A Generalizable and Accessible Approach to Machine Learning with Global Satellite Imagery

63 Pages Posted: 9 Nov 2020 Last revised: 19 Nov 2021

See all articles by Esther Rolf

Esther Rolf

University of California, Berkeley

Jonathan Proctor

Harvard University

Tamma Carleton

University of Chicago

Ian Bolliger

University of California, Berkeley - Energy and Resources Group

Vaishaal Shankar

University of California, Berkeley

Miyabi Ishihara

University of California, Berkeley

Benjamin Recht

University of California, Berkeley - Department of Electrical Engineering & Computer Sciences (EECS)

Solomon Hsiang

University of California, Berkeley; National Bureau of Economic Research

Date Written: November 2020

Abstract

Combining satellite imagery with machine learning (SIML) has the potential to address global challenges by remotely estimating socioeconomic and environmental conditions in data-poor regions, yet the resource requirements of SIML limit its accessibility and use. We show that a single encoding of satellite imagery can generalize across diverse prediction tasks (e.g. forest cover, house price, road length). Our method achieves accuracy competitive with deep neural networks at orders of magnitude lower computational cost, scales globally, delivers label super-resolution predictions, and facilitates characterizations of uncertainty. Since image encodings are shared across tasks, they can be centrally computed and distributed to unlimited researchers, who need only fit a linear regression to their own ground truth data in order to achieve state-of-the-art SIML performance.

Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.

Suggested Citation

Rolf, Esther and Proctor, Jonathan and Carleton, Tamma and Bolliger, Ian and Shankar, Vaishaal and Ishihara, Miyabi and Recht, Benjamin and Hsiang, Solomon, A Generalizable and Accessible Approach to Machine Learning with Global Satellite Imagery (November 2020). Available at SSRN: https://ssrn.com/abstract=3727121

Esther Rolf (Contact Author)

University of California, Berkeley

310 Barrows Hall
Berkeley, CA 94720
United States

Jonathan Proctor

Harvard University ( email )

1875 Cambridge Street
Cambridge, MA 02138
United States

Tamma Carleton

University of Chicago ( email )

1101 East 58th Street
Chicago, IL 60637
United States
7737026763 (Phone)

Ian Bolliger

University of California, Berkeley - Energy and Resources Group ( email )

United States

Vaishaal Shankar

University of California, Berkeley

310 Barrows Hall
Berkeley, CA 94720
United States

Miyabi Ishihara

University of California, Berkeley

310 Barrows Hall
Berkeley, CA 94720
United States

Benjamin Recht

University of California, Berkeley - Department of Electrical Engineering & Computer Sciences (EECS) ( email )

Berkeley, CA 94720-1712
United States

Solomon Hsiang

University of California, Berkeley ( email )

2607 Hearst Avenue
Berkeley, CA 94720-7320
United States

HOME PAGE: http://gspp.berkeley.edu/directories/faculty/solomon-hsiang

National Bureau of Economic Research ( email )

1050 Massachusetts Avenue
Cambridge, MA 02138
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
9
Abstract Views
145
PlumX Metrics