My temperature forecasts for the next 15 days

Jump to the forecasts or to the verification

Brief outline of the method

  1. The starting point for the forecasts are NCEP ensemble forecasts (available at a resolution of 1x1 (latitude-longitude) degrees); I use the control run and 5 ensemble members - 6 forecasts in all.
  2. The forecasts cover the period 6-384 hours after the analysis time - providing 15 days of daily forecasts.
  3. For a given location (at which daily maximum and minimum temperatures must also be available), the average maximum and minimum forecast temperatures at each forecast time are determined from the model output.
  4. Analysis of previous forecasts for each of the 15 forecast lengths (1 day, 2 days, etc) against the subsequent verifying observations for a period of up to 30 days before the start time of the current forecasts, produces a bias correction which is applied to the model-derived forecast to generate my station forecast (the 'issued' forecast). This allows the bias to vary with time of year, etc.
  5. The issued forecasts relate to the period 0600-1800UTC for maximum temperatures and 1800-0600UTC for minimum temperatures.
  • Once the observations become available, they are compared against the issued forecasts in order to analyse the error in the issued forecast, both as a function of forecast length (0-day, 1-day, 2-day, etc.) and calendar date (by combining all forecasts made for that date).
  • More details of these validation results will be posted once sufficient forecasts have been created.

Guide to the forecasts

The forecasts contain:

  • Recent observations of minimum, maximum and mean (i.e. the average of minimum and maximum) temperature for up to 10 days. Observations from the sites are extracted automatically from SYNOP reports.
  • Forecasts for 15 days ahead of
    • minimum (1800-0600UTC) temperature,
    • maximum (0600-1800UTC) temperature and
    • mean temperature.
  • An indication of the forecast confidence based on the standard deviation of the value of the error (equal to the issued forecast minus the observation) of the past 30 forecasts - for each of the forecast lengths (0 days, 1 day, 2 days, etc.). In general the confidence will get less as the forecast length increases.
    • 'Hi' denotes a standard deviation in the error of under 1.5C (a 'high' accuracy forecast)
    • '-' denotes a standard deviation in the error of 1.5-3.0C (mid-range accuracy), while
    • 'Lo' denotes a standard deviation in the error of over 3.0C.
  • Note that no account has been yet taken of the forecast uncertainty implied by the spread of the individual ensemble forecast values, as it is considered that interpolation from the gridded data to the observation location probably creates a larger uncertainty.

The latest observations

The latest forecasts

Verification

To see how good (or bad) recent forecasts actually were, see these comparisons of observations for yesterday against previous forecasts valid for yesterday:

24-hour forecasts,
48-hour forecasts,
72-hour forecasts.

The image above shows the standard deviation (SD) and the absolute value (AB) of the forecast error (where the error is the difference between the forecast and the subsequent observation), averaged across all sites for the 30-day period indicated.

  • In general the forecast quality reduces with increasing forecast length with maximum temperature forecasts prone to a larger error than those for minimum temperatures.

  • Forecasts for coastal locations have a slightly smaller error - presumably due to the smaller diurnal ranges caused by oceanic influence.

Comments?

Your comments are welcome, and can be posted to me.

Disclaimer

The forecasts listed above use model data that are kindly provided by NCEP and a computer algorithm of my own devising. Neither my employers nor myself are professional weather forecasters and, while these forecasts may give an indication of conditions to come, we cannot accept any responsibility for the accuracy of this forecast.

The weather over Europe is part of an ever-changing system. The data on these pages should be used as a guide only and there are times when any forecast beyond about 2-3 days is likely to prove unreliable. In addition, it it unwise to make inferences about other weather elements, e.g. rainfall, from temperature forecasts alone.

Always check the latest forecast from the appropriate national forecast office before undertaking any weather-dependant activity.




You are visitor number [an error occurred while processing this directive] to this page since 4 June 2008.