A Controlled Approach to Model Comparison and Improvement: Insights from the Reynolds Creek Critical Zone Observatory

Abstract:

The current generation of models has followed a myriad of different development paths, making it difficult to identify a clear path to model improvement. Model comparison studies have been undertaken to explore model differences, but have not been able to meaningfully attribute inter-model differences to individual model components because there are often too many differences among the participating models. Model comparison studies have therefore provided limited insight into the causes of differences in model behavior, and model development has relied on the inspiration and experience of individual modelers rather than on a systematic analysis of model shortcomings.

This presentation introduces a unified approach to process-based hydrologic modeling to enable controlled and systematic evaluation of different modeling approaches. Our model framework, called the Structure for Unifying Multiple Modeling Alternatives (SUMMA), formulates a general set of conservation equations, providing the flexibility to experiment with different spatial representations, different flux parameterizations, different model parameter values, and different time stepping schemes. The SUMMA framework presented here enables users to decompose the modeling problem into the individual decisions made as part of model development, and evaluate different model development decisions in a systematic and controlled way.

We present case studies to illustrate the use of SUMMA to select among competing modeling approaches, based on observed data from the Reynolds Creek CZO. Specific examples of preferable modeling approaches include adjustments to the shape of the below-canopy wind profile to improve simulations of below-canopy snowpack, and explicitly representing distributed lateral flow processes to improve simulations of riparian transpiration and streamflow. Results also demonstrate that changes in parameter values can make as much or more difference to the model predictions than changes in the process representation, emphasizing that improving model fidelity requires a sagacious choice of both process parameterizations and model parameters. Moving forward, we envisage that SUMMA can accelerate the diagnosis and correction of model structural errors, and improve the characterization of model uncertainty.