A team of international scientists has begun a project to refine predictions of El Niño-Southern Oscillation (ENSO) events by applying deep learning techniques to atmospheric data collected during the 1883 eruption of Krakatoa. The project, announced this week, represents a novel attempt to leverage a uniquely detailed 19th-century meteorological record against the backdrop of a catastrophic natural event.
The 1883 eruption of Krakatoa produced spectacular atmospheric phenomena observed globally, including vividly colored sunsets and atmospheric pressure waves that circled the Earth multiple times. These observations, meticulously recorded by weather stations and mariners, provide a rare, high-resolution snapshot of atmospheric behavior following a major volcanic disturbance. Researchers believe this data holds clues to understanding the complex interplay between volcanic eruptions and ENSO cycles.
The current research, detailed in a recent publication in Nature, focuses on using observation-informed deep learning to project ENSO patterns. The team is attempting to train a deep learning model not just on contemporary data, but also on the historical record from the Krakatoa event. This approach aims to improve the accuracy of ENSO forecasts, which are critical for predicting weather patterns and mitigating the impacts of climate variability worldwide.
“The challenge lies in bridging the gap between 19th-century observational techniques and modern data assimilation methods,” explained a researcher involved in the project. “We are essentially trying to teach a 21st-century algorithm to interpret a 19th-century dataset, accounting for the limitations and biases inherent in the historical record.”
The project’s initial phase involves digitizing and validating the historical data, a process complicated by the varied formats and quality of the original records. Researchers are cross-referencing observations from different sources to identify and correct errors, and to create a consistent dataset suitable for machine learning. The team is also working to account for the limited spatial and temporal coverage of the 1883 observations.
While the primary focus is on improving ENSO prediction, the research also has implications for understanding the broader impacts of volcanic eruptions on the climate system. The Krakatoa eruption injected massive amounts of aerosols into the stratosphere, causing a temporary cooling of global temperatures. The study aims to disentangle the effects of this cooling from the influence of the eruption on ENSO dynamics.
The project is scheduled to continue through 2026, with the next phase involving the development and training of the deep learning model. Researchers plan to compare the model’s performance against existing ENSO forecasting systems, and to assess its ability to predict both short-term and long-term climate variability. As of today, no official statements have been released regarding the project’s preliminary findings or anticipated improvements in forecasting accuracy.
