Large Models for Time Series and Spatio-Temporal Data: A Survey and Outlook

A comprehensive survey reviews the emergence of large foundation models adapted for analyzing time series and spatio-temporal data, categorizing approaches into two groups: models for time series analysis (LM4TS) and spatio-temporal data mining (LM4STD). The research consolidates recent advances in applying large language models and foundation models to temporal data across diverse domains, establishing a foundation for understanding how AI systems can process dynamic, sensor-generated information at scale.

This survey addresses a critical convergence in artificial intelligence: the application of large foundation models to temporal data analysis. As organizations generate massive volumes of time-series and spatio-temporal data through sensors and monitoring systems, extracting actionable insights from this information has become increasingly important. The research documents how recent breakthroughs in large language models are being adapted to handle temporal patterns, representing a shift from domain-specific solutions toward more generalized approaches.

The systematic categorization into LM4TS and LM4STD reflects the maturation of temporal AI research. Foundation models, originally developed for natural language processing, are proving effective at pattern recognition across time dimensions—suggesting that architectural innovations in one domain transfer meaningfully to others. This has implications for industries ranging from finance to healthcare, where time-series prediction and anomaly detection drive operational decisions.

For investors and developers, this survey establishes that large models for temporal data remain an active frontier with both established approaches and emerging opportunities. The consolidation of datasets, implementations, and tools signals a maturing ecosystem. However, the distinction between general-purpose and domain-specific models indicates that practitioners still require specialized knowledge—off-the-shelf solutions may underperform compared to tailored implementations.

Looking ahead, the development of unified temporal foundation models could reshape predictive analytics across industries. Key questions remain about computational efficiency, data requirements for effective transfer learning, and whether temporal reasoning will become a standard capability in general-purpose AI systems. The field appears positioned for continued rapid advancement as researchers refine model architectures specifically for sequential and spatial-temporal phenomena.

• →Large foundation models are increasingly adapted for time series and spatio-temporal data analysis, extending LLM capabilities beyond natural language.
• →The field categorizes into two main approaches: LM4TS for pure time series and LM4STD for data with spatial-temporal dimensions.
• →General-purpose temporal models exist alongside domain-specific solutions, suggesting the ecosystem remains in transition toward standardization.
• →Datasets, implementations, and tools are consolidating by application area, enabling faster adoption but indicating specialized knowledge remains valuable.
• →Temporal data analysis with large models spans diverse industries from finance to healthcare, making advances here broadly applicable.