Sea surface temperature forecasting improved by combining text context and spatial graphs
Textual Environmental Context and Spatial Graphs for LLM-Based Regional SST Forecasting
Artificial Intelligence
Summary
Forecasting ocean surface temperatures is tricky because conditions change both over time and across locations. The authors developed a method that gives a large language model information about past temperatures, environmental conditions, and ocean knowledge in text form, along with graphs representing spatial relationships and recent changes. This combination helps the model predict future temperature changes in a specific sea area better than other approaches. They also added a rule-based system that explains forecasts by linking trends to known environmental factors.
What this means in practice
- •For environmental monitoring teams: Improve regional sea surface temperature forecasts by integrating text-based environmental data with spatial graph representations.
- •For marine navigation planners: Use enhanced SST forecasts with spatial context to better plan shipping routes considering temperature-related ocean conditions.
Authors
Xiong Li, Xiaowei Zhou, Yanwei Yu, Qian Cui, Junyu Dong
Abstract
Sea surface temperature (SST) forecasting depends on local temporal persistence, regional spatial dependence, and environmental conditions that evolve with the forecast date. We study how these heterogeneous conditions can be presented to a large language model (LLM) for regional multi-step forecasting without serializing the full SST grid as text. We formulate forecasting as conditional numerical generation: historical SST and anomaly sequences, date-aligned environmental records, and static ocean knowledge form a textual context, while regional spatial state is supplied through continuous graph-derived prefixes. A static graph encodes persistent geographic--climatological relations, and a dynamic graph encodes recent SST correlations and localized tropical-cyclone influence. Two graph neural networks produce a target-node representation that is mapped by a spatial-prefix fusion and injected into the LLM input. On SST forecasting in the South China Sea, the complete configuration achieves the best MAE and $\Rtwo$ among the compared methods over ten forecast steps. Alongside the numerical forecast, a rule-based module matches predicted trends and environmental-factor directions with knowledge entries to return source-linked, post-hoc contextual explanations.