 percent. This targeted network improvement could be shown to be clearly superior to an unguided selection of new sites, therefore leading to substantial improvements in network coverage based on relatively small investments.

Vegetation type is an important predictor of the arctic summer land surface energy budget

“…Furthermore, year-round turbulent flux measurements (i.e. sensible and latent heat flux) are especially scarce for all vegetation types 53 . Finally, SEB observations for barren tundra are largely missing, while this type shows largest differences in surface energy fluxes to other tundra vegetation types in the limited data available for this study.…”

Permafrost and Climate Change: Carbon Cycle Feedbacks From the Warming Arctic

“…Given the size of the Arctic region, the rugged environmental conditions, and limited accessibility, it has been a challenge to detect carbon cycle change over the region as a whole. There are notable well-studied sites with a history of scientific research that have provided a wealth of mechanistic insight, but the limited number has been a barrier toward understanding and quantifying the aggregate response of the entire region (60)(61)(62)(63)(64)(65)(66). At the same time, the history of the region provides inference into past interactions between terrestrial ecosystems and the atmosphere for CO 2 , which is the largest carbon flux.…”

Gaps in Network Infrastructure limit our understanding of biogenic methane emissions in the United States

“…This approach provides a unit-less relative measure of representativeness between a location defined as the medoid of each cluster and every other location within that cluster. Although studies of this nature often define representativeness relative to a research site (Pallandt et al, 2021), this study identified positions within the landscape that are more representative of the medoid condition of the cluster and measured the distance from this medoid to understand how representative any location within a cluster is to the medoid condition. To extrapolate the cluster and distance layers across the entire US beyond the 20,000-pixel subsample, we fit a Random Forest model with the package randomForest (Liaw and Wiener, 2002) to model the first and second MDS dimension using the ecotype and climate layers as predictors.…”