Climate change is a pressing and dangerous global issue that touches all aspects of life on this planet. (IPCC, 2018). At the same time, cities are taking up more and more land around the world, reducing the land available for natural, carbon-sequestering environments. (Seto et al., 2012).
Cities around the world have considered or have enacted regulations to promote green roofs as way to help reduce urban carbon emissions impacts (City of Vancouver, 2018a).
What it does:
urbanForests.ml aims to help city managers, civil servants, and the general public identify potential urban spaces to support green roofs and estimate their ecological and economic impacts.
How we built it
We trained a machine-learning object recognition model powered by Google Vision Auto-ML to recognize different types of rooftops (ex. low-density housing, stores/warehouses/industrial sites, and skyscrapers). This was done with over 600 datapoints across 60+ images.
Challenges we ran into:
Machine Learning Model:
- Limitations with computer vision (hard to perceive depth, i.e. skyscrapers vs. low-density housing).
- Labour intensive process to label training data at scale
- Learning to integrate various frameworks (Google Cloud Services, Microsoft Azure, and custom code).
- JSP to HTML integration.
Accomplishments that we're proud of:
- Labelling over 600 bounding boxes of custom training data
- Deploying a Google Machine Learning API on a Microsoft Azure hosting platform
- Over 50% precision as measured by Google model metrics
What we learned:
- Hnat: Data training is hard, labelling data is a very labour-intensive process
- Benny: There are many unforeseen difficulties in deploying multi-platform systems (ex. mixing JSP and HTML)
- Edward: It's hard to maintain a consistent level of productivity and commitment over the entire hackathon.
What's next for urbanForests:
- More training data to provide better accuracy in low-medium density environments and across larger aerial scales
- Refine user interface
- Identify suitability for intensive vs. extensive green roofs
- Define a custom model
- Acquire HTTPS certificate for webpage
- City of Toronto, 2005: https://web.toronto.ca/wp-content/uploads/2017/08/8f39-Report-on-the-Environmental-Benefits-and-Costs-of-Green-Roof-Technology-for-the-City-of-Toronto-Full-Report.pdf /n
- City of Toronto, 2018: http://map.toronto.ca/maps/map.jsp?app=TorontoMaps_v2
- City of Vancouver, 2018a: https://council.vancouver.ca/20180725/documents/pspc15.pdf
- City of Vancouver, 2018b: https://maps.vancouver.ca/
- IPCC, 2018: https://www.ipcc.ch/sr15/
- MacDonald et al., 2016: https://dalspace.library.dal.ca/bitstream/handle/10222/76724/Green%20Roof%20Carbon%20Sequestration%20Potential%20of%20Dalhousies%20Halifax%20Campus.pdf?sequence=1&isAllowed=y
- Seto, Güneralp, and Hutyra (2012): https://www.pnas.org/content/109/40/16083.short
- Air Conditioning Units: pixabay.net, https://www.needpix.com/photo/378272/services-ac-repair-business-heating-services-cooling-services-air-purification-air-conditioning-repair-air-conditioning
- City Skyline: pixino, https://pixnio.com/architecture/city-downtown/city-cityscape-downtown-metropolis-blue-sky-architecture-aerial-urban
- City of Vancouver from Granville Island: jantiga, https://pixabay.com/photos/city-downtown-vancouver-631445/
- City of Vancouver from a Mountain: Colin Knowles, https://www.flickr.com/photos/colink/4959132756/
- Green Roof: pxhere, https://pxhere.com/en/photo/965031
- Industrial Pollution: Petr Kratochvil https://www.publicdomainpictures.net/en/view-image.php?image=16891&picture=industrial-pollution
- Rooftop Gardens: pixfuel, https://www.pxfuel.com/en/free-photo-qgqzm
- "Everything is Fine" meme is a work of parody/commentary.
- Map data used under Open Government License - Vancouver and Open Government License - Toronto (as applies).
- Unless otherwise noted, images used are from CC0 (Public Domain) sources.
- Website Template was provided from https://templated.co/ under a CC3 (Attribution 3.0 Unported) License