Improved dryland carbon flux predictions with explicit consideration of water-carbon coupling

Authors: BARNES, M.L. - University Of Indiana; FARELLA, M.M - University Of Arizona; Scott, Russell - Russ; MOORE, D.J.P. - University Of Arizona; PONCE-CAMPOS, G.E. - University Of Arizona; Biederman, Joel; MACBEAN, N. - University Of Indiana; LITVAK, M.E. - University Of New Mexico; BRESHEARS, D.D. - University Of Arizona

Submitted to: Communications Earth & Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/29/2021
Publication Date: 12/2/2021
Citation: Barnes, M., Farella, M., Scott, R.L., Moore, D., Ponce-Campos, G., Biederman, J.A., MacBean, N., Litvak, M., Breshears, D. 2021. Improved dryland carbon flux predictions with explicit consideration of water-carbon coupling. Communications Earth & Environment. 2. Article 248. https://doi.org/10.1038/s43247-021-00308-2.
DOI: https://doi.org/10.1038/s43247-021-00308-2

Interpretive Summary: There is a growing appreciation of the crucial role played by arid and semiarid ecosystems in the global carbon cycle, yet current process- and remote sensing-based models used to estimate current and future carbon uptake do not accurately represent dryland carbon dynamics. Here, we present a new remote sensing-based machine learning model to estimate dryland carbon uptake that is uniquely tuned to key features of dryland ecohydrology, including high hydroclimate variability and drought. Compared to existing models, our model more accurately represents carbon uptake and provides more realistic estimates of drought impacts on the carbon cycle in global drylands. Our results underscore the importance of including ecohydrological water-carbon coupling to avoid underestimation of total carbon uptake, interannual variability, and drought impacts on dryland carbon fluxes. We suggest models need to account for the well-established link between hydrologic and carbon cycles for accurate regional and global carbon modeling in dryland systems.

Technical Abstract: Recent studies have highlighted the dominant role of dryland ecosystems in both the trend and the interannual variability of the terrestrial carbon sink. Despite the importance of these water-limited systems, current process-based earth system models and remote-sensing-driven estimates of vegetation production do not adequately capture dryland carbon dynamics. Here, we present DryFlux, a new product that fuses in situ flux measurements with remotely sensed observations in a machine learning algorithm to estimate dryland carbon uptake (gross primary productivity; GPP). DryFlux was driven by a dense network of eddy covariance sites spanning dryland ecosystem types and climate spaces. Improved characterization of intra- and inter-annual impacts of water availability, including drought, on carbon uptake in our model resulted in more accurate predictions of interannual and seasonal variability in dryland GPP. We found that our model trained in the Southwest region of North America provides realistic estimates of drought impacts on the carbon cycle in global drylands, underscoring the process-based underpinnings of the drought terms in our model. Applying our model to Australia, we found existing global GPP products consistently underestimate interannual variability in dryland carbon uptake in response to El Niño–Southern Oscillation phases. We anticipate DryFlux will be a starting point for a more sensitive accounting of drought impacts on the global carbon cycle and an improved benchmark for earth system models in drylands. More generally, our results highlight how ecohydrological accounting for the well-established link between hydrologic and carbon cycles is likely crucial for accurate regional and global carbon modeling.