A shared vision for saving the Amazon
Bringing together humans and AI to detect deforestation.
Crowd-training AI to automate conservation efforts
Beginning on Earth Day in 2020, SAS and IIASA launched a project asking online volunteers to help identify areas of deforestation in more than 90,000 satellite images. Two years later, participants from 130 countries have successfully classified more than 1 million square kilometers of the Amazon.
The result is a crowd-trained suite of AI models that can now monitor new areas autonomously and detect deforestation with better than 90% accuracy. And unlike the human eye, computer vision doesn’t always rely on the visible spectrum, meaning the model could be applied to near-infrared images in the future for even greater accuracy.
Seeing the future of Amazon conservation
This novel approach to rainforest conservation makes it possible for experts to focus on harder-to-classify areas. It also opens the door to new technological approaches to detecting and preventing deforestation.
The AI models can be tied into new or existing deforestation monitoring systems, and future efforts that implement time-series analysis could help create improved alert capabilities to take conservationists from detection to prevention. Applying AI to satellite data also has potential applications beyond deforestation, in areas like:
Pollution
Migration
Food Security
Natural Hazards
Wildfires
Sustainable Agriculture
If we aren’t careful, we will watch our ecosystem deteriorate through a process of degradation, a small and marginal process, but collectively devastating over time. Albert Van Jaarsveld CEO International Institute for Applied Systems Analysis
The APPROACH
Humans Amplifying AI
AI doesn’t happen magically. It relies on the expertise of people as the basis for its training data. The crowd’s input gave the AI models examples to learn from, and the accuracy improved as more examples were provided.
The Modeling Cycle Simplified
IMAGE COLLECTION
Satellite images are continuously collected for areas affected by deforestation.
IMAGE CLASSIFICATION
Through crowdsourcing, thousands of people like you sort and label the images that show signs of deforestation.
MACHINE LEARNING
Models are trained to recognize areas of deforestation based on this labeled data, learning and improving at every iteration.
EVALUATION
Models are tested for accuracy. High-quality models are deployed, and lower quality models continue to reiterate.
DEPLOYMENT
Even deployed models continue to be evaluated for accuracy. The best model is run constantly to alert experts when new areas of deforestation are detected so immediate action can be taken.
THE TECHNOLOGY
Working hard behind the scenes
As computer vision models analyze millions of images every day to identify the earliest signs of new deforestation, government policymakers and aid organizations can intervene to protect these forests before it's too late.
AI driving conservation efforts with scale and speed
The global challenge to protect our planet would be impossible without the robust scalability provided by SAS® Viya®. Our open analytics platform with cloud enabled, in-memory processing gives you accurate results at incredibly fast speeds. Research data that took years to analyze can now be put to good use in a fraction of the time.
In the News
Keep the conversation going to keep our planet safer.
Our knowledge and passion can have a real impact in preserving our world. So get involved and ask others to do the same at #rainforest.