Two observers watch the Pacific Ocean. One studies surface heat maps, the other listens to the rhythm of temperature and heat stored in the water column. On their own, each notices only part of the picture, but together they can sense the approach of El Niño months before the warm current alters weather across the planet.
El Niño, the warm phase of a climate oscillation in the equatorial Pacific, brings droughts to Australia and floods to Peru, affecting fisheries and crops. An accurate forecast six months ahead gives time to prepare. Classical climate models poorly resolve small-scale processes, while regional models are limited by one-way nesting. Machine learning opens a path to more flexible analysis of large oceanographic datasets, but existing models often suffer from insufficient validation on real data and do not always capture extreme events.
Researchers from Vietnam National University built the ConvLSTM-XT model, which merges the two observers into a single architecture. A convolutional neural network (CNN) processes sea surface temperature (SST) heat maps from NOAA ERSST v5, while long short-term memory (LSTM) analyzes time series of temperature and ocean heat content (OHC) from the Chinese Academy of Sciences Oceanographic Data Center. The data span January 2000 to September 2023, confined to the Niño 3.4 region (5°S–5°N, 120°–170°W). Instead of standard matrix multiplications, the LSTM blocks use two-dimensional convolutions, allowing the model to capture both spatial and temporal patterns simultaneously.
Forecast accuracy reached 90.57% for configurations where at least one of the five quarters used real observations and the rest were forecast. When all five quarters were forecast without observational input, accuracy dropped to 83.02%. Qualitative analysis of heat maps from March to September 2023 showed that the model captured the transition from cold to warm anomalies typical of El Niño development, though it did not always exceed the 0.5°C threshold in all five periods.
The results confirm that hybrid spatiotemporal neural networks can compete with traditional models in El Niño forecasting. High accuracy with partial use of observations indicates practical value for operational monitoring. The drop in accuracy for fully autonomous forecasts points to the need for additional input variables or architectural improvements. By joining forces, the two observers already provide a several-month head start in the face of a climate anomaly.
The 1982–1983 El Niño nearly wiped out the Peruvian anchovy population, which supported the world's largest fishery. Catches plummeted from 12 million tons to less than 1 million tons.
In Kim Stanley Robinson's novel 'Forty Signs of Rain', climate models play a key role in predicting extreme weather events, and scientists struggle to make their forecasts heard by policymakers. Developing more accurate El Niño models echoes this theme: the earlier and more precise the warning, the better the chances for adaptation.