Scientists Developing New Machine Learning Methods for Easier Insight into Massive Satellite Map Data

Currently, there are over 700 imaging satellites orbiting the Earth, transmitting massive amounts of information back to ground databases every day, either for monitoring climate change or tracking health and poverty issues. But scientists also face a challenge: while geospatial data can help researchers and policymakers address key challenges, only those with significant wealth and professional knowledge can access it.

Now, a team at the University of California, Berkeley has designed a machine learning system to solve satellite image problems. They use low-cost and easy-to-use technologies that can provide access and analytical capabilities for researchers and governments around the world. This study, titled "A Universal and Accessible Method for Machine Learning Using Global Satellite Images," was published in the journal Nature Communications on July 20, 2021.

The co author of the project, Esther Rolf, stated, "Satellite images contain a wealth of data about the world, but the trick is how to transform the data into useful insights without the need for manual editing of each image. We have designed our system for accessibility, so one person should be able to run it on a laptop without the need for specialized training to solve their local problems.".

The co authors of this paper, Solomon Hsiang, Director of the Global Policy Laboratory at Goldman Sachs School of Public Policy, stated: We hope that our actions can bring global impact. The pace of development is faster than ever before. We are changing resource allocation faster than ever before. We are changing the Earth. This requires a more responsive management system that can see these things happening so that we can respond promptly and effectively.

This project is being promoted in collaboration between the Global Policy Lab led by Hsiang and the research team from the Department of Electrical Engineering and Computer Science led by Benjamin Recht. Other collaborators include Tamma Carleton, a Berkeley PhD graduate currently studying at the University of California, Santa Barbara, Jonathan Proctor currently working at the Harvard Center for Environmental and Data Science Programs, Ian Bollinger from Rhodium Group, Vaishaal Shankar from Amazon, and Miyabi Ishihara, a Berkeley PhD student.

Prev: How will intelligent transportation technology once again drive urban development?