Deep Learning Maps Green Roof Potential in Swiss Cities

Effective urban climate adaptation requires detailed spatial information about rooftops and buildings, particularly for assessing the ecosystem services provided by green infrastructure like green roofs. While green roofs are recognized for mitigating urban heat islands, existing research often only maps current green roofs or potential sites, not both comprehensively. This study introduces an advanced deep learning framework to address this gap. The framework, a modified deep convolutional neural network based on Roofpedia, leverages publicly available Swisstopo datasets, including high-resolution aerial imagery, digital surface models for slope information, and building footprints. It classifies rooftops into four distinct categories: existing green roofs, rooftops suitable for new green roof installations, rooftops with solar panels, and flat rooftops unsuitable for greening. Applied to Bern, Switzerland, the model successfully identifies realistic opportunities for expanding green infrastructure. Being fully open-source, this framework is highly transferable, offering urban planners worldwide evidence-based insights for deploying green infrastructure and enhancing urban thermal comfort.