Megadose The AI curve, compressed.

DySENSE: Dynamic Spatiotemporal ENergy Synthesis and Emulation with Satellite Imagery

· HF Daily Papers ·
DySENSE generates monthly building-energy maps while keeping the underlying city layout fixed.

The framework pairs synthetic satellite imagery with spatially aligned energy-consumption maps, using land-use, month and climate prompts to control its outputs. Tests covered New York, Washington, Norfolk and Busan. The authors report generating high-fidelity annotated synthetic data with less than 25% labelled data, improving downstream prediction by 8% mIoU. They also report prediction-error reductions of 9–18% on the CVRMSE metric compared with existing methods.

HF Daily Papers' note

score 4

Categories: Research