Operational Streamflow Forecasts in the Columbia River Basin

As the demand for water increases, the effective management of limited water resources becomes more important. Managers of highly utilized river basins must balance complex tradeoffs between water supply, hydropower, flood control, irrigation, fishing, and recreation. The private sector is involved in the development of decision support systems for water management applications. In order to quantify the risk and tradeoffs associated with medium-range (1-10 dy) operating strategies, numerical weather prediction models can employed. One way to produce an ensemble streamflow forecast is to drive a distributed hydrologic model with probabilistic outlooks of temperature, precipitation, and other meteorological variables.

In this work, forecasts from European Center for Medium Range Weather Forecasting (ECMWF) are used to force the Variable Infiltration Capacity (VIC) model. Elevations were obtained from CGIAR version 4 SRTM data; stream networks from USGS HYDRO1k; soil textures from FAO data; and vegetation types / land cover information from UMD land cover data. The surface energy balance in 0.25° (~25 km) cells is closed through an iterative process operating at a 6 hour timestep. Output fluxes from cells in the basin are combined through one-dimensional flow routing predicated on assumptions of linearity and time invariance. These combinations lead to daily mean streamflow estimates.

Operational streamflow forecasts in the Columbia River Basin have been active since October 2012. These are “naturalized” or unregulated forecasts. Moisture stored in snowpack during the winter is typically released in spring and early summer, contributing to large increases in streamflow. In 2013, increases in daily mean flow of ~2600 m³/s (~48% of average discharge for water years 1879-2012) or greater were observed at The Dalles Dam during the following periods: 29 March to 12 April, 5 May to 11 May, and 19 June to 29 June. Precipitation and temperature forecasts during these periods are shown along with changes in the model simulated snowpack. We evaluate the performance of the ensemble mean 10 days in advance of each of these three events, and comment on how the distribution of ensemble members affected forecast confidence in each situation. These forecasts can be tailored to the user's needs.