Democratizing weather forecasting: UM, NOAA collaborate on multi-model prediction systems

Democratizing weather forecasting: UM, NOAA collaborate on multi-model prediction systems

Ben Kirtman, Professor of Atmospheric Sciences, Director of the NOAA Cooperative Institute for Marine and Atmospheric Studies and Director, Center for Computational Science Climate and Environmental Hazards Program
By Kate Stein for UM Rosenstiel School

Ben Kirtman, Professor of Atmospheric Sciences, Director of the NOAA Cooperative Institute for Marine and Atmospheric Studies and Director, Center for Computational Science Climate and Environmental Hazards Program

Democratizing weather forecasting: UM, NOAA collaborate on multi-model prediction systems

By Kate Stein for UM Rosenstiel School
Ben Kirtman’s appreciation for weather forecasting began as a sleep-deprived teenager in one of the only basements in Santa Barbara, California.

That basement would flood during El Niño rains. So Kirtman’s father made his son spend nights “pump-sitting” on the basement couch. Every half-hour, Kirtman had to turn on the pump and wait for the water to go down. The experience, plus reading about an ice storm in Canada that killed many people, led him toward meteorology.

Improving forecasting “seemed like a great problem to solve,” says Kirtman, now a professor and researcher at UM’s Rosenstiel School of Marine and Atmospheric Science. “If I could help save lives, all the better.”

Kirtman leads a partnership between UM and the federal National Oceanic and Atmospheric Administration to develop long-term weather forecasts based on seven different climate models. Known as the North American Multi-Model Ensemble, or NMME, the forecast is freely available to companies and government agencies across the United States and in other countries, who use it to make decisions about everything from water resource management to energy use planning. NOAA uses the NMME forecasts to inform their official seasonal outlooks, including the status of El Niño. 

NMME forecasts must be produced every month without fail, come hurricane, polar vortex or government shutdown, as they are a critical input for NOAA’s climate and weather forecasts.

“NOAA is, by law, required to issue these forecasts,” Kirtman says. “Their congressionally mandated mission is to provide weather forecasts and warnings for the protection of life and property.”

A seasonal forecast can mean the difference between cities having extra fuel for heat or not when there’s an extremely cold winter. There are economic benefits, as well: Forecasts let farmers plan ahead for periods of drought, and help Internet and power providers prepare for possible outages from storms and hurricanes.

Kirtman says a good forecast can even make or break a romance.

“I get phone calls, you know, ‘I’m getting married May 15th, 2020. What’s the weather going to be?’”

But, he adds, rather than forecasting a particular weather event at a particular time, NMME forecasts predict if a season is likely to be warmer, cooler, wetter, or drier than normal. Currently no long-term forecast can tell you if your wedding weekend will be sunny, but it can provide a background as to what the spring season overall will look like. 

As far as Kirtman knows, there’s no other arrangement in the United States where a federal agency relies on a non-government partner for an integral service like forecasting. But over the seven-and-a-half years of NMME’s existence, UM and NOAA have never failed to release a forecast on time. Kirtman says playing a critical role to a critical federal operation is demanding but exciting. It’s raised scientific challenges that aren’t directly within his area of expertise: For instance, in addition to the forecasting system, Kirtman and his team have learned to develop safeguards for computer failures and workarounds for hurricane evacuations.

“During the shutdown [of the federal government in January], we delivered. If the shutdown lasted three more days, we would have had our first failure,” Kirtman says. “During Irma, I had to evacuate and we were still able to get our forecasts in on time.”

NMME essentially brings together climate models from researchers at NASA and the Canadian Meteorological Center, as well as NOAA and UM, to create one long-range weather forecast. 

This high-level forecasting is possible thanks to UM’s Center for Computational Science, where a supercomputer analyzes big data sets from partners around the world to create high-resolution analyses of water surface temperature, wind shear and other climatological factors.

The overall goal of NMME is reliability. Every model has unique limitations that can affect the accuracy of a forecast. By bringing together multiple models that have different limitations and draw data from different places, a forecast from NMME is typically more accurate than just one model alone could be.

Emily Becker, a climate scientist who’s contracted by NOAA to manage real-time operations of NMME, calls NMME “an incredible contribution” to what’s known as sub-seasonal to seasonal forecasting (basically, forecasts for anywhere from three weeks to nine months out). Weather centers in countries including Peru and Myanmar have begun using global NMME forecasts to support their government forecasters. Scientists are also using NMME forecasts in other areas of research such as predictions for fisheries management, drought development, and the number of hurricanes per season. 

Becker and Jin Huang, Chief of the Earth Systems Science and Modeling Division in the NOAA Climate Program Office, say they’re proud of how that work is made possible by collaboration among scientists at different institutions, gathering data from all over the world. Huang says the effort saves NOAA millions of dollars because the agency doesn’t have to spend money on developing models and maintaining the models, just supporting the people who run them.

“It’s different from the traditional idea of an operational system,” she says.

The idea of developing weather forecasts based on climate models has been around since the 1980s, when meteorologist Jagadish Shukla -- who later became Kirtman’s mentor -- released research suggesting long range weather was less chaotic and more predictable than scientists had long believed. But the first steps toward developing climate model-based weather forecasts didn’t come until decades later. The NMME project began in spring 2011 with a handful of workshops for climate modelers who wanted to collaborate on long-range weather forecasting.

It’s also led to the creation of another sub-seasonal-to-seasonal forecasting tool, SubX, which is in beta testing. SubX focuses on weather anomalies three to four weeks in the future -- shorter-term than NMME. Among various successes, SubX predicted abnormally heavy rainfall from Hurricanes Harvey and Michael, and a recent polar vortex.

Multi-model systems are “our best tool for improving the reliability of forecasts,” Kirtman says. He, Huang and Becker all say as new and more accurate climate models are developed, they want to bring them in to increase the overall reliability of the multi-model forecasts. Kirtman also wants to learn more about who uses the NMME forecasts, which are free and available online, so researchers can better suit the forecasts to real-world needs. 

He envisions a program where companies who use NMME can pay for students to study how the systems work for the companies’ particular region or product. The companies would get forecasts specific to their needs; the students would get degrees and real-world experience with the forecasting system; and the NMME collaboration would benefit from additional refinements to and research on their work.

Becker says although the logistics of keeping NMME running are complicated, participating modelers have thus far responded quickly when there’s some sort of problem. It’s a track record she, Kirtman and Huang believe they can maintain even as NMME evolves and expands.

“This is such a community effort,” Becker says. “It’s really something to be proud of.”