PIER: Physics-Informed Environmental Retrieval for Time-Series Modeling
PIER adds a physics-aware retrieval stream so environmental models do not rely on embedding similarity alone.
The framework scores retrieved cases by flux-response consistency with the target system, using local verifiers trained on physics-derived flux features. It then learns scenario-specific weights to balance that physics stream against standard embedding retrieval. In tests on 356 Midwestern U.S. lakes over 41 years, the authors report better water temperature and dissolved oxygen predictions than baseline methods. ArXiv · AI/CL/LG's note
The framework scores retrieved cases by flux-response consistency with the target system, using local verifiers trained on physics-derived flux features. It then learns scenario-specific weights to balance that physics stream against standard embedding retrieval. In tests on 356 Midwestern U.S. lakes over 41 years, the authors report better water temperature and dissolved oxygen predictions than baseline methods. ArXiv · AI/CL/LG's note
score 4