Analyses & Numerical Prediction

Operations Bulletin No. 50
GASP MODEL OUTPUT FORECAST (MOF Version 2)
November, 1999

A new derivation of Model Output Statistics (MOF) linear regression coefficients has been prepared for use in the automatic generation of seven day forecasts based on the GASP model. (Global Analysis and Prediction System). It will be run twice a day, and will provide forecasts for 689 stations.

The current operational MOF (Version 1) which is based on the LAPS_PT375 model output provides 3-hourly forecasts up to 2 days for 477 stations and will be updated to version 2 in the next few months.

It is planned to implement the 7-day product from base-date/time Thursday 25 November 1999, 0000UTC. It is expected to be available around 06:30/18:30 UTC each day.

Further information is available in the references 'Derivation of seven day model output statistics coefficients' which can be referenced and 'Model Output Forecast Reference Manual' which contains reference material for both Version 1 MOF (2-day) and Version 2 MOF(7-day) systems.

These documents provided much of the information summarised in this bulletin.

Method:

The equations are derived from a linear regression of a historical set of dependent variables provided by a numerical model (as potential predictors) and near concurrent weather elements (as independent variable predictands).

The Dependent Data Period (DDP) was 1 January 1996 to 31 May 1999 with observational data extending 8 days beyond that.

Coefficients are derived independently for each month using model and observational data in a 105-day window centred on the 15^th day of the month. Thus, the largest possible development data size was 420 cases in March and 315 otherwise. Data errors in any model record field caused the entire record to be discarded so that actual sizes were much smaller.

The GASP archive period was shorter than that available for the regional model in Version 1. It was necessary to increase the width of month-centred windows in order to boost the developmental data available. However a wider window introduces more seasonal effects such as greater variance and less representative mean values. A trigonometrical function of the Julian day was devised as an additional potential predictor to produce summer maximum and winter minimum values to partially compensate for seasonal effects.

Forecast elements will be a sub-set of the following

Daily Forecast Fields

*this is the abbreviation used in the coefficient files

Performance:

Verification of Capital City sites for the independent data period September 1999 was performed for Maximum and Minimum temperatures. These were compared with the results obtained by RFC's and Climatology for the same period.

(Results supplied by F.Woodcock (Techniques Development))

The climatology forecasts are taken from 20-year, fitted, smooth daily-mean annual curve for each city and should be more accurate than simply using each city's September monthly mean forecast.

Additionally, there seems to be a consensus that the percentage of errors greater than 5deg is a more useful broad measure of forecast value than the RMSE. This parameter is not yet available for climatological forecasts.

On average the 7-Day Maximum Temperature MOF forecasts were better than climatology at days D+1, D+2 and D+3 but worse than RFC at all intervals.The Minimum Temperature MOF forecasts were better than climatology at all intervals, comparable to RFC for intervals of D+5 or longer, and generally worse than RFC for D+1 to D+4.

These results are limited to one month's forecasts for Capital Cities and verify Maximum and Minimum Temperatures only. No verification results were available for Darwin and the results for Hobart were for D+1 to D+4 only.

Also there are few 7-day forecast quality benchmarks for comparison. Users should be aware that this product has only been tested against a small sample of situations.

Amended coefficients, extracted over a longer period of dependent data, may well improve the product in future releases.

Communications:

The operational MOF will be produced twice daily in NMOC and distributed via CMSS using ftp. All Regional offices will receive a text message which will be sent to the AIFS Product Store.

Nominated stations are bundled into state bulletins for distribution vis CMSS Product Numbers associated with the 7-Day MOF products are:

IDY00231 Australian Capital Territory

IDY00232 New South Wales

IDY00233 Northern Territory

IDY00234 Queensland

IDY00235 South Australia

IDY00236 Sydney Olympics

IDY00237 Tasmania

IDY00238 Victoria

IDY00239 Western Australia

Help Desk:

The first contact point for problems with receiptof the MOF messages should be to the NMOC Help Desk on(03) 9662 2182. This is a 24hr service.

Feedback:

For queries, comments or suggestions for improvements in content please contact:

F.Woodcock@BoM.GOV.AU (MOF development details)

R.Dahni@BoM.GOV.AU (MOF programming)

W.Skinner@BoM.GOV.AU (Daily scheduling )

Documentation:

For fuller details of derivation and implementation, the URL addresses in the references are recommended. The 'Derivation of seven day model output statistics coefficients' contains the breakdown of the verification for each capital city.

References:

Model Output Forecast Reference Manual, Version 2.0
Derivation of seven day model output statistics coefficients, Version 1.0