Extreme weather can wreak havoc on the power grid, all the more so when it strikes with little warning. In January 2017, temperatures in parts of France dropped to as low as 5 degrees, driving up demand for electric heating at the same time winds stilled — leaving wind power in a lurch. With several nuclear power plants offline for maintenance, the grid strained under the added demand.
Similarly, during the February 2021 Texas cold snap, subfreezing temperatures hobbled poorly weatherized gas- and coal-fired power plants, and ice gathered on wind turbines, leading to widespread outages, with the state “seconds and minutes” away from a months-long grid failure, according to the Texas Tribune.
Historically, forecasts have only offered a few days’ notice of severe weather, giving grid operators little time to prepare. But that’s beginning to change.
A recent paper on the latest developments in long-range forecasting suggests it is now possible to predict such severe cold snaps up to two weeks in advance, information that would be invaluable for deciding whether to keep French nuclear reactors online to shore up gaps in supply or for protecting the Texas power grid against collapse.
“The scientific methods have only just started to be there in the last few years where these forecasts could actually be useful,” said Hannah Bloomfield, a postdoctoral researcher at the University of Bristol and co-author of the paper. “That’s going to be quite helpful from a planning perspective with people thinking about energy systems.”
Recent advances have made it possible to predict some heat waves, cold snaps and tropical cyclones weeks ahead of time, the paper found, a potential boon to the grid. With enough forewarning, the power sector can better prepare for extreme weather, limit rationing, plan for surpluses and shortfalls in renewable power, and keep the price of electricity in check.
Grid operators and power producers are already taking advantage of advances in short-term forecasts — roughly three to five days into the future. Wind and solar farm operators, for instance, rely on such forecasts to know when to sell electricity to the grid. If operators commit to supplying power, and the wind calms or clouds form overhead, causing output to stall, they risk paying hefty fees.
Groups of model simulations — known as ensemble forecasting systems — have made this calculus a little easier. These systems produce large numbers of forecasts from the same model with slight tweaks to the input data. When the forecasts agree, it lends confidence to their predictions.
“Having multiple forecasts that all agree so that your forecast is rock solid, as opposed to having lots of variability in your forecasts, gives people who are bidding on this competitive market some sense of certainty,” said Julie Lundquist, an atmospheric scientist at the University of Colorado Boulder.
Great Britain’s grid operator has applied artificial intelligence to ensemble forecasting to improve predictions of solar output by one third. A simulation of the New England power grid found that improved short-term forecasts led to the grid using more solar power and less fossil fuels. A simulation of California’s power grid found that improved short-term wind forecasts produced as much as $146 million annually in savings.
Now more refined models run on increasingly powerful computers are delivering more accurate forecasts for the coming weeks and months.
With today’s improved models, scientists could predict hot spells like the 2010 Russian heat wave — when temperatures surpassed 100 degrees in parts of the country — up to three weeks in advance. Looking to the season ahead, models can’t say if it will be wet or dry, hot or cold on a given day three months from now, but they can better gauge the odds of an especially cold winter or a particularly rainy summer.
To make such predictions, scientists are focusing on slower-moving processes, such as changes in ocean temperatures, which vary little from day-to-day, but can shift significantly over the course of a season, helping to shape the weather.
“The models are including more of these processes that give long-range predictability, which requires more computer horsepower,” said David Brayshaw, a climate scientist at the University of Reading. “There’s also that fact that we're getting better at modeling the interactions between these other processes and the atmosphere.”
In a 2020 paper, scientists at the Center for International Climate and Environmental Research in Oslo and the Barcelona Supercomputing Center lamented that many renewable energy companies are still not making use of longer-range forecasts.
“Better forecasts could help renewable energy producers plan their production and curtailment with greater accuracy and over longer time frames, leading to improved performance,” they wrote. “This in turn would make renewable energy business more competitive, attracting more investments and facilitating a smoother transition towards a low carbon economy.”
Initiatives like the European Union-funded subseasonal-to-seasonal forecasting for energy (S2S4E) are working to turn better forecasting data on wind and sunshine into usable information for the power sector. One study to come out of the program showed that it’s possible to forecast above- or below-average solar power across much of Europe around two weeks out. For wind power, that extended to three weeks.
“This whole area of subseasonal and seasonal forecasting is very rapidly developing,” said Brayshaw, a co-author of the study, which was led by Bloomfield. “There’s a lot of work in Europe at the moment around this issue of climate services, about taking the data that comes out of these models and doing something with it.”
Predictions that look months into the future have more limited returns. “For multiple months, it is hard to do better for wind and solar forecasts than basically use climatic averages,” Benjamin Hobbs, professor of environmental management at Johns Hopkins University, said in an email. But scientists are trying.
Recent research has looked to slow-moving, large-scale weather phenomena to predict higher- or lower-than-usual renewable energy output for the coming months. A 2020 study found that someone seeking to gauge wind power in parts of France three months ahead of time would do better to consult a seasonal forecast than to look at what the weather is typically like that time of year. The difference is small but enough to “be useful for improving forecasting and risk management,” authors wrote.
“The main point is indeed that there is some potentially useful information in the subseasonal to seasonal forecasts,” said Riwal Plougonven, an atmospheric scientist at the Dynamic Meteorology Laboratory in France and co-author of the study, in an email. But, he added, “extracting the useful information for the energy sector, in a form that is useful and straightforward to use for stakeholders, still requires some work.”
Over the much longer term, on the scale of years or decades, experts say predicting wind and sunshine is important for deciding where to build wind farms, solar arrays and transmission lines to provide for a steady stream of power. Accurate predictions are as crucial for renewables as geology is for oil and gas, said Justin Sharp, an atmospheric scientist and owner of a consultancy supplying knowledge of weather and climate to the electric sector.
“I’m fond of saying that you cannot forecast away variability, and even perfect forecasts just tell you that there may be a problem,” he said in an email. “They don’t fix the problem. That’s done by applying meteorology upstream in the planning phase.”