[Photo by Janusz Maniak on Unsplash](https://cdn-images-1.medium.com/max/9344/0*wZbwjSJe8lFKq1JP)
Photo by Janusz Maniak on Unsplash
Using high-resolution satellite images from the Amazon rainforest and a good ol’ResNet  gives us promising results of > 95% accuracy in detecting deforestation-related land scenes, with interesting results also when applied to other areas of the world. We also show a proof-of-concept of how a real deforestation detection tool could look like, including a dashboard made with Streamlit  and data infrastructure on Google Cloud . In this article, you can follow along with these insights and the journey throughout this project.
This time I’ll start with the why 😉 (a probably worn-out reference to one of the great TED talks by Simon Sinek  and a wink to all the data science enthusiasts, and my Mom, who read my previous article ).
It probably doesn’t come as a surprise to you that we’re facing climate change, a crisis which is already affecting our lives and, unless acted upon in a swift and decisive manner, could get us all into big trouble. Furthermore, this is largely caused by human activity, with the release of greenhouse gases, which both means that we’re at fault for this but also that we could potentially fix it. If you’re still skeptical about this, there are many resources that you can look into, but I’d recommend reading Tomorrow’s guide on climate change  and Bill Gates’ latest book “How to Avoid a Climate Disaster” .
One of the contributors to climate change is deforestation. When cutting trees and plants, we are also cutting natural carbon sinks, i.e. one of the few parts of nature that still seems to want to help us even after all the crap that we put onto her. Besides getting fewer CO2-capturing trees, we are also disrupting the local ecosystem, with potential complications down the line on wildlife, food sources, and water reserves, as well as potentially filling that now emptied area with more polluting factors such as cattle or fossil fuel-powered facilities. And despite all of these disadvantages, we’re facing an increase in deforestation in some parts of the world, including in Brazil, where 2020’s deforestation rate was the highest of the previous decade .
When faced with these concerns, we might be tempted to fall into despair
No need to stress out too much there Ned.
Not a moment to stand still either.
However, while we must remain aware of the severity of the topic, we should also be stubborn optimists , in the sense of believing that we can overcome this adversity and actually do something about it.
We got this!
Fortunately, there are several technological innovations that can help us on reducing our carbon footprint. While the first one that comes to our minds might be renewable energy tech, I’d say satellites can be of great use as well. While they don’t reduce emissions on their own, satellites can inform us of our progress and guide our action plan. Getting unbiased, global insights on a frequent basis might be crucial to set priorities and to move forward with policies that have a sound base. In that sense, their importance is being recognized, as we see more and better remote sensing satellites being launched into space by the likes of Planet , Satellite Vu , and Satellogic , as well as more organizations using them for sustainability purposes, such as Climate TRACE , TransitionZero , and Carbon Mapper .
Satellite images on their own are not that practical, as we can’t really just hire some person to run a kind of CCTV surveillance on the entirety of Earth. But we can automate the extraction of insights through a machine learning pipeline. We’re in luck in this part too as a lot of progress has been made in computer vision over the last years and, with all the satellite data that we’re collecting, it’s also starting to become more and more an enticing machine learning field.
As if all of this wasn’t enough motivation, this started as a group project between me and Karthik Bhaskar for the Full Stack Deep Learning — Spring 2021 online course. So, we also have the curiosity to work on a machine learning project in a somewhat complete manner, from start to end.