Using Tiling Schemes to Evaluate Earth System Models at the Field-Scale

The processes that govern the terrestrial water, energy, and biogeochemical cycles operate at spatial scales far below the spatial resolution of existing land surface models (LSMs). To handle this discrepancy, LSMs commonly partition each grid cell into representative tiles to parameterize field-scale sub-grid spatial heterogeneity. In principle, each grid cell’s tile output can be evaluated to assess model performance at the field scale. However, this approach is uncommon. Instead, usually the grid cell output resulting from the weighted average of the tiles is used. In this presentation, we will discuss how this approach is a missed opportunity. We will illustrate how existing tiling schemes can be exploited to more adequately use data from existing observation networks to evaluate model performance.

We will introduce this approach by evaluating an offline simulation of the GFDL land model over the contiguous United States (CONUS) using available in-situ observational networks. The model is run between 2004 and 2014 at a 10 km spatial resolution. This approach to model evaluation is possible due to the new method used in defining the tiles for each macroscale grid cell in the GFDL land model. The tiles are defined by clustering available 30-meter environmental data (USGS NED and POLARIS) according to their field-scale soil and topographic attributes—each 30-meter grid cell is assigned a tile. The land model then simulates these tiles and their subsurface spatial interactions via the exchange of water, energy, and nutrients.

Since each 30-meter grid cell is assigned a tile, available in-situ observations of states and fluxes of the water, energy, and biogeochemical cycles can be used to evaluate model performance at the tile level. We will discuss results from a preliminary evaluation of the model results of soil moisture, leaf area index, soil carbon, latent heat flux, and sensible heat flux. For this evaluation we will use a suite of in-situ observation networks including COSMOS, AmeriFlux, and the International Soil Carbon Network. We will show how the evaluation results vary depending on whether observations are compared to the co-located macroscale grid cell average or its corresponding tile.