Identifying Potential Vertical Spaces for Nature-Based Solutions using Geospatial and Artificial Intelligence Analysis

Abstract

Cities globally are faced with many challenges such as the urban heat island effect, urban storm water management, and climate change. Nature-based solutions (NBS) are able to mitigate the impact of these aforementioned challenges, along with providing a variety of ecosystem services and co-benefits. Retrofitting NBS within the dense urban fabric may be challenging, which is why vertical spaces (facades and roofs) should be considered. This research aimed to create a novel 'vertical greening toolbox' capable of finding potential vertical spaces for NBS, using Amsterdam as a case study. The toolbox makes use of text-to-mask image segmentation with the Segment Anything Model (SAM) to find potential suitable facades in Street View Imagery (SVI). Potential suitable roofs are found by applying geospatial analysis techniques on the Dutch building information dataset (BAG). A greening potential score (GPS) is then calculated for both facades and roofs and combined into a mean GPS for aggregated areas to indicate their vertical greening potential. This study highlights a scalable, reproducible harmonization of facade and roof data using deep learning and geospatial analysis to find potential vertical spaces for greening. This, to the best of our knowledge, has not been done before. The Amsterdam case study indicates high facade greening potential for districts with mostly residential buildings. For roofs, industrial districts showed the highest greening potential, and an estimated roof area of 10.0 km2 (38%) was identified as highly suitable. Still, limitations apply to these findings. The toolbox was evaluated, though proper validation is needed to test its robustness. Future research should be focused on adding facade surface area estimation and increasing the flexibility of the roof greening potential estimation.