240 likes | 365 Views
A knowledge-based system to generate internet weather forecasts. Dr Harvey Stern, Bureau of Meteorology, Australia. … a work in progress . Introduction. A “pilot” knowledge-based system for the generation of internet forecasts is described. The system has been developed for Victoria.
E N D
A knowledge-based system to generate internet weather forecasts. Dr Harvey Stern, Bureau of Meteorology, Australia
Introduction • A “pilot” knowledge-based system for the generation of internet forecasts is described. • The system has been developed for Victoria. • Forecasts generated include those for public, aviation, marine and media interests.
Description of the System. • At the core of the system is an algorithm, written in HTML (and incorporating some JavaScript). • The algorithm combines statistical interpretation of NWP output with other knowledge. • The statistical interpretation component includes identification of the synoptic type.
Synoptic Typing • The basis for the system is identification of the expected synoptic type. • The direction, strength and curvature of the isobaric (surface) flow determine the type. • These characteristics are determined from a grid of forecast pressure values.
Generating the Output • Data associated with the identified type are statistically analysed. • In a Perfect Prog mode, statistical relationships so derived are used to generate forecasts. • HTML Code (incorporating JavaScript) is generated. • The Code is uploaded to a Web Site.
The “Bank” of Experience • Ramage proposed an “iterative” approach to locking in improvements in forecasting. • This is, indeed, the approach adopted here. • Thereby, the skill increases as new knowledge is incorporated. • Hence, progress is made towards the realisation of Ramage’s dream.
Multi-lingual Feature • An increasing component of the WEB is in languages other than English. • Chinese may become the common language of the WEB. • The system has a component that generates a forecast summary in Chinese.
Verification of System • Five skill measures are used. • These are MIN, MAX, QPF, Precip/No Precip (P/NP), & BRIER. • Skill measures are positive for forecasts better than climatology.
A Preliminary Trial • Conducted (last April) on an earlier (and much abbreviated) version of the system. • Evaluation limited to one month, one location (Melbourne), and to day one.
A Subsequent Trial • Conducted (last November). • Evaluation extended to include days one to seven. • Still only for one location.
Skill of Subsequent Trial’sQuantitative Precipitation Forecasts
System Performance • Skill Measures all show system forecasts better than climatology. • They also are better than persistence. • They are (on most measures) inferior to official forecasts, especially for days 1 and 2; - with the notable exception of minimum temperature forecasts for days 3 to 7, inclusive. • System forecasts would be expected to improve as new knowledge is incorporated.
Future Plans • Extensively verify the system, covering all current observing sites. • Enhance the sophistication of the statistical analysis. • Incorporate new forecaster knowledge. • Extend the multi-lingual feature to Australian indigenous languages.