BAMOS Jun 2018
21
Figure 5. The Bureau of Meteorology’ s current Cray supercomputer.( Source: Tim Pugh, Bureau of Meteorology)
Along with the dramatic advances in numerical weather prediction, there has been a corresponding improvement in public weather forecasts. For example, on average, current 5-day forecasts for Melbourne’ s maximum and minimum temperature have approximately the same skill as 1-day forecasts from 50 years ago( e. g. Figure 4( shown on pg. 22); Stern and Davidson, 2015). Large maximum temperature errors of 5oC or more have decreased from approximately 10 % of the total number of forecasts forty years ago, to less than 1 % in recent years. Forecasts of rainfall are still difficult and, although they too have improved, more development is required.
Nowcasting has also benefited from advances in observational technology. Australia now has automatic weather stations reporting as frequently as every minute, 10-minute geostationary satellite imagery, a network of weather radars measuring rainfall and in some cases wind profilers that monitor the upper-level winds and temperature, and wind measurements from aircraft.
5. Technological Requirements
The computer model used by the BoM to forecast the weather has approximately 1 million lines of code. A typical 10-day forecast requires the computer to solve the governing laws of physics every few minutes at more than 50 million points on the globe( at approximately 25 km spacing). Moreover, within a given 6-hour time window, 3-4 million observations of varying quality are assimilated.
The peak speed of the supercomputer currently used by the BoM( Figure 5) is 1.66 × 1015 floating point operations per second 3. The computer model used by the BoM typically uses 2160 computer nodes with 128 GB of memory per node. In rough terms, this configuration is one million times faster and has half a million times more memory than a fast home PC( in 2016).
6. Future Directions and Challenges
• International Collaboration— without the sharing of meteorological data and scientific knowledge, Australia’ s capacity to forecast the weather would be much poorer.
• Improving parameterisations especially for localised heavy rain.
• Ensemble prediction— constructing and using ensemble forecasts including how to effectively communicate the meaning of probabilistic forecasts to the public.
• High impact weather— high-resolution, short-term forecasts, particularly for severe weather and its impacts.
• Urban weather— improved air quality and localised flooding predictions.
• Convergence of weather, seasonal and climate models— the atmospheric models used for weather prediction, seasonal prediction, and climate projections are converging towards a single unified modelling approach known as seamless prediction.
• Computational demands.
• Tailored information and effective communication— meteorological information needs to be tailored to make it useful to the user community. In the future there will be greater emphasis on adding value to the raw computer model forecasts. Social media and mobile platforms are useful and targeted means of conveying forecasts and warnings to the public and specific user groups.
• Advances in remote sensing— remote sensing applications( both space and surface based) continue to develop at great pace. There is a challenge to capture and process the larger data volumes and translate the extra data into weather analysis and prediction, particularly into nowcasting and warning services that protect and benefit the community.
• Citizen observations— many people have mobile communication devices that can take and send measurements of pressure, and allow people to report unusual or extreme weather.
• Maintenance of the observational network.
• Verification— comprehensively verifying all forecasts made by both the public and private sectors is important.
3
A floating point operation is a mathematical operation such as addition or multiplication on a decimal number.