Clever computing for better wind forecasts

Posted 27 July 2017

Weather forecasting is a number-crunching affair: under the watchful eyes of meteorologists, data from weather stations around Ireland get digested by massively powerful computers and are mixed with global models to work out the future probabilities of cloud, rain, wind and occasionally even sunny weather in locations around the country. But forecasting something as variable as wind is tricky - particularly if you want to predict windiness at a specific location, like a wind farm.

Once again computers can help, and UCD researcher Dr Conor Sweeney is looking at how fine-tuning the weather forecast could help local wind-farms manage their resources. A native of Dublin, Dr Sweeney started out his career as an engineer, working in Italy and the US designing machinery for material handling. He enjoyed the experience of living overseas, but the work began to lose its appeal.

"The shine wore off - it didn't involve an awful lot of thought beyond a formulaic approach to design, so I wanted to do something that I found interesting," he recalls. A PhD in Trinity re-ignited his curiosity as he worked on a noisy problem in aeronautics. "We were interested in the noise given off by an aircraft when it lands, and how to design aircraft so they could be more silent - we were trying to come up with a new way of solving how the air flows around awkward shapes," says Dr Sweeney. "We wrote a computer programme that solved the equations in a different way, it was a lot more efficient and worked well when you had awkward shapes."

That computing experience came in handy for his next project - working with Met Éireann and the Dublin Institute of Advanced Studies on ways to predict the effects of climate change on Ireland. "We wanted to run computer models to downscale what was going to happen," he explains. "I knew a lot about computational models - and it turns out the governing equations are the same."

From there it was a logical move into further work on forecasting, and after a brief round trip to Africa on a motorbike, in 2007 Dr Sweeney came on board at (opens in a new window)UCD School of Mathematical Sciences. There he has been looking at ways to improve wind forecasting, because being able to predict with high certainty the levels of wind at specific wind-farms would help balance inputs into the national electricity grid, and so ultimately make wind energy more cost-efficient.

"The problem with wind is that you can't turn it on and off," says Dr Sweeney. "EirGrid has to strike a balance between wind usage and using gas turbines - and they don't need this uncertainty."

Looking at a regional weather forecast - which can get down to a resolution of 2.5km in Ireland - will not necessarily tell you much about local effects at individual wind farms, explains Dr Sweeney. "There could be loads going on within that 2.5km that would have a local impact - there could be a valley or a hill that the model hasn't accounted for - so at the wind farm, the wind could behave differently to the forecast model."

In a project funded by (opens in a new window)Science Foundation Ireland, Dr Sweeney is part of a team at UCD that has been figuring out how to scale down the wind forecast for specific sites using relatively little computing power - so a solution could be run on a desktop computer. The trick is to compare forecast data to the data on what then happened at the location, to see the 'hit rate' of the forecast, he explains.

"You have the forecast and later you get the observations, so you know what happened," he says. "If the forecast says it is always going to be 10 metres per second and it's always 8 metres per second, then you subtract 2 - we are using ideas as simple as that, how the forecast compares with the observations."

Taking other factors into account, such as wind direction and using more advanced statistical methods, can increase the skill of the forecast even further, he adds. The project is currently applying the approach at seven locations around Ireland to localise the wind forecast. "That has worked very well," says Dr Sweeney, who says they hope to engage wind farms at a later point. And the solutions they develop could be applied to wind farms anywhere in the world: "It's a big problem; a lot of countries just take the outputs of computer models - but it's easy to apply these simple, quick processes. The extra number crunching you are doing is very small but the added forecast skill potentially is of big value."

Dr Conor Sweeney was interviewed by freelance journalist Dr Claire O'Connell

-----------------------------------------------------------------------------------

UCD Climate & Meteorology Centre (UCD School of Mathematical Sciences)

UCD is now producing regular weather forecasts. Hourly predictions of rainfall, wind, temperature and pressure for the next two days are available on the website. The forecasts are updated four times each day, and the charts are available at(opens in a new window)http://mathsci.ucd.ie/met/mcc-forecast.html