Towards the effective communication of weather and climate information—harnessing new technologies to integrate material from various sources on the web to generate new products
It proposes effective integration of material within the weather and climate enterprise, highlights the application of rapidly-changing technologies to bring new and powerful tools to disseminate and receive information, and notes efforts being undertaken to better communicate with different populations.
With this background, results of a "real-time" trial of a system used to generate forecasts, by mechanically integrating (that is, combining) judgmental (human) and automated predictions are documented.
The approach utilised by the system is to integrate material from various existing sources on the web to automatically generate improved forecasts for Melbourne and other central Victorian localities. These have yielded an increase in the accuracy of predictions for a broad range of weather elements.
The purpose of the paper is to discuss the application of the aforementioned system to automatically generate new and enhanced weather and climate products in English, Australian Indigenous, and other languages.
For example, after a trial of nearly five years, the mean square error (MSE) of the generated temperature predictions for Melbourne has averaged 0.72°C less than the MSE of corresponding official forecasts.
Similarly, the accuracy of the Melbourne Day-1 to Day-7 rainfall forecasts so generated has also been increased by means of the mechanical integration process, mechanically integrated forecasts of whether or not it was going to rain being correct 6.2% more often than the corresponding human (official) product.
Furthermore, the accuracy of the Melbourne Day-1 to Day-7 thunderstorm forecasts so generated has also been increased by means of the mechanical integration process, the Critical Success Index (CSI) of the mechanically integrated forecasts of thunderstorms being 1.1% higher than that of the corresponding human (official) product.
However, the accuracy of the Melbourne Day-1 to Day-7 fog forecasts so generated has not been increased, the CSI of the mechanically integrated forecasts of fog being 0.1% lower than that of the corresponding human (official) product.
Looking further ahead, solid statistical relationships exist between historical monthly climate anomalies in Victoria and various measures of the ENSO, Indian Ocean Dipole, and Madden-Julian Oscillation phenomena.
Using this approach over the past year, the correlation coefficients between forecast probabilities that the Melbourne total monthly rainfall, minimum temperature and maximum temperature would be in tercile three, and the corresponding observed total monthly rainfall, minimum temperature and maximum temperature 1961-1990 deciles, were respectively +0.28, +0.25, and +0.32.
The probability that the three correlation coefficients would all be at least +0.25 by chance is 0.8%, suggesting that it is most unlikely that the skill displayed by the experimental monthly climate outlooks arose by chance.
Examples of automatically generated text:
"South Channel Island Forecast for Tue-10-8-2010 Cloud increasing. A few showers developing during the afternoon. Following a chilly night, a cool day. Mainly light to moderate wind, from an easterly direction at about 20 to 30 km/hr during the morning and continuing to be from an easterly direction, but at about 15 to 25 km/hr, during the afternoon. Gusts to around 40 km/hr. A smooth to slight sea with waves averaging up to 0.5 m."
"... The Madden-Julian Oscillation (MJO) is presently operating in Phase 4. This is reflected in the near-equatorial enhanced convection being found over the over the western part of what is referred to as the Maritime Continent (Indonesia). Following Phase 4, the region of enhanced convection often moves from the western part of Indonesia to the eastern part of Indonesia. In Melbourne, at this time of the year, a combination of the MEI, the DMI, and the MJO Phase, such as what we have operating now, suggests, over the following 30 days: RAINFALL: There is a 51% chance of it being wet, a 31% chance of normal rainfall, and a 18% chance of it being dry ..."
Supplementary URL: http://www.bom.gov.au