Machine learning-based extreme event attribution.

Area of Science:

• Climate Science
• Machine Learning
• Extreme Weather Events

Background:

• Rising global temperatures are increasing the frequency and intensity of extreme weather events.
• Accurate attribution of these events to anthropogenic climate change is crucial for understanding impacts.
• Established attribution methods can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate a novel machine learning-based approach for extreme event attribution.
• To quantify the influence of global mean temperature (GMT) on specific extreme heat events.
• To assess the potential for machine learning to provide rapid and cost-effective event attribution.

Main Methods:

• Utilized convolutional neural networks (CNNs) to create dynamically consistent counterfactual scenarios.
• Applied the CNN model to a recent extreme heat event in southcentral North America (2023) and historical events.
• Compared machine learning-derived attribution estimates with results from established methods.

Main Results:

• Temperatures during the 2023 southcentral North America heat event were estimated to be 1.18°–1.42°C warmer due to global warming.
• Similar events are projected to occur 0.14–0.60 times per year at 2.0°C above preindustrial GMT.
• Learned relationships between daily temperature and GMT were influenced by seasonality and daily meteorological conditions.

Conclusions:

• Machine learning, specifically CNNs, provides a viable and efficient tool for extreme event attribution.
• The findings align with established attribution techniques, validating the ML approach.
• This method enables faster and more accessible analysis of climate change's role in extreme weather.