Bridging the middle ground between medium range weather forecasting and seasonal climate outlooks: two-week day-to-day weather forecasts and monthly climate outlooks

In Australia, the Bureau of Meteorology issues day-to-day weather forecasts out to the end of week one, whilst its Seasonal Climate Outlook (SCO) is issued about two weeks prior to the beginning of the season for which the outlook is valid.

Work aimed at automatically generating day-to-day weather forecasts out to the end of week two, based upon statistical interpretation of the output of NOAA's Global Forecasting System 16-day NWP model, is presently underway. Furthermore, work aimed at automatically generating worded monthly climate outlooks, based upon statistical relationships between historical monthly climate anomalies and various measures of the ENSO, Indian Ocean Dipole, and Madden-Julian Oscillation phenomena, is also presently underway. These works have as their motive a desire to bridge the middle ground between medium range weather forecasting and the SCOs. It is the purpose of this paper to report preliminary findings.

A "real-time" trial of a system used to generate Day-1 to Day-7 forecasts, by mechanically integrating (that is, combining) judgmental (human) and automated predictions, has been ongoing since 20 August 2005. The approach yielded an increase in the accuracy of forecasts for a broad range of weather elements. In August 2006, the forecast period was extended to Day-10, by combining climatology and automated predictions, in order to generate the Day-8 to Day-10 component of the forecasts, which have shown some skill, albeit of a low level, during the ongoing “real-time” trial. In January 2009, the system was extended so as to provide forecasts out to 14 days in order to assess that capability. Some skill was possessed by the forecasts out to Day-12 and the skill was shown not to have been achieved by chance.

To illustrate, the correlation coefficient between the Day-11 observed and forecast amounts of precipitation (expressed as a departure from normal) was +0.204 and the probability that a correlation coefficient of at least +0.204 was achieved by chance was only 0.92%.  Similarly, the correlation coefficient between the observed and forecast Day-11 maximum temperatures (expressed as a departure from normal) was +0.256 and the probability that a positive correlation coefficient of at least +0.256 was achieved by chance was only 0.15%. Furthermore, the correlation coefficient between the Day-12 observed and forecast amounts of precipitation (expressed as a departure from normal) was +0.217 and the probability that a correlation coefficient of at least +0.217 was achieved by chance was only 0.60%. Similarly, the correlation coefficient between the observed and forecast Day-12 maximum temperatures (expressed as a departure from normal) was +0.198 and the probability that a positive correlation coefficient of at least +0.198 was achieved by chance was only 1.93%.

However, there was little skill shown by the Day-13 and Day-14 predictions.

The monthly climate outlook has been generated for too short a time to obtain a “real-time” performance measure. Nevertheless, the level of correlation between historical monthly climate anomalies, and various measures of the ENSO, the Indian Ocean Dipole, and Madden-Julian Oscillation phenomena, suggest that useful skill would be displayed in a “real-time” trial.