MOSAIKS: A Generalizable and Accessible Approach to Machine Learning with Global Satellite Imagery
An aerial photo enhanced with lidar data of Māngere Mountain in Auckland, New Zealand. Image from Daniel Coe
Policy Context
MOSAIKS is a new approach developed at UC Berkeley and UC Santa Barbara that applies machine learning to globally available satellite imagery to estimate economic and socio-economic conditions on the ground. The method relies on a simple mathematical summary of the information in a satellite image that can be used to predict a wide range of ground conditions, from ecosystem transformations to economic development. In the past, to do this kind of work a big team with huge quantities of computing power was needed, meaning that many people, particularly in remote or data-poor communities, could not afford to participate. The MOSAIKS approach allows one person with a standard laptop computer and basic statistical training to use satellite imagery to solve the problems that are relevant in their local context. The goal is to provide an accessible and affordable way for everyone from researchers to government officials to local NGOs to use satellite imagery to monitor environmental change, evaluate program interventions, and measure other outcomes of interest.
Study Design
In a forthcoming publication, CEGA affiliates Solomon Hsiang, Tamma Carleton, and collaborators develop, test, and demonstrate the power of MOSAIKS as an affordable approach to predicting diverse ground conditions from satellite imagery. The release of this paper will enable users to replicate the team’s predictions of seven “tasks” (e.g., population density, average income) using satellite imagery. However, to truly make the MOSAIKS tool accessible and easily implementable for users who are interested in predicting other ground conditions in different localities across the globe, they aim to provide a user interface (i.e., API) that enables anyone to repeat the simple MOSAIKS pipeline in their own context. CEGA is currently supporting the MOSAIKS team to build out this data infrastructure, expand the universe of predictable tasks, and to identify new use cases and collaborations with potential users of the platform. To learn more, explore the Global Policy Lab MOSAIKS homepage, or watch Solomon Hsiang’s keynote address on MOSAIKS at CEGA and the World Bank’s 2020 Infrastructure for Development (Infra4Dev) conference.
Results and Policy Lessons
Results from the CEGA supported scale-up of MOSAIKS are forthcoming.
