J4.3 CONUS-Scale Model Evaluation and Benchmarking for National Streamflow Prediction

Monday, 11 January 2016: 2:00 PM
Room 242 ( New Orleans Ernest N. Morial Convention Center)
Aubrey Dugger, NCAR, Boulder, Colorado; and D. Gochis, A. RafieeiNasab, J. McCreight, D. Yates, L. Karsten, K. Sampson, W. Yu, M. Somos, F. Salas, and D. Maidment

Coupled land surface-hydrologic models suffer from highly dimensional parameter and variable spaces, making model calibration and verification somewhat of an art in balance. When applying such complex models to an area as diverse as the contiguous U.S., we have a formidable challenge in digesting the quantities of output variables and attempting to translate these outputs into a cohesive understanding of system behavior. Meeting this challenge requires the development and implementation of a clear model evaluation and benchmarking strategy. In this presentation, the R-based, community ‘rwrfhydro' model evaluation system is described in the context of the evaluation of several configurations of a national hydrologic analysis and prediction system. The community WRF-Hydro system is being adapted into 4 different operational configurations for national hydrologic prediction at various timescales: (1) analysis and assimilation, (2) short-range, (3) medium-range configurations and (4) ensemble long-range configuration. After providing an overview of the structure and function of the ‘rwrfhydro' tool, results from analysis and assimilation, short-range and medium range national configurations of WRF-Hydro are presented. Results from these model configurations are contrasted against both pilot-basin experimental test cases and a prototype version of WRF-Hydro that was configured during the National Flood Interoperability Experiment during the summer of 2015. We evaluate hydrologic analyses and forecasts at ~1000 USGS GAGES-II reference basins distributed across CONUS. These reference basins allow us to examine streamflow with minimal impacts due to unaccounted for human influences (e.g., reservoirs, diversions). We first stratify model prediction skill at a range of timescales by basin geophysical setting, climate, and size. We then attempt to attribute streamflow error to likely sources based on background analysis of precipitation biases, land surface model processes, and routing processes. We evaluate streamflow and secondary water fluxes (soil moisture, snow, evapotranspiration) under alternative model physics at a select set of representative basins to test our hypotheses of dominant error sources. Finally, based on our results, we propose an error-reduction strategy for the next generation CONUS-scale hydrologic forecasting system planned by the National Weather Service using the WRF-Hydro framework.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner