Floor space is a key variable used to understand the energy and material demands of buildings. Using recent data sets of building footprints, we employ a random forest regression model to estimate the total floor space (conditioned and unconditioned) of the North American building stock. Our estimate for total floor space in 2016 is 88033 (+15907/-21861) million m2, which is 2.9 times higher than current estimates from national statistics offices. We also show how floor space per capita (m2 cap-1) is not constant across the North American region, highlighting the heterogeneous nature of building stocks. As a critical variable in integrated assessment models to project energy and material demands, this result suggests that there is much more unconditioned floor space than previously realized. Furthermore, when estimating material stocks, flows, and associated embodied carbon emissions, total floor space per-capita estimates, such as those presented in this study, offer a more comprehensive approach in comparison to national statistics that do not capture unconditioned floor space. This result also calls for an investigation as to why there is such a vast difference between estimates of conditioned and total floor space.
Keywords: North America; building stock; floor space; machine learning.