Improved Regression-Based Streamflow Forecasting Considering Large-Scale Climate Variability

The Natural Resources Conservation Service (NRCS) produces water supply forecasts for most of the streams in the western United States. The NRCS produces forecasts starting in January, when snow course measurements of snow water equivalent were available. Although the seasonal streamflow volume forecasts made by the NRCS are useful, many water supply managers need information at the beginning of the water year in October and would like to see forecasts in the form of a monthly hydrograph. An investigation into the effect of decadal scale variability, as reflected by the Pacific Decadal Oscillation (PDO), show important relationships that may be useful in forecasting. Data from three basins, the Sandy, Skykomish, and Rogue Rivers were split based on the warm and cool phases of the PDQ and correlated to the Southern Oscillation Index (SOI) as a measure of the inter-annual climate phenomenon El Nino-Southern Oscillation (ENSO). All three basins have similar annual hydrographs with a peak in the winter around November due to direct winter runoff and a peak in the spring or summer due to snow runoff. The results show that in cool phases of the PDO, seasonal streamflow is above average and is significantly correlated with the SOI. However, in warm phases of the PDO, streamflow is lower than normal and not as influenced by the SOI as measured by the correlation coefficient. Further, the PDO influences the distribution of flow within the year. As a result, a new seasonal streamflow volume forecasting method is proposed. The new method fits regression equations for both phases of the PDO and mixes the two forecasts by the probability of the current state of the PDO. The model was verified by comparison to a control model that was fit to all of the data, and by a water year 2000 forecast. The results show that the mixed seasonal streamflow volume forecasts better estimate the historical mean. Further, the disaggregated mixed volume forecasts resulted in better estimates of the historical monthly mean and reduced the overall variability of the forecasts