Image credit: Tomorrow.io
By Susan Biagi
Climate change is fueling weather-related disasters, which can be sudden and catastrophic. A serious event can knock out power, demolish homes, and take lives with little warning. Understanding weather patterns could help scientists better predict when and where a weather event will unfold, giving people and utility companies more time to prepare.
To gain that knowledge, scientists are turning to artificial intelligence. AI is already used to help predict weather, but accuracy diminishes when predicting long term. While AI excels at analyzing images and data to train models, predicting weather has proven more difficult. So many variables are, well, variable. Air temperature, pressure, wind, precipitation, and other factors change constantly. Finding patterns within them is difficult, especially when looking two or more weeks into the future.
The ability to predict an impending disaster can mitigate damage and destruction, which is exacerbated by power outages. Researchers at the State University of New York in Albany (UAlbany) are working on AI models that correlate conditions that could indicate a weather crisis long before it occurs. The research team is training the AI model to cross-reference historical power outage data with local weather data, in an effort to predict the impact a specific weather event will have on a power company’s infrastructure.
For example, if the AI algorithms determine that conditions are right for a severe ice storm to blow into a region, it can alert meteorologists and utility company engineers, who can prepare the power lines for the impact. In the short-term, the goal is to lessen the likelihood of a severe power outage. Long-term, the AI system could help utility companies determine where to invest in grid-hardening to avoid or minimize power disruptions.
Power to the People
UAlbany also created the Wind Extremes Forecast System (WEFS), in collaboration with the Consolidated Edison Company of New York (Con Ed). WEFS uses machine learning and numerical weather prediction (NWP) modeling to predict wind speeds or gusts that are likely to cause power outages in specific counties or towns. Predicting the location of an outage allows energy companies to mobilize resources and restore or redistribute power more quickly.
The WEFS looks at winds of 30 to 50 mph—two danger thresholds—and considers precipitation, vegetation, and past outages. That information is correlated with real-time weather data collected every five minutes from a network of more than 125 weather stations that use Doppler LiDAR and radiometers to sample local weather conditions. When the AI model detects conditions that are ripe for an outage, it notifies the team.
Data to Help Weather the Storm
While much work remains to hone long-term predictions, this kind of weather insight is already helping companies plan around weather in the coming 10 to 14 days. Tomorrow.io created a weather platform that is taking businesses by storm. The Tomorrow.io Platform combines quantitative precipitation forecasting with real-time local weather data and geospatial information to create a weather prediction tool that can be customized for specific verticals and businesses. It allows companies to see how near-term weather events can impact their operations. Customers can view and track their weather concerns on a dashboard, which posts and sends alerts when relevant concerns emerge.
Tomorrow.io also developed a weather API that can be integrated into an application, system, or program to provide detailed weather data to a business. Transportation companies, sports organizations, and energy companies use the insights from the weather API to run their operations. From canceling a sporting event to warning delivery fleets of gusty winds along their routes to gearing up for potential power outages, knowing what weather lies ahead can improve preparedness and operational efficiency.