1-Year Lead Time Experimental Streamflow Forecasts for the Columbia River at The Dalles
Center for Science in the Earth System Climate Impacts Group
Department of Civil and Environmental Engineering
University of Washington
Alan F. Hamlet
Contact Information for Alan F. Hamlet
Dennis P. Lettenmaier
Last Updated 10/18/2005
The forecasts provided here are experimental and are provided to familiarize researchers and potential users with the methods and performance of the techniques used. The forecasts are offered with no guarantee of appropriateness for any particular purpose. Use of these forecasts for any purpose other than academic research is therefore entirely at the discretion of the user, and the authors and the JISAO Climate Impacts Group accept no responsibility for the consequences of such use.
1)Historical temperature and precipiation data are now available from 1915-2003. Having more data increases the sample size of ENSO based streamflow forecasts. For example, 27 warm ENSO events occurred between 1916-2003 as opposed to 18 warm ENSO events between 1950 and 2003. The new data set also includes a number of very dry years in the 1930s and 1940s that enhance the understanding of drought risks.
2)For the 2006 forecast, we have used an explicit estimate of a range of Nino3.4 anomalies forecast by the ECMWF Nino3.4 plume. This is change from past years, for which a categorical forecast of warm neutral or cool ENSO were used.
3)For the past two forecast iterations (2005 and 2006), streamflow forecasts on this site have been bias corrected. This allows a direct comparison with observed naturalized streamflow values.
4)New methods have been developed to predict the year to year value of the PDO index with long lead times, creating a more objective forecast based on the relationship to the PDO index. The forecasts are now composited according to the interannual value of the PDO index as warm, neutral, or cool PDO. The longer temperature and precipitation record also provides more realizations in each of these categories, which improves estimates of uncertainty.
5)The UW experimental West-Wide streamflow forecasting system has been operating on a quasi operational basis through water year 2005 with forecast updates once a month. Review the forecast performance for 2005 at: UW West Wide Seasonal Hydrologic Forecast System
6) In all likelihood, this will be the final year that these special long-range forecasts for The Dalles will be produced as a separate forecast product here at the UW. Long range streamflow forecasts comparable to those found here will continue to be available from the UW West Wide Hydrologic Forecasting System, however. If you need assistance in finding a comparable product on the West Wide site, please feel free to contact me.
Links to Forecast Methodology and Forecast Archive
Real-Time Experimental Forecasts for Water Year 2006
Download bias corrected VIC Data
Download bias corrected VIC Data in MAF
The forecast for water year 2006 displayed below was produced on about Sept 15, 2005 based on a forecast of NINO3.4 anomalies between -0.4 and 0.6 for the winter of 2005-2006 (left panel below). The ENSO composite forecast is further composited based on neutral PDO conditions to produce the PDO/ENSO forecast in the right panel below.
For this forecast, hydrologic initial conditions from September 1 from the UW West Wide Hydrological Forecast System were used to estimate the initial soil moisture.
The figure below shows a forecast based on an assumption of Nino3.4 index anomaly between -0.4 and 0.6 (no assumption of warm or cool PDO). The heavy black lines on the figure are the maximum and minimum simulated streamflow for the period from 1916-2002, and are provided as a common reference for the forecasts. The long term mean for all years from 1916-2002 is shown in red and the ensemble mean is shown in blue. The forecast (left panel below) shows a central tendency (blue line) near the long term mean and a range of uncertainty similar to the climatology. This probability distribution is typical of ENSO neutral years, and there are no strong signals of above or below average streamflow.
New forecasting techniques for the PDO index (Newman et al. 2003) predict neutral PDO conditions for winter 2005-2006 based in part on the persistence of conditions last year PDO index, The ENSO/PDO composite is essentially the same as the ENSO composite in this case.
What is the Performance Record of the ENSO Forecasts Since 1998?
This figure from the data archives of the IRI shows the time history of the NINO3.4 time series (sst anomalies in degrees C). The retrospective definition of ENSO state that we use for forecasting is based on 0.5 standard deviations from the long-term mean (+/- 0.47 degrees C) of the NINO3.4 index averaged from December to February PDO ENSO Tables. We have been interpreting ENSO forecasts delivered in about the first week of June since 1997. ENSO forecasts for water year 1998, 1999, 2000, 2002, and 2003 (warm, cool, cool, neutral, warm respectively) made in June proved accurate. The ENSO forecast for 2001 predicted the highest likelihood of ENSO neutral conditions in June, but during the winter the sea surface temperatures exceeded the threshold for the cool ENSO category, so our interpretation of the ENSO forecast in June was in error. The streamflow forecast for 2001 would have been even more in error if we had had a correct forecast of cool ENSO, so the incorrect ENSO forecast was not the predominant source of forecast error in 2001 as it turned out. For water year 2004, the ENSO forecast available in June, 2003 for the winter 2003-2004 proved to be incorrect. By October 1, however, an updated forecast of ENSO neutral conditions was released. This forecast proved to be accurate.
How did we do in water year 2005?
The long-range forecast in 2005 proved to be very skillful, with the observations very closely matching the ensemble mean. Forecast Archive.
What's Happening with the PDO?
New methods developed by Newman et al. 2003 have resulted in skillful predictions of the PDO index with long lead times using simple regression techniques. These methods exploit the fact that the PDO is very strongly autocorrelated in time and is affected by ENSO variability in a systematic manner. Modest warm PDO conditions last winter and an expected neutral ENSO event this winter result in a forecast of PDO neutral conditions for mid winter 2005-2006.
Research using retrospective forecasts has shown that the skill of forecasts using an interannual estimate of the PDO index are comparable to those produced using perfect knowledge of predominantly warm or cool PDO epochs. (See the presentation at: hamlet_hydrologic_predictability_cig_water_wrks_kelso_oct_2003.ppt ) We use these PDO forecasts composite the ENSO streamflow forecast according to the interannual forecast of the PDO index for the coming year (as above). That is, only the historic years for which PDO index was in the neutral category are shown.
Newman M., Compo GP, Alexander MA, 2003, ENSO-forced variability of the Pacific decadal oscillation, J. of Climate, 16 (23): 3853-3857