Characterizing subsurface controls on the Arctic ecosystem carbon cycling across scales using geophysical, in-situ and remote sensing datasets

Recent studies suggest that warming climate has significant impacts on the Arctic ecosystems, which could in turn create feedbacks to the climate system. Predicting Arctic ecosystem evolution and feedbacks requires a mechanistic understanding of complex subsurface processes (e.g., permafrost degradation and carbon decomposition) and characterization of key controls on those processes - such as soil moisture, active layer thickness and aqueous geochemistry - from native scales (where the processes occur) to landscape scales considered in climate models. Among others, developing such a predictive understanding requires two advances that we consider in this study: (1) approaches to characterize heterogeneous geocryological and hydrogeochemical properties in high resolution, and their uncertainties, inter-correlations, and correlations to land surface variables; (2) identification of possible zonation that can distinguish between suites of subsurface properties. Our study is carried out at the Department of Energy, Office of Science, Next Generation Ecosystem Experiment (NGEE-Arctic) Site in Barrow, Alaska. Given the importance of quantifying heterogeneous subsurface properties in both high resolution and over large spatial scales, we take advantage of geophysical datasets. We first develop and implement a multiscale Bayesian method to estimate subsurface properties using disparate datasets. We consider in-situ, surface geophysical and remote-sensing datasets along two 500m transects that traverse various ice-wedge polygon features. The method is based on site-specific correlations between geophysical datasets and in-situ measured subsurface properties. The Markov-chain Monte-Carlo sampling is used to compute the posterior distributions. For example, thaw depth and soil moisture are estimated by integrating ground-penetrating radar (GPR), electrical resistivity tomography and in-situ point data. Snow depth coverage is estimated by integrating GPR and point data. The validation results show that the method successfully captures the spatial distribution of those properties within the confidence intervals. For the second objective, statistical clustering/classification methods are used to identify zonation that distinguishes between suites of subsurface properties potentially important for predicting carbon cycling. We consider different types of zonation, including polygon feature (i.e., trough, rim or center), polygon-type classification and soil texture boundaries. To assess the relative importance, statistical tests are performed. Results are compared to the flux measurements at different locations, suggesting a potential link between the identified zonation and carbon flux. Results also show that the zonation approach is useful for guiding subsequent data acquisition and model parameterization strategies at Arctic lowland sites such as the Arctic Coastal Plain.