Location: Northwest Watershed Management Research
Title: The use of similarity concepts to represent sub-grid variability in hydrologic and land-surface models: case study in a snowmelt dominated watershed Authors
Submitted to: Journal of Hydrometeorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: May 12, 2014
Publication Date: June 1, 2014
Citation: Newman, A., Clark, M., Winstral, A.H., Marks, D.G., Seyfried, M.S. 2014. The use of similarity concepts to represent sub-grid variability in hydrologic and land-surface models: case study in a snowmelt dominated watershed. Journal of Hydrometeorology. DOI: 10.1175/JHM-D-13-038.1. Interpretive Summary: This paper compares different methods of spatial aggregation to evaluate the impact of spatial simplification schemes typically used in large-scale hydrologic modeling in land-surface models. The best result was based on physiographic information such as elevation and land-cover. All aggregation methods tend to over-estimate streamflow, but are closer to actual streamflow in dry years. If we consider only daily streamflow, aggregated methods are not statistically different from high-resolution (10m) grid-based simulations.
Technical Abstract: This paper first compares two fully distributed hyper-resolution (10m) simulations with 48 three common approaches used in hydrologic and land-surface models (LSMs): the “lumped” approach (no sub-grid variability), disaggregation by elevation bands, and disaggregation by 50 vegetation types. These two sets of end members are then compared against two variants of the K-Means approach: an a posteriori case where clusters are identified based on output from 52 hyper-resolution simulations and an a priori case where clusters are based solely on physiographic information. The a priori case is the candidate method for implementation into 54 LSMs implemented over larger spatial scales. Evaluation of these methods in the Reynolds Mountain East (RME) catchment (Idaho, 56 USA) illustrates that the a priori clustering method is able to capture the aggregate impact of fine-scale spatial variability with O(10) instead of O(1000) simulation points. Catchment-58 average simulations of sensible heat flux, latent heat flux, and net radiation are generally within 10-15% for the two hyper-resolution and two K-means approaches. Examination of observed and 60 simulated streamflow timeseries show that the a priori method tends to overestimate peak streamflow in wet years, while in dry years it is closer to observed peak flow. The a priori 62 method generally reproduces the observed peak flow better than all other simple disaggregation approaches. It also has more similar recession limbs to the hyper-resolution simulations and 64 observations. Finally, there is no statistically significant performance improvement from the a priori simulation to the hyper-resolution distributed simulations for daily streamflow statistics.