Snowmelt and Streamflow Forecasting in a Milieu of Climate Warming: Can Operational Models Handle It?

Abstract:

Forecasting of snowmelt and resultant streamflow in mountainous regions is done operationally using two styles of models: statistical and conceptual simulation models. The former relies entirely on empirically-derived statistical relationships between predictor variables (snow water equivalent, precipitation, etc.) and the target streamflow quantity. The latter, while not explicitly based on empirical statistical relationships, nevertheless has some empiricism built into model components and parameterizations. Both model styles are therefore based to a greater or lesser extent on stable relationships between climate and hydrologic response.

In addition, both styles of models rely on repeatable patterns of weather behavior in the future, after the forecast issuance date. For statistical models, future weather is implicit in the fitted model coefficients and the width of the prediction error bounds. For conceptual simulation models, future weather is represented by the ensemble of historical weather traces used to force the model into the future and generate an ensemble of possible streamflow predictions given the current watershed state.

Both of these facets of model dependence on climate arouse concern regarding the ability of these models to continue to represent hydrologic response adequately in a rapidly warming climate. Specific susceptibilities include: (1) Earlier disappearance of snowpack, leading to removal of the primary forecast signal; (2) Changes in future weather patterns, leading to changes in snowpack accumulation or increases in variability and persistence, and (3) Changes in fundamental hydrologic input-output behavior.

This presentation diagnoses these climate warming vulnerabilities in models used for operational snowmelt-streamflow forecasts, illustrated with specific examples.