GISS Lunch Seminar
Speaker: Nate Chaney (Duke Univ.)

HydroBlocks: An efficient and effective modeling framework to represent field-scale heterogeneity in Earth system models

The representation of multi-scale heterogeneity within the land components of Earth system models (i.e., land surface models) remains a persistent challenge. The emergence of “tiling” schemes over 30 years ago provided a path forward by parameterizing sub-grid heterogeneity via histograms of clusters (i.e., tiles). What was meant initially to divide a grid cell into different land cover types has now become a hierarchical approach to account for sub-grid heterogeneities in land use, lakes, glaciers, hillslopes, and soils, among others. As the land surface modeling community leverages big environmental datasets and moves towards increasing sub-grid spatial complexity in process-representation (e.g., lateral flow), the underlying requirements and assumptions of classical tiling schemes have become obsolete. In this presentation, I will introduce the HydroBlocks modeling framework that addresses this challenge by building on the tiling scheme to enable a sub-grid approach that effectively and efficiently represents land heterogeneity in Earth system models.

To assemble the sub-grid tiles or clusters, HydroBlocks leverages field-scale data from satellite remote sensing using hierarchical clustering methods. By directly learning the clusters from multi-dimensional covariate space, this approach avoids the pitfalls of traditional tiling approaches. The spatial location of these clusters has a significant value as they make it possible to determine how different clusters might interact. By identifying the clusters and their interactions, HydroBlocks can be used to model groundwater, surface flows, and potentially boundary layer interactions between clusters. To this end, the current HydroBlocks approach uses the Noah-MP 1-D vertical land surface model to resolve the core physics in each cluster/tile; the spatial interactions among the clusters are then resolved as divergence terms within the vertical land surface model. Other implementations of the HydroBlocks framework exist within the Geophysical Fluid Dynamics Laboratory (GFDL) LM4.2 model. Finally, the spatial mapping of the clusters enables direct one-to-many mapping of the modeling results to physical space making it possible to directly evaluate the modeling results at the field-scale. In summary, HydroBlocks provides a clear path to robustly represent heterogeneity in land surface models, allows for increasing complexity in sub-grid spatial interactions, and makes it possible to leverage multi-scale observations to improve Earth system models.