Decoupling Long- and Short-Term Patterns in Spatiotemporal Inference.

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Environmental monitoring relies on sensors for smart city applications, but high costs lead to sparse data collection.
• Inferring environmental data at unmonitored locations (spatiotemporal inference) is crucial for fine-grained measurements.

Purpose of the Study:

• To develop a method for accurate spatiotemporal inference of environmental data, addressing limitations of sparse sensor networks.
• To investigate and model distinct short-term and long-term temporal patterns in environmental data.

Main Methods:

• Decoupled modeling of short-term and long-term temporal patterns.
• Utilized a joint spatiotemporal graph attention network for short-term pattern analysis.
• Employed a graph recurrent network with a time skip strategy for long-term dependencies.

Main Results:

• The proposed method effectively captures complex spatiotemporal relationships.
• Demonstrated superior performance compared to existing methods across four real-world datasets.
• Successfully inferred environmental data at non-sensor locations with high accuracy.

Conclusions:

• The decoupled approach for analyzing short- and long-term patterns significantly enhances environmental data inference.
• This method offers a scalable solution for achieving fine-grained environmental monitoring in smart cities.
• Achieved state-of-the-art results, highlighting the effectiveness of the proposed spatiotemporal inference technique.