Researchers from the Massachusetts Institute of Technology (MIT) have developed a new method that can make long-term predictions regarding the risk of extreme weather events more accurate. The new technique combines machine learning with dynamical systems theory to make better predictions about extreme weather events such as floods and tropical cyclones in specific areas.
Currently, policymakers heavily rely on global climate models when assessing a community’s potential exposure to extreme weather conditions. However, these models only provide a coarse resolution and can only make predictions decades or even centuries into the future. The existing models can be used to predict future weather conditions for a region like the northeastern U.S., but not for specific cities like Boston.
To make the predictions more specific, policymakers usually combine a coarse model’s large-scale forecasts with a finer-resolution model. The fine-resolution model is used to estimate how often Boston is likely to experience damaging floods as the climate warms. Nevertheless, this risk analysis is only as accurate as the predictions from the coarser climate model.
The MIT researchers have now developed a method that can correct the predictions from such coarse climate models using machine learning combined with dynamical systems theory. The new technique nudges a climate model’s simulations into more realistic patterns across large scales. When paired with smaller-scale models to predict specific weather events such as floods or tropical cyclones, the approach made more accurate predictions regarding how frequently particular locations will experience those events over the coming decades.
While large-scale climate models currently simulate weather features such as the average temperature, humidity, and precipitation globally, they usually average out features every 100 kilometers or so due to the enormous computing power required. Running these model simulations can take massive computing power, and in order to simulate how weather features will interact and evolve over periods of decades or longer.
These models, however, do not account for crucial processes such as storms and clouds, which happen over smaller scales of a kilometer or less. To enhance these coarse climate models’ resolution, scientists usually try to fix a model’s underlying dynamical equations, which describe how phenomena in the atmosphere and oceans should interact physically.
The new method developed by the MIT researchers takes a different approach. Instead of attempting to correct the equations, the technique aims to correct the model’s output. The approach overlays an algorithm on a model’s output or simulation and nudges the simulation towards something that more closely represents actual world conditions.
The algorithm is based on a machine-learning scheme that learns associations within historical weather data worldwide reflecting fundamental dynamics among weather features. The algorithm then uses these learned associations to correct a model’s predictions, improving our understanding of how extreme weather events will look in a future climate. Predicting extreme weather events is crucial for planning and engineering designed to mitigate the potential impact of these events.
The technique promises to significantly improve the capacity to predict and prepare for extreme weather events linked to climate change. The research team’s results appear in the Journal of Advances in Modeling Earth Systems. The work was partially supported by the U.S. Defense Advanced Research Projects Agency.