Resolution and Spatial Compatibility issues in Land Surface Modeling

In offline simulations involving land surface models (LSMs), numerous questions have been raised related to scaling issues associated with grid resolution, as well as how to make comparisons between simulations or observations that produce a value for an area against those that are valid for a single point. Using the framework established by the Land Data Assimilation Systems (LDAS) project, we assess the loss of accuracy when grid resolution is degraded and how area soil moisture relates to single point values.

One method to offset a loss of accuracy, due to coarse grid resolution, is to introduce a tiling or mosaic approach. In this approach, the area of a grid box is broken down into subparts related to the dominant vegetation types in the grid box. The same meteorological forcing is use for each "tile", however, different model parameters, based on vegetation type, are used by the LSM. The results for the individual tiles are then averaged, using area weighting, into a single value for the grid box. We look at the changes in the surface fluxes when altering the maximum number of tiles allowed per grid. For LDAS's 1/8° resolution over the USA, no single grid box has more than ten different types of vegetation, so ten is the maximum number of tiles. The average number of different vegetation types, and thus number of tiles per grid, for the USA, is roughly five. At the 1/8° resolution, using more than three tiles per grid has proven to be excessive. However, as the grid resolution increases, we see larger differences and find that at higher resolution it is advisable to consider using five tiles per grid.

In order to aid in the comparison of model areas to point measurements we have performed another set of experiments. These experiments use forcing data, observed by the individual stations, where the soil moisture is observed, to force the LSM. Comparing the model results valid at the point to those valid for an area helps us to better understand how to compare both sets of model results with the point soil moisture observations.

When changing grid resolution and when comparing model results valid for an area with those valid for just the singular point, we have found that precipitation differences create the largest influence upon the results, for both soil moisture and the surface fluxes. Our results imply that for offline simulations it may only be necessary to have higher resolution precipitation estimates, where as the rest of the forcing data can be at a coarser resolution. This would also imply that when comparing area estimates of soil moisture, either from model results or satellite observations, to point estimates, it is imperative to have some estimate of the local spatial scale of precipitation.