Predicting the effects of global warming with AI systems

The enormous technological advances that the world has experienced in recent years now give us a clearer idea of where we are when we talk about climate change and what the possible future scenarios are. At the same time, it provides us with great tools to deal with it.


Recent studies indicate that in the next 20 years, the global temperature will increase by at least 1.5°C, but may reach 4°C. The environmental consequences are enormous and monitoring them becomes very difficult and costly with traditional tools. At Dymaxion Labs, we work to contribute to the generation of relevant information for an effective approach to the issue.


Forest fire monitoring with machine learning


Global warming generates very favorable conditions for the emergence and prolongation of fires: more recurrent heat waves, longer hot and shorter cold seasons. Unfortunately, it is increasingly common to see forest fires with devastating environmental, social, and economic consequences.


In order to perform early detection and effective monitoring, we developed the open-source packages; satproc and unetseg, which detect burned areas through specialized algorithms.


This is what a fire looks like on a satellite image. Source: Dymaxion Labs


Satproc focuses on the processing of geospatial images and vector files to solve object detection and semantic segmentation problems. Unetseg, on the other hand, is a deep learning model training and prediction package, based on the U-Net architecture, for semantic segmentation problems on geospatial images.


Detection of open dumps


The growing presence of open dumps particularly affects what is known as the global south. According to a World Bank study, in Latin America and the Caribbean in particular, open dumps account for approximately 27% of waste disposal and treatment.


The environmental impact is negative, but it is also negative from a social and economic point of view, affecting mostly vulnerable populations in the surrounding areas. In joint work with the Bunge & Born Foundation, we developed a project for the detection and monitoring of landfills in a scalable and low-cost way.


Thanks to the use of Machine Learning techniques for satellite image processing, we were able to identify large areas quickly and with few resources. We start from examples of open dumps identified by the human eye, which generally come from the work of municipalities and NGOs.


The algorithms we create look for common patterns and then generate predictions of possible as-yet-undetected landfills over vast tracts of territory. In addition, they allow us to generate valuable insights for accurate monitoring and periodic reporting of landfill locations.


Only the actions and policies that are implemented will be able to avoid the worst scenarios; to do so, it will be key to have accurate and relevant information. Our solutions seek to contribute to building a more sustainable and fair future.


Author:Federico Bayle, Co-founder, and CEO of Dymaxion Labs.


​​All views expressed on this blog post are the author's and do not necessarily reflect or represent the opinions of the AI for Climate Initiative or its founders.