Real-time demonstration and evaluation of over-the-loop medium-range ensemble streamflow forecasting

The US National Weather Service River Forecasting Centers are beginning to operationalize short range to medium range ensemble predictions that have been in development for several years. This practice contrasts with the traditional single-value forecast practice at these lead times not only because the ensemble forecasts offer a basis for quantifying forecast uncertainty, but also because the use of ensembles requires a greater degree of automation in the forecast workflow than is currently used. For instance, individual ensemble member forcings cannot (practically) be manually adjusted, a step not uncommon with the current single-value paradigm, thus the forecaster is required to adopt a more 'over-the-loop' role than before. The relative lack of experience among operational forecasters and forecast users (eg, water managers) in the US with over-the-loop approaches motivates the creation of a real-time demonstration and evaluation platform for exploring the potential of over-the-loop workflows to produce usable ensemble short- to-medium range forecasts, as well as long range predictions. We describe the development and early results from this platform, which results from a collaboration between NCAR and two federal water agencies, the US Army Corps of Engineers and the US Bureau of Reclamation. Focusing on small to medium sized headwater basins around the US, and using multi-decade series of ensemble streamflow hindcasts, we assess the skill of daily-updating, over-the-loop forecasts driven by a set of ensemble atmospheric outputs from the NCEP GEFS for lead times from 1-15 days.