Accurate forecasts of the streamflow expected during late spring and summer in the Upper Klamath River Basin in southern-central Oregon and northern California are used by water management agencies to balance water allocations for agriculture, aquatic habitat, and hydropower-production needs. Streamflow forecasts are also used by irrigation farmers for planning. The forecasts are typically made twice a month starting as early in the water year as December. Multiple regression equations relating real-time snowpack and precipitation conditions to seasonal streamflow volumes have been used for many years in forecasting. However, with warming temperature trends and lower snowpack, such forecasts based on historical data could become less reliable in the future. If the timing and relation of snowpack and precipitation are outside of the range of the historical data used to create the equations, the forecasts become extrapolations. Statistical forecast equations are also limited in their ability to forecast streamflow in groundwater-dominated basins having inter-annual lag. As an additional method for seasonal streamflow forecasting, a physical-process-based hydrologic model employing the Precipitation-Runoff Modeling System (PRMS) was developed in cooperation with the U.S. Bureau of Reclamation for the Upper Klamath Basin in this study. The model was calibrated for the portion of the basin draining into Upper Klamath Lake. PRMS is a deterministic, distributed-parameter, physical-process-based modeling system developed by the U.S. Geological Survey. It simulates daily streamflow, snow, solar radiation, evapotranspiration, surface-water, and groundwater processes within the basin. A model calibration and validation period for water years 2000–15 and water years 1984–99, respectively, was used. The model was calibrated and validated using measured streamflow, snowpack, evapotranspiration, and solar radiation data sets. Interpolated daily precipitation and air temperature data from 32 meteorological stations within and surrounding the Upper Klamath Basin were used as model input. Performance statistics, used to evaluate how well simulated daily streamflow matched with measured streamflow included percent bias, percent relative error, and root-mean-square error. The statistics were computed annually, monthly, for October–March, and for April–September. With the exception of the October–March period, percent bias statistics were all within plus or minus 5-percent for both the calibration and validation periods. Limitations to using the model are error in the precipitation and air temperature input time series data, which include measurement error and error in the spatial interpolation method. Other errors include measured daily streamflow data, which were adjusted for consumptive use losses to make them more closely resemble natural streamflow for calibration.

The model developed for the Upper Klamath Basin can be used to forecast streamflow from the Sprague and Williamson River Basins and inflow to Upper Klamath Lake. Reliable forecasts at these locations are needed for managing water for irrigation, ecosystem health, and power production. Using the models in a forecast application requires assembling model input data sets of anticipated daily precipitation and minimum and maximum air temperature for the period after the date the forecast is made and the end of the forecasted period. These climate data sets can be based on historical or synthetic records, at the discretion of the forecaster. With the Ensemble Streamflow Prediction method, a suite of streamflow scenarios is simulated using multiple years of climate data as model input. The forecasted streamflow is determined from knowing the exceedance probabilities of the simulated streamflows. In this study, the model and the Ensemble Streamflow Prediction method were used to forecast the volume of inflow to Upper Klamath Lake for a 6-month period from April 1, 2015, to September 30, 2015, using a range of climate data sets based on El Niño Southern Oscillation (ENSO) criteria. Because 2015 was a warm phase ENSO period, climate data for 10 warm phase ENSO years from 1980 to 2010 were used as input to the model. The simulated April–September 2015 UKL inflow volume based on measured 2015 climate data was 482,000 acre-feet, which was very close to the 50th percent exceedance probability computed from 10 simulated scenarios that used warm phase ENSO climate input data from 1980–2010.