AI for Accurate Rooftop PV Potential

In this challenge, participants will develop a computer vision model that analyzes aerial imagery provided by swisstopo to improve these estimates. Teams can focus on detecting existing PV panels, identifying rooftop obstacles through image segmentation, or combining both tasks into a unified solution. Participants are encouraged to leverage modern deep learning techniques, including foundation segmentation models (SAM) and multimodal approaches.

Switzerland has set ambitious climate goals, and expanding rooftop solar installations is a key part of achieving them. While existing datasets (Sonnendach) estimate the photovoltaic potential of rooftops, they do not take into account already installed solar panels and rooftop structures like chimneys, skylights, and dormers that reduce usable surface area. The resulting solutions should provide more realistic estimates of installable rooftop solar capacity and support the development of the energy transition strategy.