Monitoring the Planet with AI and Satellite Imagery: Satlas | Written by Favyen Bastani | August 2023

The Allen Institute for AI is thrilled to introduce Satlas, a cutting-edge platform that utilizes AI-generated geospatial data from satellite imagery. Currently offering three data products that are updated monthly, Satlas provides timely and accurate information for various applications, such as renewable energy infrastructure monitoring and disaster relief planning. With its high-accuracy deep learning models, Satlas ensures reliable and up-to-date global snapshots of geospatial data. Furthermore, Satlas aims to expand its data products in the future and provide tools for other teams to create similar geospatial data solutions.

Full Article: Monitoring the Planet with AI and Satellite Imagery: Satlas | Written by Favyen Bastani | August 2023

Introducing Satlas: AI-Powered Geospatial Data Exploration

The Allen Institute for AI is thrilled to announce the launch of Satlas, a groundbreaking platform that allows users to explore global geospatial data generated by AI from satellite imagery. Satlas currently offers three data products that are updated on a monthly basis. This data is released under an open license and can be visualized on the Satlas Map or downloaded for offline analysis. Additionally, the Institute plans to release more geospatial data products in the future.

The Importance of Timely Geospatial Data

Timely geospatial data plays a crucial role in informing decisions related to emissions reduction, disaster relief, urban planning, and more. As renewable energy infrastructure continues to expand worldwide, accurately tracking this growth across political boundaries becomes vital for resource allocation. However, high-quality global geospatial data products are often difficult to find. Manual curation of these products is a tedious process that involves aggregating, cleaning, and correcting regional datasets from multiple countries, assuming these datasets even exist. Moreover, existing geospatial data on renewable energy infrastructure is fragmented and only up-to-date in limited areas.

The Power of Satellite Imagery and Deep Learning

An alternative approach to obtaining geospatial data is analyzing satellite images, such as those captured by Landsat and Sentinel-2, which cover most of the Earth every few days. However, manual analysis of these images is impractical due to the sheer volume of data. Automatic analysis has historically been error-prone due to the low resolution of these images, where a single pixel corresponds to a 100 m² plot of land.

Fortunately, modern deep learning methods have proven to be highly accurate at extracting data from satellite imagery. With enough training examples, deep learning models can accurately identify features like wind turbines as well as humans can. To power Satlas, the Allen Institute for AI has developed high-accuracy deep learning models for each geospatial data product. These models are then applied to Sentinel-2 images every month to provide an up-to-date global snapshot of each data product.

The Potential of Satlas

The geospatial data products offered by Satlas have immense potential for a wide range of applications in planetary and environmental monitoring. For example, Skylight at AI2 is already exploring the use of marine infrastructure data to improve the classification of vessel movement trajectories. The Institute is actively seeking other use cases for Satlas and welcomes collaboration in exploring its full potential.

Training Data and Foundation Models

Developing high-accuracy deep learning models relies on a large number of high-quality training examples. The Allen Institute for AI has manually labeled thousands of wind turbines, solar farms, offshore platforms, and tree cover canopy percentages in Sentinel-2 imagery and has openly released these training examples and the corresponding model weights. To maximize accuracy, foundation models for Sentinel-2 have been leveraged, which were pre-trained on a large-scale remote sensing dataset called SatlasPretrain. This dataset combines terabytes of Sentinel-2 images with millions of labels across various tasks such as land cover segmentation, crop type classification, and building detection. The diversity of images and tasks in SatlasPretrain enables these foundation models to learn descriptive representations of Sentinel-2 satellite images that are robust across different seasons and geographies.

Super-Resolution: Enhancing Low-Resolution Satellite Imagery

The Allen Institute for AI has also embarked on enhancing the detail of low-resolution but frequent satellite imagery captured by Sentinel-2, a process known as super-resolution. Using deep learning models, the Institute generates high-resolution images from multiple low-resolution images of the same location captured at different times. These models combine information across low-resolution images to predict sub-pixel details. Output images from the current super-resolution model have been computed globally and can be viewed on the Satlas website. The Institute plans to continue exploring methods to improve and quantify accuracy in this area.

What’s Next for Satlas?

In the short term, the primary goal is to expand the geospatial data products offered by Satlas. The Institute is currently exploring models for mapping urban land use, crop types, and land cover, with the aim of incorporating a subset of these into Satlas by the end of 2023. Efforts to improve the accuracy of the existing data products are also underway. In the long term, the Institute plans to release tools that simplify the process of building similar geospatial data products, including annotating examples, training models, and deploying them.

Satlas is revolutionizing the way we explore and analyze geospatial data. With its AI-powered capabilities and commitment to expanding the range of available data products, Satlas promises to be an indispensable resource for researchers, decision-makers, and anyone interested in harnessing the power of geospatial information for a sustainable future.

Summary: Monitoring the Planet with AI and Satellite Imagery: Satlas | Written by Favyen Bastani | August 2023

The Allen Institute for AI has launched Satlas, a platform that uses AI to analyze satellite imagery and generate global geospatial data. Satlas currently offers three data products, including offshore wind turbine detection. The data is available for visualization on the Satlas Map or for download, and an open license allows for offline analysis. The institute plans to release more geospatial data products in the future. The accurate and up-to-date global data can be used for emissions reduction, disaster relief, urban planning, and other applications.

Frequently Asked Questions

1. What is Satlas?

Satlas is a cutting-edge platform that combines AI technology with satellite imagery to monitor the planet effectively. It enables users to gain valuable insights into various aspects of the Earth’s condition and track changes over time.

2. How does Satlas work?

Satlas utilizes advanced artificial intelligence algorithms to analyze satellite imagery captured from various Earth observation satellites. These algorithms can classify and detect objects, changes, patterns, and more on the Earth’s surface, providing accurate and up-to-date information to users.

3. What can I monitor with Satlas?

With Satlas, you can monitor a wide range of factors related to the Earth’s environment, including but not limited to: land use and land cover changes, deforestation, urban expansion, agricultural activities, natural disasters, climate patterns, and much more.

4. Can Satlas be used for research purposes?

Absolutely! Satlas is a powerful tool for scientific research. Its AI capabilities and vast imagery database allow researchers to study and analyze various environmental phenomena, contributing to a better understanding of our planet and potential solutions for global challenges.

5. Is Satlas accessible to the general public?

Yes, Satlas is accessible to anyone who wants to explore and understand our planet better. Whether you are an individual, a researcher, or a business entity, you can make use of Satlas to access reliable and detailed information about the Earth’s condition.

6. How often is the satellite imagery updated?

The satellite imagery in Satlas is regularly updated. The frequency of updates may vary depending on the satellite sources and the specific regions of interest. However, Satlas strives to provide the most recent and accurate data available to its users.

7. Can I download or export the data from Satlas?

Yes, Satlas allows users to download and export the data obtained from its platform. This feature enables users to conduct further analysis, integrate the data into their workflows, or visualize it using other tools and software.

8. How secure is my data on Satlas?

Satlas takes data security and privacy seriously. It employs industry-standard security measures and encryption protocols to safeguard user data. Additionally, Satlas adheres to strict privacy policies and regulations to ensure the confidentiality and integrity of user information.

9. Can Satlas be accessed on mobile devices?

Absolutely! Satlas is designed to be accessible on various devices, including mobile phones and tablets. You can conveniently access Satlas and its features on the go, allowing for greater flexibility and convenience in monitoring the planet.

10. Is there customer support available for Satlas users?

Yes, Satlas provides dedicated customer support to assist users with any inquiries or issues they may encounter. The support team is available via email, phone, or through the Satlas website’s chat feature, ensuring a seamless user experience.