SEMINAR: Analyzing the urban built environment from space

SEMINAR: Analyzing the urban built environment from space

Seminar Abstract 

In recent years, the availability of remote sensing data on the urban built environment has increased dramatically, as wide-area remote sensing has become more commoditized with new commercial satellite companies entering the market. On the other hand, advances in machine learning and computing power have lead to successes in computer vision and time series tasks that are of interest to analyzing data on buildings such as imagery and meter time series. This talk presents work on analyzing the urban built environment at several spatial scales, from cities to blocks to individual buildings situated in urban environments. First, we present a framework based on generative adversarial networks (GANs) to characterize the spatial distribution of urban built areas and relate it to population density and to proxies to energy use using data on more than 10,000 largest cities worldwide. Going down one further level of detail to the city block level, we present a method using convolutional networks to compare urban environments across cities from cheaply-obtained satellite data paired with land use surveys for supervised training. We see this as a way of cheaply creating land use maps in places where no survey data is available. Finally, we discuss implications of this work and challenges of machine learning methods to on-going work on building energy benchmarking.

Seminar Speaker(s) 

Adrian Albert

Adrian Albert is currently a postdoctoral scholar in the Civil Engineering department at MIT, and a visiting scientist in the Applied Program at SLAC. His work is on machine learning methods for analyzing energy and infrastructure systems in cities. Prior to that, he spent two years as a Senior Data Scientist at Redwood City-based startup C3 IoT, working on machine learning for predictive maintenance of large-scale power systems and industrial machinery. He obtained his PhD in Electrical Engineering from Stanford University, with a thesis on machine learning algorithms for smart grid applications. He holds a MS degree in Management Science and Engineering from Stanford University, and an MS in Astrophysics, and BS degrees in Physics and in Applied and Computational Mathematics from Jacobs University Bremen, Germany. 


Jun 9, 2017 -
12:00pm to 1:00pm