The Accuracy of Weather Predictions, from the Next Day to the Next Season—an Illustration from Australia

In another paper also submitted for presentation to a conference at the 2017 AMS Annual Meeting (28th Conference on Weather Analysis and Forecasting / 24th Conference on Numerical Weather Prediction), "Quantifying How Positional and Timing Errors in the Prediction of Synoptic Scale Systems Extract a Penalty When Verifying Weather Forecasts", the author briefly touches upon trends in the accuracy of day-to-day weather forecasts for Melbourne, Australia, out to the end of Week-2.

For the purpose of the current paper, the accuracy of predictions for Melbourne of four weather elements are considered in detail, this time out to the end of Week 4. The four elements considered are minimum temperature, maximum temperature, probability of precipitation and amount of precipitation. 

The accuracy of official seasonal climate outlooks for Australia is also considered.

The accuracy of the predictions of the four weather elements, especially for lead times out to Day-7, has increased substantially over the years.

To illustrate, the frequency of major Day-1 maximum temperature forecast errors (greater than 5 deg C) is shown to have fallen from about 30 per year during the 1960s and 1970s to fewer than 5 per year now. The percent variance in the observations explained by these Day-1 forecasts has increased from around 50% during the 1960s and 1970s to between 80% and 90% more recently. Day 5-7 maximum temperature forecasts explained just 25% of the variance some 15 years ago, but now explain approximately 50% of the variance. 

Amount of precipitation forecasts have also increased in accuracy. Day-1 amount of precipitation forecasts, having explained just 30% of the variance some 15 years ago, now explain  about 60% of the variance.  

The day-to-day weather forecasts for Melbourne out to the end of Week-4 are weather forecasts that have been generated by an algorithm that interprets the output of the ECMWF Ensemble Control Model. It is found that little skill is evident for day-to-day predictions of weather with lead times beyond Day-14.

The analysis is then further extended to review the accuracy of official seasonal climate outlooks for Australia that are issued by the Bureau of Meteorology. Regarding the accuracy of these seasonal climate outlooks, three elements are considered - precipitation, and overnight and daytime temperature. In each case, some skill is evident, although that level of skill varies from State to State and with time of the year.

To illustrate, the precipitation outlook, which has been issued officially since the late 1980s, displays greater skill in the second half of the calendar year than in the first half and also displays slightly greater skill in areas covering the tropics than in more southern regions.