NOTE: The data required to rebuild this work can be found here with the Stanford Libraries data repository. You may also feel free to cite the data source independently and reference it using this DOI: https://doi.org/10.25740/sz846yd5641. Download the zip file and unzip it into this directory to rebuild the structure required.
This work utilizes a new data collection pipeline provided by prior work 
Data computed at intermediate steps is included, such that if you are only interested in rebuilding the final figures and results you may do this using the 3c_feature_selection.R file. We recommend using an editor like R studio to help manage dependencies and configure the path to the current working directory of this repository.
To examine how the data was split and cleaned, you may instead explore the two jupyter notebooks: 2d_missing.ipynb and 3a_datasplit.ipynb. Additional to the paper you may find completeness counts for each dataset and its temporal coverage of each building found in the directory 2d.
The collection process itself was computed by using data/footprints.geojson as an input into the our data collection pipeline, between the dates of January 2018 and January 2021.