An Efficient Scaling Technique For Predicting Fine Resolution Terrestrial Hydrologic And Carbon Dynamics

Abstract:

Limited computational resources typically limit the spatial resolution of watershed-scale hydrologic-biogeochemical models, thus necessitating the upscaling of topographic, biotic, and abiotic parameters. Upscaled models, however, may not capture nonlinear interactions between processes accurately, leading to significant biases in the solutions. In this presentation, we consider a sampling-based scaling approach that maps coarse-resolution solutions to the high-resolution solutions. The approach, called Proper Orthogonal Decomposition Mapping Method (PODMM), trains a reduced order model with coarse- and fine- solutions, here obtained using a high-resolution watershed-scale model. The model, PAWS+CLM, is a quasi-3D watershed model that has been validated against observations in many temperate watersheds; here we applied PAWS+CLM to construct coarse- (7 km) and fine- (200m) resolution simulations for the Clinton River basin in Michigan, USA. After the training stage, subsequent fine-resolution solutions were approximated based only on coarse-resolution solutions and the ROM, leading to significant computational speedup. We demonstrate that fine-resolution soil moisture, latent heat flux, and net primary production are accurately reproduced, with up to 80% bias reduction compared to solutions obtained using coarse-resolution model. In addition, the subgrid distributions of the ROM solutions closely resemble those obtained using the fine-resolution model. This method can potentially bridge the intrinsic difference in scales between different processes by allowing efficient upscaling and downscaling of spatial solutions.