Wendell Cathcart, MEM
It’s easy to take the technological marvel that is satellite imagery for granted when apps like Google Maps allow us to casually swipe between continents. For most of us (GIS-aficionados excluded), satellite imagery just doesn’t seem to have much bearing on our lives. This may soon change, however, as recent developments in machine learning promise to unlock valuable energy information hidden in troves of satellite images.
The field of Machine Learning (a sub-field of Artificial Intelligence) underwent a revolution in 2012 with the emergence of “Deep Learning,” a technique that enables computers to mimic the brain’s ability to learn and accurately identify subtle patterns in data. When applied to object detection in satellite images, Deep Learning models can examine a region to identify automobiles, solar PV, and other energy assets that shed light on the region’s energy consumption and production. This type of analysis would be prohibitively slow to perform by hand and its accuracy can rival active surveying methods.
Interestingly, Deep Learning models can be trained to predict the economic well-being of a region by comparing satellite images of the region during the daytime and nighttime. The model finds features on the ground to explain relative economic prosperity (well-lit at night) and poverty (low-lit at night), and when combined with other data, can generate surprisingly insightful predictions on test (daytime) images. With poverty, development, and energy-consumption inextricably linked, it’s important that we use all of the tools at our disposal to monitor economic development and guide policies that ensure prosperity coincides with renewable energy.