The Use of Hydrological Similarity Concepts to Represent Sub-grid Variability in a Snowmelt Dominated Watershed

A significant modeling challenge is to adequately represent the impacts of small-scale variability in cryospheric and hydrological processes on surface fluxes and streamflow at larger spatial scales. Sub-grid variability in large-scale land surface modeling is traditionally represented by disaggregating a grid-box or watershed into elevation bands and/or vegetation types, with uniform forcing data used for all sub-grid tiles. This approach clearly neglects the spatial complexity in natural landscapes. This presentation evaluates different watershed disaggregation schemes in the Reynolds Mountain East (RME) sub-catchment in the Reynolds Creek Experimental Watershed (RCEW); Idaho, USA. A benchmark 10-m distributed Noah-MP simulation of the watershed has been performed for reference and to provide input for an initial a posteriori aggregation analysis. The new disaggregation approach uses the K-means clustering algorithm to develop a best estimate of the number of sub-basin tiles and the grid-point membership distribution. This aggregation method is then compared back to the traditional approaches and the distributed simulation. Results show that using a posteriori inputs, 6-7 clusters with non-uniform forcing has the best streamflow reproduction of the distributed simulation as compared to the traditional aggregation types (lump, elevation band and vegetation class). This is due to the fact that the K-means clustering properly captures areas of high water input and runoff that may cut across traditional sub-grid tiles. It is also found that the non-uniform forcing simulations perform better than the uniform cases. Analyses of various a priori inputs to the clustering algorithm indicate that it may be possible to recreate the a posteriori analysis and performance with the correct combination of inputs, as examined through temporal and spatial transposibility tests.