Experimental Empirical Medium-Range Forecasts


Welcome to the CDC experimental medium range forecast web site. We are currently making wintertime and summertime forecasts of anomalous 250 and 750 mb streamfunction out as far as six weeks ahead. Although we hope others will find these forecasts interesting, they are very experimental and the Climate Diagnostic Center is not responsible for any loss occasioned by the use of these forecasts.

Forecasts are done for two different averaging periods. Weekly forecasts are made for week 2 (the average of 8-14 days from the forecast initialization) and week 3 (the average of 15-21 days from the forecast initialization). 15-day average forecasts are made for weeks 3/4 (the average of 16-30 days from the forecast initialization) and weeks 5/6 (the average of 31-45 days from the forecast initialization).


We make two types of forecasts, which we label the LIM(Linear Inverse Modeling) forecast and the Combined forecast. Our LIM forecast procedure (Winkler et al. 1999, 2001; Newman et al. 2001) is similar to that used to make predictions of tropical IndoPacific sea surface temperature anomalies (Penland and Sardeshmukh 1995). A detailed description of the empirical method is available for download and a more thorough documentation of the success of this model in medium-range forecasting is available online. In summary, a dynamical operator of evolving low-frequency atmospheric perturbations is derived from 30 years of NCEP Reanalysis streamfunction and diabatic heating. In turn this operator can be used to generate low-frequency forecasts at any forecast time.

Our second forecast, the Combined forecast, is made by combining the LIM technique with the week 1 streamfunction forecast made by the NCEP MRF ensemble. For example, a week 2 Combined forecast is made by

  • 1) making a week 1 LIM forecast
  • 2) replacing the week 1 LIM streamfunction forecast with the week 1 MRF ensemble streamfunction forecast, but
  • 3) keeping the week 1 LIM tropical heating forecast
  • 4) making another one week LIM forecast (= week 2) from these week 1 states
The procedure to make the week 3 combined forecast is identical, except that in step 4 a two week LIM forecast (= week 3) is made. We are currently not making a combined forecast for the 15-day averages.

On this webpage we apply both techniques to realtime, low-frequency streamfunction and diabatic heating in an attempt to provide credible medium-range forecasts. Throughout the winter (Dec 1 - Feb 28) and summer (Jun 1 - Aug 31) we will be updating these forecasts daily.

About the Data

The streamfunction used in this study is measured at 750 and 250 hPa in the Northern Hemisphere. The diabatic heating is integrated from surface to tropopause and is restricted to the Tropics. Both fields are dynamically consistent, that is the large-scale mass and vorticity budgets are satisfied, as determined from the generalized baroclinic chi problem (Sardeshmukh 1993). Also, each variable has the first 3 harmonics of its seasonal cycle removed and is filtered with either a 7-day or 15-day running mean. For winter forecasts the streamfunction is then projected onto its leading 30 EOFs and the heating onto its first 7 EOFs. During summer we use the leading 30 EOFs of streamfunction and 20 EOFs of heating.

Future plans

We hope to extend our model to cover the entire year. We have also deliberately made it fairly simple, with little information about the vertical structure of the atmosphere. Adding more levels (particularly some surface information) and/or different variables may improve these forecasts. At least, we plan to examine whether this can be done. Finally, we may try to use this forecast method to predict more useful quantities such as surface temperature and perhaps precipitation.

Links from this page

  • Submitted manuscripts: Winkler et al. 2001 and Newman et al. 2001
  • Short article (from Conference proceedings) discussing this work.
  • A more thorough analysis of the forecast skill of the LIM forecasts
  • Diabatic heating web page (the Chi problem)
  • Tropical SST linear inverse model forecasts


    Penland, C., and P. D. Sardeshmukh 1995: The optimal growth of sea surface temperature anomalies. J. Climate, 8, 1999-2023.

    Sardeshmukh, P.D. 1993: The baroclinic chi problem and its application to the diagnosis of atmospheric heating rates. J. Atmos. Sci., 50, 1099-1112.

    Winkler, C.R., M. Newman, and P.D. Sardeshmukh, 1999: An empirical low-frequency forecast model incorporating diabatic forcing. Proceedings, 24th Climate Diagnostics and Prediction Workshop, Tucson, AZ, 303-306.

    Winkler, C.R., M. Newman, and P.D. Sardeshmukh, 2001: A linear model of wintertime low-frequency variability. Part I: Formulation and forecast skill. Submitted to J. Climate. Manuscript available.

    Newman, M., P. D. Sardeshmukh, and C. R. Winkler, 2001: Tropical heating impacts on extratropical predictability beyond the medium range. Submitted to Nonlinear Processes in Geophysics. Manuscript available.

  • main page week 2 and 3 weeks 3/4 and 5/6 TECHNIQUE verification heating data

    Document maintained by: Matt Newman
    Last modified: Thu Jun 14 14:20:14 MDT 2001