Climate Impacts Group home page Land Surface Hydrology Research Group home page National Oceanic and Atmospheric Administration GEWEX Americas Prediction Project

One year lead-time experimental streamflow forecasts for the Columbia River at The Dalles

Forecast Methods

Background and Description of Methodology

The methods used to produce these retrospective forecasts are described in more detail by Hamlet and Lettenmaier (1999a).

Investigation of interactions between recurrent patterns of climate variability and Columbia River streamflow have revealed strong correlations between Columbia River flow from April-September and two recurrent climate phenomena: the El Niño/Southern Oscillation (ENSO) (see, for example, Battisti and Sarachik) and the Pacific Decadal Oscillation (PDO) (Mantua et al. 1997).

These climate patterns, and particularly the phase of ENSO, can be forecast with lead times of six months to one year, and are used here in conjunction with a macro-scale hydrology model driven by resampled historical meteorological data to create streamflow forecast ensembles for the coming water year. The forecasts for the water year (Oct-Sept) can typically be delivered in June of the preceding year, when good climate forecasts are typically available (Hamlet and Lettenmaier 1999a)

Several web sites present more information on Pacific Northwest climate variability and ENSO forecasting in more detail:

The Climate Categories

Combinations of two climate indicators are used to create six climate categories for investigations of the relationship between streamflow and climate states and for retrospective forecasting.

The ENSO climate categories are created using the NINO 3.4 index averaged from December-February. Anomalies more than 0.5 standard deviations above the mean are assigned as ENSO positive (El Niño), and anomalies more than 0.5 standard deviations below the mean are assigned to ENSO negative (La Niña). Other years are considered ENSO neutral. The phase of ENSO can be forecast with lead times of about one year using coupled ocean atmosphere models and/or statistical procedures.

PDO, on the other hand, is assigned a bimodal phase based on decadal scale oscillations that have occurred in the historic record. From 1948-1976 the PDO index was predominantly negative (cool phase), and from 1977 to approximately 1998 the PDO index was predominantly positive (warm phase) (Mantua et al, 1997). Although a true PDO phase shift from warm (positive index values) to cool (negative index values) in 1998 has not been conclusively confirmed, the forecasts have been made on this basis.

The interannual value of the PDO index cannot be forecast with skill at the present time. By assuming persistence of the current decadal phase, however, and identifying transitions by streamflow events more than 1.5 standard deviations from the mean in opposition to the dominant trend, the bimodal phase of the PDO can be "forecast" with some apparent skill.

(see Predicting PDO Transitions)

Six climate categories shown below in the scatter plots are created using combinations of the ENSO and PDO phase. Red triangles in the figures show average April-Sept hydrologic response of water years in the climate category. Blue circles show the hydrologic response of those water years not in the category. The red and blue horizontal lines show the average difference between the those years in the category and those outside the category. Climate Category 1, for example, shows a strong signal for low-flow response from the Columbia Basin.

click image to enlarge

Figure 1 Snake River system.

Creating the Streamflow Forecast

The Variable Infiltration Capacity (VIC) macro-scale hydrology model covering the Columbia Basin above The Dalles at 1/4 degree resolution is used with gridded meteorological data derived from station data to simulate streamflow associated with different water years. The model exhibits some bias under certain conditions, but successfully captures the variability in streamflow at The Dalles from 1948-1997 quite well (Hamlet and Lettenmaier, 1999b). Technical information on this hydrologic model is available from links to the UW Hydrology Home Page below:

Streamflow outcome during the forecast year is primarily a function of initial soil moisture conditions and meteorological events (precipitation and temperature) which will occur in the forecast period (Oct-Sept for these forecasts). To incorporate these variables in the forecast method, the hydrology model is initialized using a meteorological time series from the historic record which is similar in character to the year preceding the forecast year. Then meteorological data from different years within the forecast climate category are used to drive the initialized hydrology model to create the forecast ensemble members. So, for example, water year 1998 was forecast as Climate Category 4 (cool PDO/El Niño), so water years from Climate Category 4 were used to drive the hydrology model after initialization using water year 1974 (an extremely wet year similar in character to 1997).

Retrospective Streamflow Forecast Archive

Retrospective forecasts from water year 1951 to 2000 have been produced and are available at: