7th International Conference on Southern Hemisphere Meteorology and Oceanography

15.4

Trends and volatility in the accuracy of temperature forecasts

Shoni S. Dawkins, Bureau of Meteorology, Melbourne, Vic., Australia; and H. Stern

The purposes of the current paper are twofold:

(1) To investigate the relationships between the accuracy of Melbourne day-1 maximum temperature forecasts and the time of year (month), and the type of initial synoptic situation.

(2) To expand the study to incorporate a preliminary analysis of trends in the accuracy of forecasts out to 7 days.

There has been a great deal of work carried out on trends in the accuracy of Melbourne's temperature forecasts, largely on account of the sharp improvement that can be documented since the 1970s (Stern, 1996). The aforementioned work came to some preliminary conclusions about how that accuracy varies with the type of synoptic situation. A more recent paper (Stern and Dawkins, 2003) utilised forecast accuracy data to price the cost of a financial product that could be used to guarantee the reliability of temperature forecasts.

It was found that the accuracy of the forecasts varies substantially, depending both upon the time of year and the type of weather pattern.

Regarding the time of the year, the highest day-1 Root Mean Square (RMS) errors, of about 3.5 deg C, are registered during the summer months, namely, December, January and February. This is attributed to these months being associated with the highest inter-diurnal variability. By contrast, RMS errors during the winter months are below 2 deg C for day-1 forecasts.

Regarding the type of weather pattern, larger errors are associated with anticyclonic northerly flow on the forecast day, often during summer with an approaching front. Smaller errors are associated with post-frontal southerly cyclonic situations, and particularly during the winter months.

It has also been found that the trend towards forecast improvement, already established since the 1970s, accelerated during the past year or two. Specifically, the 1980s was a decade of rapidly decreasing errors, a feature attributed to advances in both NWP and remote-sensing technologies.

Interestingly, there appears to have been little further improvement during the 1990s, but a sharp improvement in day-1 forecasts is registered in 2001. This is attributed, at least in part, to an influx of additional meteorologists and the implementation of new techniques.

Forecasts out to day-2, day-3 , and day-4 have steadily improved. Indeed, the day-4 maximum temperature forecasts are now showing an RMS error of 0.5 deg C less than the day-1 forecasts were registering during the 1960s and 1970s. Forecasts out to days 5, 6, and 7 are also improving, but to a lesser extent.

extended abstract  Extended Abstract (136K)

Session 15, Weather and Forecasting III
Friday, 28 March 2003, 9:00 AM-11:30 AM

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page