J4.2 Use of a Priori Benchmarks to Evaluate the Performance of a Physically Based Hydrologic Model

Monday, 11 January 2016: 1:45 PM
Room 242 ( New Orleans Ernest N. Morial Convention Center)
Andrew J. Newman, NCAR, Boulder, CO; and N. Mizukami, M. Clark, A. W. Wood, L. D. Brekke, and J. R. Arnold

Model benchmarking using a priori performance expectations helps define an expected upper limit of performance expectations, aids in the identification of model areas needing improvement, or even model workflow issues. A priori benchmarks are particularly useful in large sample (large scale) hydrologic modeling endeavors. For example use of a large sample a priori benchmark helps to set expectations of performance across a large range of hydroclimatic conditions for a given forcing dataset, or aid in quick identification of basins with technical implementation issues such as incorrectly indexed forcing data time series, where manual examination of 100's of basins is impractical. In this presentation we use a large sample, watershed scale, a priori benchmark database to evaluate the performance of the Variable Infiltration Capacity (VIC) hydrologic model using a variety of calibration configurations.

The benchmark database consists of hundreds of basins across the CONtinental United States (CONUS) with a corresponding optimized conceptual Snow-17/SAC/Unit Hydrograph (UH) modeling system. This modeling system is highly flexible, and, when optimized, a priori information on the expectations of model performance. We configured VIC to run in a similar mode to the Snow17/SAC model: lumped watershed response with unit hydrograph routing. VIC was then run for four calibration configurations: 1) default VIC parameters with only calibrated UH, 2) basic soil parameter and UH calibration, 3) additional soil parameter and UH calibration, and 4) additional calibration of transpiration. We will discuss the results in the context of 1) model agility, e.g. can any of the VIC configurations perform as well as the conceptual model, and 2) if not, can we begin to understand why. We will also discuss how the benchmark is useful to identify errors in the iterative process of setting up models for many basins, running calibration simulations, and diagnosing the results.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner