Our weather blows in from the vast North Pacific where there is little information available about day-to-day atmospheric conditions. Even satellite images can't completely bridge the gulf.
Add to the equation B.C.'s towering mountain peaks and convoluted coastline, and it is not surprising that weather forecasting here is more difficult than most other places in North America.
But now campus researchers led by Roland Stull, head of the Atmospheric Sciences Programme in the Dept. of Geography, are applying a new method of weather forecasting that could overcome these hurdles. Known as the UBC Ensemble Forecast System, it is showing promising results.
Even under the best conditions, current forecasting techniques can make accurate predictions just three or four days in advance, and only for large weather systems, Stull said.
This is because the weather is a very complex, non-linear and chaotic system in which a small change in one place can create large outcomes somewhere else--like the proverbial flutter of an African butterfly's wing that causes a tornado in Texas.
"What we are trying to do is cheat chaos by taking what we know about chaos theory and using it to improve forecasting methods," Stull said. "It just might be the best system for B.C. I think we can make a difference here."
Stull and graduate students Josh Hacker and Henryk Modzelewski use computers to create ensemble forecasts, ones that result in a number of outcomes instead of a single prediction.
This is where chaos theory enters the picture. Instead of sticking strictly to current weather conditions as a starting point for predictions--as conventional forecasting does--they make a series of slight alterations to the statistics. These numbers serve as a basis for complex calculations that result in a spread of likely outcomes.
For example, a conventional forecast might state categorically that overnight temperatures will remain above freezing at 2 C.
An ensemble forecast for the same night, however, may predict a cluster of low temperatures that could include the likelihood of frost--say 3 to -4 C -- information that could be crucial to an industry such as agriculture.
By averaging out the ensemble predictions he arrives at a more accurate forecast than any one categorical forecast. He also uses two different numerical models to make calculations, doubling the number of ensemble members.
To combat the forecasting difficulties posed by B.C.'s topography, Stull and his colleagues use a fine-mesh forecast grid over population centres and areas of commercial activities.
The fine-mesh forecasts are made at grid locations spaced just five kilometres apart, compared with the 35-kilometre grid used by the Atmospheric Environment Service.
This allows them to better capture the often substantial differences in weather across mountains, valleys and coastlines.
Improved weather forecasts could have significant impact in B.C. Some areas that could benefit include: flood warnings, wind forecasts for log-boom towing and forest-fire fighting, precipitation forecasts for watersheds where B.C. Hydro has dams, air pollution forecasts for the Fraser Valley, snowfall forecasts for highway maintenance, and avalanche predictions for railroads and ski resorts.
"If we can improve predictability by even one day, for example, by warning of a serious storm, then we could help save property and lives," Stull said.
The UBC Ensemble Weather Forecast System is not yet perfected. For one thing, it can consume hours of computer time--up to 15 hours per day for some detailed forecasts.
It is also difficult to get exact measures of current weather for use as a starting point due to instrument error and gaps between weather stations. Choosing which slight initial differences should start their calculations is also a scientific challenge.
Stull and his colleagues share their forecasts with the Pacific Weather Centre, but with the understanding that this is just early research and not ready for public dissemination.
However, centre forecasters can gain insight on storm dynamics from these forecasts, and can provide feedback on the model capabilities, Stull said.