TerraNova: A Foundation Model for the Anthropocene
TerraNova is built to learn Earth-system fields and national indicators in one shared representation.
The paper says the model trains on 1,024 records: 512 gridded physical fields and 512 country-level societal indicators. Its architecture uses separate encoders for location, country, time, and task, then fuses them with cross-modal transformers. The authors claim the frozen backbone can reconstruct dense fields from sparse observations and adapt to unseen variables in minutes on consumer hardware. ArXiv · AI/CL/LG's note
The paper says the model trains on 1,024 records: 512 gridded physical fields and 512 country-level societal indicators. Its architecture uses separate encoders for location, country, time, and task, then fuses them with cross-modal transformers. The authors claim the frozen backbone can reconstruct dense fields from sparse observations and adapt to unseen variables in minutes on consumer hardware. ArXiv · AI/CL/LG's note
score 5