AI Climate Project

For the last year our Institute has been developing AI Climate, a remote sensing, machine-learning tool which seeks to map, monitor and assess exposure of informal settlement and low-income communities to climate change shocks and stresses in vulnerable areas of the Global South. At this stage we are focusing on river flooding.

From geospatial imagery and other datasets, the AI Climate tool generates an up-to-date flood-risk heat map available in a dedicated website.  The AI-Climate tool will merge knowledge of the local community and public officials through a crowdsourcing co-design component and will inform participatory scenario planning in order to develop strategic and equitable interventions with socially vulnerable urban and peri-urban communities.

This initiative is a response to two major trends: on one hand, the combination of impacts of climate change, fast expansion of informal settlements, and budgetary and trained-personnel limitations especially in secondary and smaller cities; and on the other, rapidly increasing geospatial imagery availability and falling processing costs, which make it possible to cut time, expense and improve quality of documentation to craft policies, spatial plans and interventions aiming at safer, resilient cities.

Starting with adaptation strategies for informal settlements and flooding, our vision is that, with further evolution, AI Climate will be used to help plan strategies for both climate change adaptation and mitigation in cities across the world, through participatory processes that include free, equitable access to the tool.

In 2020, we received seed money to start a Proof of Concept of the AI Climate tool, a process now about 80% complete.  The Proof of Concept is testing state-of-the-art technology that processes geospatial imagery and georeferenced datasets, deriving analytics-ready layers from these datasets. Specifically, we use convolutional neural networks (U-net architecture) for classification and prediction for vulnerable communities and hybrid models for flooding susceptibility.

The machine learning component of the AI Climate tool is being carried out with data from Tegucigalpa and in San Pedro Sula, Honduras.  Although still in the initial stages, we provided just-in-time scenario information and “flooding susceptibility” mapping to our partners in Honduras for the Iota Hurricane last November.

Our Institute has a long tradition of innovative ideas.  In this arena, we started using geospatial data, remote sensing techniques combined with Machine Learning while in Arusha, Tanzania, on a project for the Agha Khan Foundation in 2013.  By training an algorithm, we identified villages in large swaths of land using military aerial images.

Today, we are an interdisciplinary group aimed at bridging and integrating planning, science and technology points of view and solutions. Our collaborators include Dymaxion Labs, Machine Learning and Remote Sensing specialists in Argentina; the Dean of Sciences of the National University of Honduras and the Director of the University’s Institute of Earth Sciences; and our Institute, I2UD, for vision, management and funding.  We have an advisory group comprised of Alfredo Stein, a professor at the University of Manchester, Adriana Larangeira, an architect and urban planner based in Rio de Janeiro, and Víctor Pajuelo Madrigal, a senior remote sensing and machine learning manager at Satellogic in Barcelona. Within I2UD, the project is being led by Alejandra Mortarini and Carlos Rufín.