How realistic are the spatial patterns simulated from field-scale resolving land surface models?

Emerging field-scale resolving land surface models (LSMs), such as HydroBlocks, aim to model the water, energy, and biogeochemical cycles (e.g., surface energy partitioning) at 10-100 meter spatial scales over continental extents. However, there has yet to be a concerted effort to evaluate the realism of the simulated field-scale spatial patterns. This presentation challenges the scientific community to evaluate the modeled multi-scale spatial patterns of contemporary land surface models more critically. Here, we present an approach to evaluate the modeled multi-scale spatial patterns of land surface temperature (a linchpin state variable in the land surface energy and water cycles) of the HydroBlocks LSM using GOES-16 land surface temperature over the contiguous United States (CONUS).

To perform this evaluation, HydroBlocks is run at an effective 30-meter spatial resolution over CONUS at an hourly time step between 2015 and 2020. The domain is then split into 0.5 arcdegree grid cells (~50 km) and a series of spatial statistics are computed (e.g., spatial variance and correlation length) at hourly, daily, monthly, and annual time steps. These spatial statistics are also calculated using the GOES-16 land surface temperature product at the available time steps (with 80%+ spatial coverage per 0.5 degree grid cell). GOES-16 provides hourly observations of land surface temperature over CONUS at a 2 km spatial resolution. The simulated and observed spatial statistics are then compared between 2017 and 2020 for each macroscale grid cell over CONUS. The results show a poor correlation between the two at hourly time scales but show marked improvement over larger time scales. In any case, the surprisingly weak correlation between the observed and simulated spatial statistics reinforce the need to think more critically about the spatial uncertainty chain in land surface models. More importantly, this work reemphasizes the need to make simulated spatial patterns an integral part of the evaluation and calibration of macroscale land surface and hydrologic models moving forward.