84th AMS Annual Meeting

Monday, 12 January 2004
Sources of skill and error in long range Columbia River streamflow forecasts: a comparison of the role of hydrologic state variables and winter climate forecasts
Hall 4AB
Alan F. Hamlet, University of Washington, Seattle, WA; and A. Wood, S. Babu, and D. P. Lettenmaier
In the Columbia River Basin (CRB) in the Pacific Northwest (PNW) skillful winter climate forecasts based on the El Niño Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO) are available with lead times of up to about six months (i.e. June 1). Using a hydrologic model and near real time estimates of hydrologic state variables such as soil moisture and snowpack, probabilistic forecasts of streamflow at various basin locations can be produced starting in June (preceding the water year) and proceeding through the snow accumulation season. A useful standard of comparison for these experimental forecasts is a January 1 ESP forecast, which is comparable in skill and reliability to the statistical streamflow forecasts relating winter snow accumulation to spring and summer streamflow that are used operationally for water management throughout much of the western US. In order to understand the role of the various sources of hydrologic predictability, we generate probabilistic forecasts using both climate forecasts and estimates of hydrologic state variables on Oct 1, Nov 1, and Dec 1, and compare them, using a quantitative skill metric, to Jan 1 ESP forecasts based solely on hydrologic state variables. On Oct 1, the skill of the streamflow forecasts is primarily attributable to skill in the winter climate forecasts, with a relatively small contribution from the estimated initial soil moisture state. As snow accumulates in the basin, however, the experimental forecasts on Nov 1 and Dec 1 are increasingly influenced by the persistence of the hydrologic state, becoming more skillful and less sensitive to errors in the climate forecasts. These experiments show that Oct 1 forecasts based on PDO/ENSO forecasts, while frequently better than climatology, are generally not as skillful as Jan 1 ESP forecasts, and that the error characteristics of the Oct 1 forecasts are generally less desirable than Jan 1 ESP forecasts. Results for the Nov 1 and Dec 1 forecasts, however, suggest that by about Dec 1, the skill of the PDO/ENSO forecasts is comparable to the Jan 1 ESP forecasts in most years. Very long-range forecasts made on June 1, which estimate fall soil moisture based on current hydrologic conditions are also shown to be nearly as skillful as the Oct 1 forecasts, although increased likelihood of misclassification of PDO/ENSO state with increased lead time adds some uncertainty to the forecasts. The skill of the PDO/ENSO forecasts described above is also contrasted with the results from previous studies assessing the skill of ensemble streamflow forecasts produced using the same hydrologic model but generated using ensemble predictions from NCEP’s Global Spectral Model, with a lead time of six months.

Supplementary URL: