Gwangju Institute of Science and Technology r

Image: Integrating AI with weather forecasting models can make wildfire forecasting systems more accurate, according to a new study by an international team of researchers. Their new model improves forecast accuracy with lead times of up to seven days, enabling earlier preparation and better allocation of resources for wildfire risk mitigation.
look more

Credit: Jin Ho-Yoon of Korean GIST.

Raging wildfires around the world are causing enormous economic damage and loss of life. Knowing in advance when and where widespread fires are likely to occur can improve fire prevention and resource allocation. However, available forecasting systems provide limited information. Furthermore, there is not enough lead time to get useful local details.

Scientists are now applying deep learning algorithms to better predict wildfire hazards in the western United States. Researchers from South Korea and the United States have developed a hybrid method that combines AI technology and weather forecasting to improve extreme fire risk predictions at a finer scale (4km x 4km resolution) up to a week, Increased usefulness of containment and management.

“We tried a number of approaches to integrate machine learning with traditional weather forecasting models to improve wildfire risk predictions. It’s a big step forward because it shows the potential of such efforts to enhance fire hazard prediction.” Lead author Lackhun Son, Ph.D., a recent Ph.D., said: He received his PhD from the Gwangju Institute of Science and Technology (GIST) in South Korea and is currently working at the Max He Planck Institute for Biogeochemistry in Germany. “Fire hazard prediction could be further improved with constant developments in both earth system models and recent AI developments.” he adds.

Data-driven AI methods have shown great capabilities for inferring things, but explaining why and how inferences are reached remains a challenge. This led to AI being labeled a black box. “But when we combine AI with computer models based on physics, we can now diagnose what is going on inside this black box.” Co-author Professor Simon Wang of Utah State University, USA, said: “AI-based predictions related to extreme levels of fire hazard are well based on strong winds and specific geographic features such as high mountains and canyons. US West, traditionally difficult to solve with coarser models”

Computational efficiency is another big advantage of this method. The traditional method of predicting fire risk at finer spatial resolution, a process called ‘regional downscaling’, is often computationally intensive, expensive and time consuming. “It required comparable computational resources during the development stage, but once the AI ​​training task is complete, i.e., it is run once for the first time, it becomes possible to use that component in a weather forecast model to generate the rest of the forecast. It only took a few seconds to complete the season,” Co-author Professor Kyosung Lim of Kyungpook University in South Korea said: Therefore, the newly developed AI-based method, which can make accurate high-resolution forecasts in a shorter time, was much more cost-effective compared to traditional forecasting systems.

“In this study, the AI ​​was only tested for fire hazard prediction in the western United States. In the future, it may be applied to other types of extreme weather and other parts of the world.” Co-author Philip J. Rasch, Ph.D., Pacific Northwest National Laboratory and University of Washington, said: “The flexibility of our AI method helps predict weather-related features.”

This research Journal of Advances in Modeling Earth Systems September 22, 2022.

***

reference

DOIs: https://doi.org/10.1029/2022MS002995

Author: Son Rakhun^1,8, Po Lung Ma²Hyron Wang²Philip J. Rush^2,3Shiyu (Simon)^Five one^FourKim Hyonjun^5,9,10Jung Ji Hoon⁶Kyosan Sunny Lim⁷Yoon Jin Ho^8,

Affiliation:

¹ Max Planck Institute for Biogeochemistry Division of Biogeochemistry

² Pacific Northwest National Laboratory

³ Department of Atmospheric Sciences, University of Washington

^Four Department of Plants, Soils, and Climate, University of Utah, Logan

^Five Moon Soul Korea Advanced Institute of Science and Technology Graduate School of Future Strategy

⁶ College of Earth and Environmental Sciences, Chonnam National University

⁷ Department of Earth System Science, Kyungpook National University

⁸ Graduate School of Global Environmental Engineering, Gwangju University of Science and Technology

⁹ Department of Civil and Environmental Engineering, Korea Advanced Institute of Science and Technology

^Ten Institute of Industrial Science, University of Tokyo

About Gwangju Institute of Science and Technology (GIST)

Gwangju Institute of Science and Technology (GIST) is a research-oriented university in Gwangju, South Korea. Established in 1993, his GIST has become one of the most prestigious schools in South Korea. The university aims to create a strong research environment to spur scientific and technological progress and foster collaboration between international and national research programs. With the motto “Proud Creator of Future Science and Technology”, GIST consistently wins his one of the best university rankings in South Korea.

Website: http://www.gist.ac.kr/

About the author

Jin-Ho Yoon is Professor of Geoscience and Environmental Engineering at GIST, Korea. His group focuses on understanding and predicting weather and climate extremes under climate change. Yoon’s group also analyzes aerosol-cloud-precipitation interactions to understand cloud distribution and properties. Prior to coming to GIST, Pacific He was a Scientist (Level 3) at the Northwest National Laboratory. In 2004, Professor Yoon completed his Ph.D. He holds a PhD in atmospheric science from Iowa State University, USA.

---

---

journal

Journal of Advances in Modeling Earth Systems

Survey method

Computational simulation/modeling

Research theme

not applicable

article title

Deep learning significantly improves county-level fire forecasting in the western United States.

Article publication date

September 22, 2022

Disclaimer: AAAS and EurekAlert! EurekAlert! is not responsible for the accuracy of news releases posted. Use of information by contributors or via the EurekAlert system.