Trends in the skill of weather prediction at lead times of 1-14 days

Unique, multi-year datasets of weather observations and official and experimental predictions are used to document trends in weather forecast accuracy and the current level of forecast skill specifically for Melbourne, Australia. The data are applied to quantify prediction skill out to Day-14 for maximum and minimum temperature, and for precipitation amount and probability. An innovative statistical analysis is applied to the data. This analysis clearly demonstrates the need for long time series of forecasts in order to reliably assess long-term trends. The accuracy of the current official Day 5-7 forecasts is found to be similar to that of Day-1 forecasts from 50 years ago. The accuracy of experimental Day 8-10 forecasts is comparable to that of the Day 5-7 forecasts, when they were first officially provided 15 years ago. Some overall skill, albeit limited, is evident out to Day-14 and significance testing indicates that it is unlikely that this apparent skill arose by chance. The results provide evidence of deterministic weather forecast skill out to the hypothesised 15 day limit on such predictions. However, in so doing, the results raise the possibility that the limit may be breached at some stage in the future.