A POD Mapping Approach to Emulate Land Surface Models

Abstract:

Existing land surface models (LSMs) describe physical and biological processes that occur over a wide range of spatial and temporal scales. Since simulating LSMs at a spatial scale to explicitly resolve the finest resolution processes is computationally expensive, upscaling techniques are used in LSMs to capture effect of subgrid heterogeneity. However, routinely employed linear upscaling techniques that allow LSMs to be simulated at coarse spatial resolution can result in large prediction error. To efficiently predict fine-resolution solutions to LSMs, we studied the application of a reduce order model (ROM) technique known as the “Proper Orthogonal Decomposition mapping method” that reconstructs temporally-resolved fine-resolution solutions based on coarse-resolution solutions for two case studies. In the first case study, we applied POD approach on surface-subsurface isothermal simulations for four study sites (10⁴ [m²]) in a polygonal tundra landscape near Barrow, Alaska. The results indicate that the ROM produced a significant computational speedup (>10³) with very small relative approximation error (<0.1%) for two validation years not used in training the ROM. In the second case study, we illustrate the applicability of our ROM approach at watershed scale (1837 km²) model that is substantially more heterogeneous and demonstrate a hierarchical approach to emulating models at spatial scales consistent with mechanistic physical process representation.