Terra: Multimodal Spatio-Temporal Earth Science Benchmark from NeurIPS 2024

Terra is a multimodal spatio-temporal benchmark for Earth science applications developed by CityMind-Lab and presented at NeurIPS 2024. It provides global-scale data across multiple modalities to support the development of advanced environmental and geographic models. The dataset was released in late 2024 to address the need for standardized benchmarks in the Earth science domain.

Use Cases

• Benchmarking spatio-temporal models using multimodal Earth science features
• Global-scale environmental monitoring using time-series geospatial data
• Developing cross-modal representation learning for Earth observation tasks

Strengths

• Peer-reviewed benchmark from NeurIPS 2024
• Global geographic scope
• Multimodal data integration

Limitations

• Lack of publicly documented column schemas in the repository metadata
• High complexity inherent in aligning multimodal spatio-temporal features

Provenance

Source

CityMind-Lab, NeurIPS 2024

Freshness

Last updated in November 2025; contains data relevant to the 2024 NeurIPS publication.

Geography

Global