NCAR plays key role in NAS study on improving seasonal forecasts

Staff contribute expertise in cyberinfrastructure, decision support

NCAR's Anke Kamrath recently applied her expertise in "big data" to help a national panel devise a strategy to strengthen seasonal weather prediction.

The director of operations and services for NCAR's Computational and Information Systems Laboratory (CISL), Kamrath served on a committee sponsored by the National Academy of Sciences on sub-seasonal to seasonal weather predictions (S2S) – defined as predictions two weeks to 12 months in advance. She led the cyberinfrastructure section and provided input into building workforce capacity.

"I enjoyed working with committee members who came from around the country, as well as Canada and the United Kingdom," Kamrath says, adding that she learned from their expertise in different scientific areas. For example, she was struck by the challenge of obtaining observational data for sea ice depth, "an important factor in the forecasting."

Anke Kamrath
Anke Kamrath of CISL. (©UCAR. This image is freely available for media & nonprofit use.)

More-accurate S2S predictions are seen as critical for helping industry, agriculture, and government protect lives and property from an increasing number of extreme weather events. But S2S forecasts currently often fall into a gap between weather forecasting and climate modeling.

The recently published report concludes that S2S models must seamlessly combine weather forecasting with more variables from climate or Earth system modeling, which integrate atmosphere, ocean, land, and other components.

"In seasonal forecasting, climate and weather forecasting are converging," Kamrath says. "In fact, S2S forecasting is more similar to the demands of climate modeling than weather modeling."

S2S challenges

Improving seasonal forecasting is a big data challenge, says Kamrath, since it requires assimilating more than a billion data points into multiple "ensemble" runs of an S2S model and crunching the data with supercomputers. Meanwhile, improvements in computer and storage capacities haven't kept up with the needs of S2S modelers, and increased investments are needed.

NCAR's Yellowstone supercomputer
Supercomputers such as the Yellowstone system at the NCAR-Wyoming Supercomputing Center are required to crunch the massive amounts of data included in a seasonal weather prediction model.  (©UCAR. Photo by Carlye Calvin. This image is freely available for media & nonprofit use.)

She sees great potential value in improving S2S forecasts. The U.S. Navy was a key sponsor of the report, due to its interest in planning on these timescales. The information also is expected to benefit emergency and city planning, energy and water management, agriculture, insurance risk management, and other sectors.

Studies have examined the use of seasonal forecast information for a number of years. However, NCAR Scientist Rebecca Morss, who was part of a panel during the committee's visit to NCAR, notes that in many cases it's not yet clear how people can best turn the predictions and early warnings into action, given the uncertainties involved.

For example, current seasonal forecasts might predict that the odds are greater for a wetter-than-usual season, but it's unlikely we'll be seeing forecasts a season in advance that predict rain on a particular Sunday, given the innate variability of weather.

"One of the questions is at what point do you actually make specific decisions to position resources for a potential catastrophic event?" Kamrath says. "That's not clear yet." For example, "At what probability of a severe drought does one reposition resources to address the impact?"

Developing future scientists and mathematicians

Kamrath also contributed to the workforce section of the report, which concluded that universities aren't producing enough scientists and mathematicians who can not only use models and data assimilation techniques but also write code and further develop these capabilities.

Our organization plays an important role in executing the S2S strategy, she notes, from our modeling and risk communication work to our programs focused on inspiring and developing the next generation of research professionals, who will need to understand a rapidly evolving field that spans many disciplines. CISL internships, for example, give undergraduate and graduate students hands-on experience exploring research and development challenges in high-performance computing for the geosciences.

Scientists at NCAR already are working on monthly and seasonal forecasts through projects such as the North American Multi-Model Ensemble (NMME). Postdoctoral Fellow Karen McKinnon (NCAR Advanced Study Program) also recently published research on predicting heat waves in the eastern U.S. 50 days in advance.

The S2S committee met five times in person over 18 months, including one meeting at NCAR during which several NCAR scientists made presentations. Morss and Julie Demuth (MMM) discussed social science and decision support, while Rich Loft (CISL) talked to the committee about optimizing models for supercomputers.

As with other National Academies reports, congressional staff and federal agencies will be briefed on the findings.

The report, Next Generation Earth System Prediction: Strategies for Subseasonal to Seasonal Forecasts is available for purchase or as a free PDF.


Jeff Smith, Science Writer and Public Information Officer