From Fires out West to Hailstorms in the Central US: Unraveling the Link with Machine Learning
Wildfires in the western United States have become more frequent and stronger. As western wildfires begin earlier each year, they have begun to occur at the same time severe weather systems hit the Central United States (CUS). Previous physics modeling of single events showed a possible connection between these seemingly unrelated occurrences. Researchers applied machine learning (ML) techniques to understand the connections between the fires in the western US (WUS) and the occurrence of large hail (size: ≥ 2.54 cm) in the CUS. They found that the ML models that considered temperature and moisture over the fire regions, fire power and burned area demonstrated high prediction accuracy (> 90).
Both fires and severe storms generate substantial economic losses across the United States and the globe. The confirmed impact, based on long-term observational data, of western fires on severe storms and weather hazards in downwind states suggests severe weather forecasts may consider wildfires in the affected states. This includes Wyoming, South Dakota, Nebraska, and Kansas. Both wildfire and severe thunderstorm potentials are projected to increase in a warming future. The impact of wildfires on severe storms in downwind regions may be increasingly critical as climate warming continues.
Researchers applied two machine learning techniques- Random Forest (RF) and Extreme Gradient Boosting (XGB), to understand the relationships between the fires in the western US and the occurrence of large hail (size: ≥ 2.54 cm) in the CUS based on 20 years of the observational data. They first identified co-occurrences of CUS large hail and WUS fires from March to September over the period of 2001–2020. They considered the input variables including temperature and moisture in the fire region, the westerly wind over the plume transport path, and the fire features (i.e., the maximum fire power and burned area) to the ML models. They validated the ML models using five-fold cross-validation. The built ML models demonstrate high prediction accuracy (> 90), particularly in four CUS states: Wyoming, South Dakota, Nebraska, and Kansas. The temperature and moisture in the fire region, the westerly wind over the plume transport path, and the fire features are identified as key contributing variables. The statistical results confirm the impact of WUS fires on severe weather in the CUS revealed from researchers’ previous physics modeling of a single event. As a warming climate is projected to lead to more frequent and severe wildfires, their influence on severe weather in downwind regions may increase.