Scaling approaches for improving estimates of biogeochemical cycling in spatially heterogeneous ecosystems

Background/Question/Methods

Upscaling ecological information to larger scales in space and downscaling remote sensing observations or global model simulations to finer scales remain grand challenges in ecological and Earth system science. Here, we compare and contrast multiple scaling methodologies to improve our understanding of biogeochemical cycling in tundra and boreal wetland ecosystems that are characterized by pronounced spatial variability. Downscaling involves inferring subgrid information from coarse-scale observations. This is an ill-posed problem, for which Regularization is a classic mathematical approach. We apply two-dimensional Tikhonov Regularization (TR) to estimate subgrid surface patterns of normalized difference vegetation index (NDVI) and leaf area index (LAI) in a tundra ecosystem near Abisko, Sweden, from coarse-scale remote sensing information and subgrid statistics. We also demonstrate applications of information theory-based scaling approaches and multiresolution analysis for improving estimates of ecosystem function across scales in space.

Results/Conclusions

TR-downscaled estimates of Gross Primary Productivity (GPP) differ from estimates from fine-scale observations by <10% if the impacts of vegetation edge on canopy carbon uptake are simulated. We describe how a derived probability distribution of fine-scale information can be used to upscale tundra carbon flux estimates with minimal (<5%) bias. We then apply a multiresolution analysis to fused Landsat and MODIS products to investigate the spatial scales of change in a thawing permafrost region, Innoko, AK, since the beginning of the satellite record. We find evidence of changing vegetation patters at two canonical scales: fine (meters) scales that correspond to changes in thaw feature edges, and larger (kilometers) scales that correspond to the characteristic dimension of hills in the area. Scaling approaches are the handshake between plot-level ecological inference and global understanding, and ongoing scaling efforts can only improve the representation of ecological dynamics in Earth system models.