Reed, D. E., J. M. Poe, M. Abraha, K. M. Dahlin, and J. Chen. 2021. Modeled surface-atmosphere fluxes from paired sites in the Upper Great Lakes Region using neural networks. Journal of Geophysical Research Biogeosciences 126:e2021JG006363.
The eddy covariance (EC) method is one of the most widely used approaches to quantify surface-atmosphere fluxes. However, scaling up from a single EC tower to the landscape remains an open challenge. To address this, we used 63 site years of data to examine simulated annual and growing season sums of carbon fluxes from three paired land-cover type sites of corn, restored-prairie, and switchgrass ecosystems. This was also done across the landscape by modeling fluxes using different land-cover type input data. An artificial neural network (ANN) approach was used to model net ecosystem exchange (NEE), ecosystem respiration (Reco), and gross primary production (GPP) at one paired site using environmental observations from the second site only. With a mean spatial separation of 11 km between paired sites, we were able to model annual sums of NEE, Reco, and GPP with uncertainties of 20%, 22%, and 8%, respectively, relative to observation sums. When considering the growing season only, model uncertainties were 17%, 22%, and 9%, respectively for the three flux terms. We also show that ANN models can estimate sums of Reco and GPP fluxes without needing the constraint of similar land-cover-type, with annual uncertainties of 12% and 10%. These results provide new insights to scaling up observations from one EC site beyond the footprint of the EC tower to multiple land-cover types across the landscape.
Associated Treatment Areas:
GLBRC Scale-up FieldsSign in to download PDF back to index