Skip to content Skip to footer

MIT scientists have developed a method to “correct” the predictions made by climate change models, thus enabling more accurate risk analysis of extreme weather events. Specifically, they have combined machine learning with dynamical systems theory to fine-tune global climate model predictions for the long-term. This enables policymakers and planners to assess community-specific risks of extreme weather such as flooding. The refined forecasts will be particularly beneficial when paired with smaller-scale models to predict specific weather events such as heat waves, cyclones, or floods.

The team’s “correction scheme” uses algorithms to nudge a climate model’s output towards more realistic output. While the finer components, such as storms or clouds, are not directly resolved in the models, the large-scale dynamics that generate these events are corrected. Through machine learning, the algorithm uses past weather data to make the necessary adjustments. The machine-learning scheme adjusts phenomena such as temperature, humidity, and wind speed.

The primary focus is on correcting how an extreme weather feature will appear in the coarse model compared to real-world conditions. The method learns dynamics, which then leads to correct outbreak statistics (e.g., frequency of rare extreme events).

The team tested the machine-learning scheme on an existing climate model run by the U.S. Department of Energy, applying eight years of past data to train their new algorithm. The corrected model then produced climate patterns that more closely matched observing conditions over the past 36 years. Notably, the team managed to accurately reproduce the frequency of extreme storms in specific global locations.

The correction approach can be applied to any global climate model, making predictions more accurate and helping determine where and how frequently extreme weather will occur in the face of rising global temperatures. This will allow for better preparation and mitigation strategies against extreme weather events due to climate change.

The next step is to analyze future climate scenarios, modifying the correction scheme to incorporate future greenhouse-gas emission scenarios. The work was partially supported by the U.S. Defense Advanced Research Projects Agency and their results were published in the Journal of Advances in Modeling Earth Systems. The study was conducted by researchers at MIT and the Pacific Northwest National Laboratory in Washington state.

Leave a comment

0.0/5