Embedded Files
My current research uses explainable machine-learning techniques to study the influence of climate change on extreme weather events and their impacts.
Extreme weather events (e.g., heatwaves, floods, droughts) can have destructive and costly impacts on human health, electrical infrastructure, food security, water supply, and ecosystem health. To prepare for future impacts, it’s critical that we understand (1) how climate change will influence extreme weather events and their impacts in the coming decades, and (2) what physical processes are be responsible for driving these changes. In the following projects, we leverage explainable artificial intelligence techniques to analyze large amounts of climate data to investigate how global warming influences extreme weather events today—and what impacts we can expect in the future.
Recent and ongoing research projects
Climate change and extreme weather
Climate change and extreme weather
Decades of research has shown that climate change is making extreme weather events more frequent and more intense. However, since each extreme weather event results from complex interactions between the atmosphere, the land surface, and the background climate, quantifying the precise impact of global warming on any individual extreme weather event remains challenging.
To isolate the impact of global warming on individual extreme weather events, we develop a new approach that uses a machine learning model to predict temperature based on the atmospheric pressure patterns and the global mean temperature. After training this machine learning model, we use it to ask "what if?" questions to understand how the temperature of an extreme heatwave would have been different if the same weather conditions had occurred in a climate with a different level of global warming.
We use this approach to study several record-breaking extreme heat events including the 2023 Texas heatwave and the 2003 Europe heatwave (shown on the right). We find that global warming intensified the Europe 2003 heatwave by up to 1.8℃ (3.2℉), and this heatwave would occur on average once per year in a world with +2.5C of global warming relative to 1850.
Impact of global warming on the August 2003 heatwave in Western Europe
(figure adapted from Trok et al. 2024)
Future heatwaves pose substantial health risks
Future heatwaves pose substantial health risks
Evidence suggests that climate-related deaths are increasing in many regions of the world, particularly for heat-related deaths. Periods of extreme heat can cause mortality rates to surge which can strain local healthcare services, potentially amplifying the health risks. To prepare for future impacts, we need to quantify peak mortality rates that can be expected during plausible future heatwaves.
In work led by Prof. Chris Callahan, we combine statistical and machine learning-based approaches to estimate the mortality rates that would occur if recent extreme heatwaves were to recur under a range of future global warming levels. To do this, we first use a causal inference approach to establish an empirical relationship between daily mean temperature and weekly mortality rates across different subregions of Europe. Then, we use the machine learning-based approach developed in Trok et al. 2024 to predict changes in heatwave temperature, and combine these approaches to derive counterfactual mortality estimates for several historical heatwaves at different levels of global mean temperature. At future levels of warming, we find that peak weekly heat mortality rates are comparable to peak COVID-19 mortality rates across Europe.
Exploring land-atmosphere interactions with machine learning
Exploring land-atmosphere interactions with machine learning
For the same reason that sweating keeps us cool, the amount of moisture on the land-surface can exert a controlling influence on the near-surface air temperature. When soils are dry, the sun warms the land-surface which can be an important driver of extreme heat events (sometimes increasing temperatures by several degrees Fahrenheit). While these processes are well-understood, there remains uncertainty about the extent to which soil moisture regulates near-surface temperature in different regions of the world.
To isolate these land-atmosphere interactions, we train a machine learning model to predict regional temperature using daily maps of atmospheric pressure and soil moisture. Then, we use this model to derive the nonlinear relationships between soil moisture and temperature over different regions of the world. We find large regional differences in the sensitivity of temperature to soil moisture, with southern Australia exhibiting the strongest sensitivity. These results have important implications for regional climate impacts, particularly as climate change alters the global distribution of soil moisture and precipitation.
Page updated
Report abuse