Satellites turn “concrete”: tracking cement with satellite data and neural networks

The Covid crisis has demonstrated the need for alternative data, in real-time and with global coverage. This paper exploits daily infrared images from satellites to track economic activity in advanced and emerging countries. We first develop a framework to read, clean and exploit satellite images. We construct an algorithm based on the laws of physics and machine learning to detect the heat produced by cement plants in activity. This allows to monitor in real-time if a cement plant is functioning. Using this information on more than 500 plants, we construct a satellite-based index tracking activity. Using this satellite index outperforms benchmark models and alternative indicators for nowcasting the activity in the cement industry and in the construction sector. Exploring the granularity of daily and plant-level data, using neural networks yields significantly more accurate predictions. Overall, combining satellite images and machine learning allows to track industrial activity accurately.