Lead scientist Clinton Stipek has developed a machine learning algorithm capable of predicting building height with an impressive accuracy of within three meters. According to a report published by ORNL, Stipek's algorithm pushes past prior limitations by analyzing building morphology features.
Previous methods, which needed extra information such as roads or internal occupancy counts, now seem antiquated compared to Stipek's approach that uses an open-source code, allowing for height estimations to be conducted on a global scale. As reported in the findings of their research in Scientific Reports, "We developed an open-source code that uses building morphology features at a building level to predict a height," said Stipek, as noted by ORNL. The model eschews old barriers, offering a tool that sidesteps issues like defining a building or incorrectly measuring shadows.
Fellow researcher Taylor Hauser, stimulated by the challenge of refining building height data for the USA Structures project, recognized the need for improved accuracy. While manually correcting shadow-induced errors in buildings' height data used by organizations such as FEMA, Hauser realized the potential for a new tool. "I began to develop a tool that can generate a significant number of measurements collectively known as building morphologies to describe aspects of a building particularly in relationship to neighboring buildings," Hauser told ORNL.
Collaborating, Hauser and Stipek incorporated no less than 65 different building morphology features into their height calculation framework. Utilizing a gradient boosting tree algorithm termed XGBoost, the duo were able to cluster data, arriving at conclusions about a structure's height. "The results point to what Hauser refers to as a hidden signal to predict the third dimension, height, just by using two-dimensional building morphologies," Hauser explained in discussing their approach to drawing inferences from the gathered data, as per ORNL.
This development is more than just an academic achievement; it has real-world utility. It's being integrated into major initiatives like LandScan, a population mapping program, and Global Building Intelligence, aiming to attribute comprehensive building features. As cities expand and evolve, it's estimations like these that offer insights crucial for infrastructure and urban development planning. The data derived from such machine learning tools have applications that stretch far beyond their initial intent, as Hauser puts it: “We can learn more about city planning, buildings, roads, infrastructure. We’re trying to fill in gaps where information doesn’t exist,” in a statement obtained by ORNL.
