PiCo: Physics-Informed Concept Probing for Geo-Foundation Models
We propose Physics-Informed Concept Probing (PiCo), a post-hoc framework for explaining GFM predictions.
ProofPaper ↗
Key points
- Geo-foundation models (GFMs) are typically evaluated by downstream task predictive performance, revealing little about which physical features they are sensitive to or use in prediction.
- PiCo identifies concept vectors in a GFM's embedding space by generating synthetic multispectral satellite observations from a 3D radiative transfer model coupled with a game engine.
- We evaluate PiCo on three GFMs with different input modalities using seven physical concepts covering forest leaf optics and canopy structure.
- For hierarchical forest type and tree species classification, concept representations retain 70.0-93.8% of full-embedding accuracy.
Sources (1)
- [1]PiCo: Physics-Informed Concept Probing for Geo-Foundation ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 09:46 PM
We propose Physics-Informed Concept Probing (PiCo), a post-hoc framework for explaining GFM predictions.
Geo-foundation models (GFMs) are typically evaluated by downstream task predictive performance, revealing little about which physical features they are sensitive to or use in prediction.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 14, 2026Sparse Transformer: a Controlled Testbed for Mechanistic Interpretability
- Oct 14, 2026When Is Smaller Greener? Right-Sizing Language Models for Agentic AI
- Oct 14, 2026More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning
- Oct 14, 2026Orbia: Do Generated Videos Form a Consistent 3D World?
- Oct 13, 2026Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon Agents
- Oct 13, 2026OneDemo: Unified Policy Learning and Data Generation from a Single Demonstration