Monitoring Deforestation in the Amazon with Multimodal Learning

The Multimodal Learning for Earth and Environment Challenge (MultiEarth 2022) aims to monitor and analyze deforestation in the Amazon rainforest using satellite remote sensing, specifically synthetic aperture radar (SAR) with optical sensors. The challenge addresses the difficulty of gathering clear images in humid and cloudy weather conditions. The challenge also involves manually labeling deforestation label maps, which is time-consuming and requires collaboration with external organizations. The MultiEarth 2022 challenge received 54 submissions across three sub-challenges and was successful in promoting collaboration between MIT Lincoln Lab, MIT campus, and the Department of Air Force.

• Deforestation in the Amazon rainforest is a significant problem that contributes to reduced biodiversity, habitat loss, and climate change.
• Satellite remote sensing, specifically SAR with optical sensors, offers a powerful tool to track changes in the Amazon, even in difficult weather conditions.
• The MultiEarth 2022 challenge involves manually labeling deforestation label maps, which is time-consuming and requires collaboration with external organizations.
• The MultiEarth 2022 challenge received 54 submissions across three sub-challenges and was successful in promoting collaboration between MIT Lincoln Lab, MIT campus, and the Department of Air Force.

The challenge organizers initially attempted to manually label deforestation label maps by overlaying high-resolution Planet imagery and drawing binary polygons following the boundaries of the deforested regions. However, they realized that this was too time-consuming and contacted Scale AI for help. A single time slice contained roughly 10,000 polygons, and Scale AI was able to label the remaining 10 time slices. With the help of Scale AI and other collaborators, the MultiEarth 2022 challenge was successful in providing binary deforestation label maps from 2016 to 2021.

Despite international efforts to reduce deforestation, the world loses an area of forest that is equivalent to the size of 40 football fields every minute. Deforestation in the Amazon rainforest accounts for the largest share, contributing to reduced biodiversity, habitat loss for many of the world’s most threatened animals and insects, and more rapid climate change.Satellite remote sensing offers a powerful tool to track changes in the Amazon. The Multimodal Learning for Earth and Environment Challenge (MultiEarth 2022) is the first competition aimed at monitoring and analyzing deforestation in the Amazon rainforest at any time and in any weather or lighting condition. Weather conditions in the Amazon are often humid and cloudy, making it difficult to gather clear images.In this workshop, learn from Miriam Cha, Research Scientist in the Artificial Intelligence Technology Group at MIT Lincoln Laboratory, about multimodal representation learning for the earth and environment. Cha will discuss the benefits and challenges of synthetic aperture radar (SAR), a sensor that transmits microwave signals and then receives back the signals that are returned from the earth’s surface. Although SAR images are extremely clear, they can be difficult for humans to interpret due to noise and lack of spatial correlations. This is why MultiEarth 2022 is using SAR with optical sensors for maximum effect.