Australian Weather News   Site:  Home   Daily Weather Summaries    Search   Feedback   About 
Weather:  Forecasts   Current   Recent & climate   Severe   Know-how   Mobile
Warnings & Forecasts  8-day guidance forecasts   Forecast charts: Rain  Pressure  Heat/Cold  Storms  Model links
About Operational Consensus Forecasts

Full Technical summary
Detailed description including verification data (31 March 2005)
Description of upgrades (15 July 2009)

These forecasts are provided by the Bureau of Meteorology Operational Consensus Forecasts system and should be used in conjunction with official forecasts.

Operational Consensus Forecasts (OCF) provide the best-yet automated forecasting guidance for specific sites around Australia. OCF combines the output from up to 8 computer weather prediction models, gives weighting to those that have recently performed best, and learns from its mistakes in real time by making routine comparisons with actual weather. It was developed by the Weather Forecasting Group at the Bureau of Meteorology Research Centre (BMRC), and in verification it generally performed better than the official (human) Bureau forecasts. It became operational on 8 March 2005 and is now available unofficially for the community.

OCF provides forecasts four times daily of up to 7 weather elements for over 600 locations around Australia for up to 8 days ahead. The elements forecast are:

  • Rainfall amount (mm) in the 24 hours FROM 9am daily (note FROM)
  • The percentage of models forecasting more than 0.2mm of rain compared to the total number of models used in preparing the forecast. So 75% tells you that 3 out of 4 or 6 out of 8 models used to prepare the forecast for this day were forecasting more than 0.2mm rain.
  • Maximum air temperature FROM 9am (note FROM)
  • Minimum air temperature to 9am
  • Minimum grass (ground, terrestrial) temperature to 9am
  • Hours of sunshine during the calendar day
  • Evaporation amount (mm) in the 24 hours FROM 9am daily (note FROM).

The locations for which forecasts are available are the Bureau's observing network sites. Because OCF uses direct feedback from the measurements at these sites, forecasts are only given for elements where at least 50% of observations over the past 30 days are available at the site. Most sites report rainfall and the maximum and minimum air temperature, but grass minimum, sunshine and evaporation are measured by relatively fewer stations and in most districts there will only be one or two sets of forecasts for these elements. Sunshine, in particular, is now measured at relatively few locations.

The models from which OCF is derived are the 00, 06, 12 and 18 UTC (4 and 10am and 4 and 10pm EST) runs of the Australian ACCESS-G, ACCESS-R and high-resolution ACCESS-C models, as well as the overseas ECMWF, GFS, UK and JMA models. However, OCF is not simply an averaging of all available computer guidance for each location.

Models suffer biases, both over time and in different situations. Further bias is introduced when output from a model is used to produce a forecast for a specific site and time. OCF overcomes these bias problems by using a bias-correction scheme using a 30-day running window of observations for each site for which it forecasts. A bias estimate is calculated for each component model over this period, and is then removed from the current forecasts. In short, the forecasts remove, as far as possible, known real-time bias for each model, then produce forecasts that favour the models that have performed best over the past 30 days.

Critical factors in the preparation of OCF twice daily are the arrival times of the models. The limited area models (ACCESS-R and ACCESS-C) become available 2 to 4 hours after observation time but only have a 36 to 72 hour outlook. The main global models take 6 to 10 hours to become available, but extend to 8 days or more. In the operational real world, model runs are also sometimes delayed. There is therefore no "perfect" time to begin the compositing process that produces OCF, and the compromise that has been struck also has an eye to presenting the latest information at the times it is most needed in the preparation of forecasts (early morning and late afternoon).

Verification of OCF was carried out during the 12 months to 31 May 2004, and further work has continued as models have been upgraded and refined. In summary, OCF outperformed official Regional Forecasting Centre maximum temperature forecasts for days 1 and 2, but was worse at days 3 and 4. However, it outperformed official minimum temperature forecasts at all projection hours.

AWN presents the OCF data district by district in a similar format to the AWN Daily Climatic Data for each month. The districts are the official rainfall districts for Australia, and are similar to the Bureau's forecast districts.