Nature Communications volume 12, Article number: 4392 (2021 ) Cite this article
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.
Addressing complex global challenges—such as managing global climate changes, population movements, ecosystem transformations, or economic development—requires that many different researchers and decision-makers (hereafter, users) have access to reliable, large-scale observations of many variables simultaneously. Planet-scale ground-based monitoring systems are generally prohibitively costly for this purpose, but satellite imagery presents a viable alternative for gathering globally comprehensive data, with over 700 earth observation satellites currently in orbit1. Further, application of machine learning is proving to be an effective approach for transforming these vast quantities of unstructured imagery data into structured estimates of ground conditions. For example, combining satellite imagery and machine learning (SIML) has enabled better characterization of forest cover2, land use3, poverty rates4 and population densities5, thereby supporting research and decision-making. We refer to such prediction of an individual variable as a task. Demand for SIML-based estimates is growing, as indicated by the large number of private service-providers specializing in predicting one or a small number of these tasks.