An intercomparison of approaches for seasonal streamflow prediction in California case study watersheds

Wednesday, April 22, 2015
Andrew W Wood, National Center for Atmospheric Research, Boulder, CO, United States, Pablo A Mendoza, University of Colorado at Boulder, Boulder, CO, United States and Martyn P Clark, NCAR, Boulder, CO, United States
Seasonal streamflow forecasts represent a critical component of strategies for anticipating and managing drought, and are particularly central to management decisions in snowmelt-dependent regions such as California. Operational streamflow forecasts at seasonal lead times in the western US are created via two primary techniques: (1) regress future streamflow on situ observations of rainfall, snow water equivalent, river flow, and occasionally climate system indices that exhibit teleconnections with local watershed climate; and (2) run ensemble hydrologic model simulations that harness runoff predictability through combining initial watershed moisture anomalies with historically observed weather sequences for the forecast period. Since these approaches were operationalized, however, new climate prediction datasets (eg, CFSv2, CFSR), physically-oriented hydrologic models, and statistical techniques have emerged that suggest opportunities for improved operational seasonal forecasting. This presentation describes an in-progress effort to intercompare a range of variations in streamflow forecasting, from traditional statistical approaches to model-based and hybrid approaches that leverage new climate forecasting datasets and modeling resources. Based on 30+ year hindcast simulations for each variation, we present interim results for a small number of California watersheds and comment on the potential for improved seasonal streamflow forecasting through method, model and dataset enhancements.