Propagating subsurface uncertainty to the atmosphere using fully-coupled, stochastic simulations

Feedbacks between the land surface and the atmosphere, manifested as mass and energy fluxes, are strongly correlated with soil moisture, making soil moisture an important factor in land-atmosphere interactions. We show that uncertainty in subsurface properties propagates into atmospheric variables, and therefore reduction of uncertainty in hydraulic conductivity (K) propagates through land-atmosphere feedbacks to yield more accurate weather forecasts. Using ParFlow-WRF, a fully-coupled groundwater-to-atmosphere model, we simulate responses in land-atmosphere feedbacks and wind patterns due to subsurface heterogeneity with ensembles generated by varying the spatial location of subsurface properties, while honoring the global statistics and correlation structure. This approach is common to the hydrologic sciences but never-before used in atmospheric simulations. We clearly show that different realizations of K produce variation in soil moisture, latent heat flux and wind for both point and domain-averaged quantities. Using a single random field to represent the “actual” case, we sample varying amounts of K data and incorporate subsurface data into conditional Monte Carlo ensembles to show that the difference between the ensemble mean prediction and the actual saturation, latent heat flux and wind speed are reduced significantly via conditioning of K. By reducing uncertainty associated with land-atmosphere feedback mechanisms, we also reduce uncertainty in both spatially distributed and synoptic wind speed magnitudes, thus improving our ability to make more accurate forecasts important for many applications such as wind energy.