FireAndStructure

An initial exploration of Vibrant Planet's California structure graph, a network of about 13.6 million buildings in which buildings within 100 m of each other are connected, and how building connectivity relates to wildfire risk.

The exploration looks at:

• How much data there are and the general characteristics of the connectivity attributes (neighbor counts, clustering coefficient, separation distances), overall and by ecoregion and county
• How many buildings are in areas where building connectivity is likely to matter for fire spread: outside the Riley et al. (2025) wildland risk area and in areas with less than 10% tree canopy cover
• Whether building connectivity is correlated with county median family income and racial/ethnic minority percentage
• Whether building connectivity is already captured by CarbonPlan's Open Climate Risk building-level risk estimates (Mariposa, Shasta, and Sacramento counties)

Key findings

• The neighbor counts (buildings within 10, 50, and 100 m) are strongly correlated with each other; the 50 m and 100 m counts are close to redundant (Spearman 0.94).
• Rural counties such as Mariposa have the highest average clustering coefficients and urban counties such as Los Angeles and San Francisco the lowest, because urban buildings have many neighbors spread far enough apart that they are not connected to each other.
• Of all buildings, 39% are outside the Riley et al. (2025) wildland risk area but within 500 m of it. After also removing buildings in areas with 10% or more tree cover, about 8.1 million buildings remain.
• Building connectivity is not strongly correlated with Open Climate Risk's 2047 risk estimates in the three counties examined (strongest Spearman correlation −0.41), and the direction differs between counties, suggesting connectivity could add information not already captured.
• About 102,000 buildings (0.75%) have no building ID or connectivity values. They are concentrated in rural counties and are likely buildings with no other buildings within 100 m.

These are first-pass results; see the notebook for details and caveats.

Repository contents

File	Description
Data_Exploration.ipynb	Main notebook with the full exploration, figures, and notes
node_summaries.py	summarize_nodes(): summary tables, histograms, and correlation matrices of building attributes, overall or by a grouping such as county or ecoregion
landcover.py	mean_tree_canopy() (and nlcd_cover_fractions()): summarize land cover rasters in a neighborhood around each building, reading the raster in strips to limit memory
risk_correlations.py	correlate_connectivity_with_risk(): match a county's Open Climate Risk building points to building footprints and correlate risk with connectivity
environment.yml	Conda environment with the package versions used
Dockerfile	Builds a Docker image with the environment, notebook, and helper modules (see Docker)

Setup

The code uses Python 3.12 with packages from conda-forge, listed with their versions in environment.yml. You can set it up with conda or run it in Docker.

Conda

conda env create -f environment.yml
conda activate firestructure

Then start Jupyter from the repository folder and open Data_Exploration.ipynb:

jupyter notebook

Docker

The Dockerfile builds an image with the same environment, the notebook, and the helper modules. The data are not included in the image; your local Data/ folder is attached when the container runs.

Build the image from the repository folder:

docker build -t firestructure .

Run it, attaching your Data/ folder:

docker run --rm -p 8888:8888 -v "$PWD/Data:/home/mambauser/FireAndStructure/Data" firestructure

Then open the http://127.0.0.1:8888/tree?token=... link printed in the terminal.

Notes:

• Files the notebook saves to Data/ are kept on your computer. Edits to the notebook itself are lost when the container stops, because the notebook is copied into the image. To keep your edits, attach the whole repository folder instead: -v "$PWD:/home/mambauser/FireAndStructure".
• Rebuild the image after changing the notebook or the .py files so the image has the latest versions.
• The full notebook needs more memory than Docker Desktop's default. Increase it under Docker Desktop → Settings → Resources → Memory (for example, to 12–14 GB).

Running the notebook

1. Download the manual datasets into a Data/ folder in the repository (about 6.1 GB in total):

   File	Size	Source
   ca_vpdc_nodes.parquet	2.4 GB	Vibrant Planet dataset page
   ca_vpdc_edges_0_10.parquet	1.8 GB	Vibrant Planet dataset page
   RDS-2025-0006.zip	1.7 GB	Riley et al. (2025) archive page
   Mariposa-County-06043.geojson, Shasta-County-06089.geojson, Sacramento-County-06067.geojson	6 MB, 30 MB, 141 MB	Open Climate Risk

   Only these two structure graph files are needed; the full structure graph dataset (all edge files) is over 30 GB.

2. Run the notebook from top to bottom. Later sections depend on variables created earlier (for example points, counties, and nodes_filtered2).

3. The other datasets are downloaded (about 330 MB: county boundaries, ecoregions, and the two ACS tables) or read remotely the first time the notebook runs and are saved to Data/ so later runs are faster: ca_counties.gpkg, ca_ecoregions.gpkg, riley2025_BP_2011ClimateRun_CA.tif, nodes_filtered_tree_canopy_2025_100m.parquet, and ca_county_acs_2024.csv. Delete a saved file to recalculate it.

Notes:

• The full node table is large (about 13.5 million building footprints), so the notebook needs a fair amount of memory; it was developed on a computer with 16 GB of RAM.
• The first tree canopy run reads the USFS tree canopy raster over the internet and can take a long time. The result is saved afterward.
• The tree canopy year (veg_year) and neighborhood radius (radius_m) are set at the top of the notebook.

Data

The data used in this project are not included in this repository. The Data/ folder is ignored by git; download the datasets below into it to run the notebook.

Dataset	Source	License	How to get it
California structure graph (building nodes and edges)	Vibrant Planet Data Commons	CC BY-NC-SA 4.0	Download ca_vpdc_nodes.parquet and ca_vpdc_edges_0_10.parquet manually from the dataset page
Wildfire risk components (burn probability, flame length probability)	Riley et al. 2025, USDA Forest Service Research Data Archive	Public domain	Download RDS-2025-0006.zip manually from the archive page
Open Climate Risk building-level risk (per county)	CarbonPlan	ODbL (building-level data)	Download manually for each county from the Open Climate Risk map (see below)
County boundaries (2023 TIGER/Line)	U.S. Census Bureau	Public domain	Downloaded by the notebook
Median family income and race/ethnicity by county (ACS 2020–2024 5-year, tables B19113 and B03002)	U.S. Census Bureau	Public domain	Downloaded by the notebook
Ecoregions 2017	RESOLVE	CC BY 4.0	Downloaded by the notebook
NLCD Tree Canopy Cover (v2025.6)	USDA Forest Service	Public domain	Read remotely by the notebook

Open Climate Risk county downloads

The notebook compares building connectivity with Open Climate Risk building-level risk (rps_2047) for individual counties. For each county:

1. Open the Open Climate Risk map and navigate to the county.
2. Select the county in the Risk in the region section of the sidebar and download the building-level data.
3. Save it in Data/ as <County>-County-<FIPS>.geojson. The notebook currently uses:
   • Mariposa-County-06043.geojson
   • Shasta-County-06089.geojson
   • Sacramento-County-06067.geojson

More about the data and methods: Open Climate Risk documentation and explainer.

Attribution

California structure graph data: © 2025 Vibrant Planet. Distributed by Vibrant Planet Data Commons. Licensed under CC BY-NC-SA 4.0. https://www.vpdatacommons.org/datasets/ca-stucture-graph-data

Wildfire risk components: Riley, Karin L.; Zimmer, Scott N.; Kodra, Evan; Grenfell, Isaac C.; Dillon, Gregory K.; Scott, Joe H.; Jaffe, Melissa R.; Olszewski, Julia H.; Vogler, Kevin C.; Finney, Mark A.; Short, Karen C.; Jolly, W. Matthew; Brittain, Stuart E.; Callahan, Michael N. 2025. Spatial datasets of probabilistic wildfire risk components for the conterminous United States (270m) for circa 2011 climate and projected future climate circa 2047. Updated 01 July 2025. Fort Collins, CO: Forest Service Research Data Archive. https://doi.org/10.2737/RDS-2025-0006 The originator notes these data are intended for national-scale strategic planning; applicability to smaller areas varies by location.

Open Climate Risk: CarbonPlan, Open Climate Risk. https://carbonplan.org/research/climate-risk. Building-level data licensed under the Open Database License (ODbL); see the data access documentation.

Ecoregions: Dinerstein, E., et al. 2017. An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm. BioScience 67(6): 534–545. https://doi.org/10.1093/biosci/bix014 Data: RESOLVE Ecoregions 2017, https://ecoregions.appspot.com/, licensed under CC BY 4.0.

Tree canopy cover: USDA Forest Service, NLCD Tree Canopy Cover, version 2025.6. https://data.fs.usda.gov/geodata/rastergateway/treecanopycover/

County boundaries: U.S. Census Bureau, 2023 TIGER/Line Shapefiles: Counties. https://www.census.gov/geographies/mapping-files/time-series/geo/tiger-line-file.html

Income and race/ethnicity: U.S. Census Bureau, American Community Survey 2020–2024 5-year estimates, tables B19113 (median family income) and B03002 (Hispanic or Latino origin by race). https://www.census.gov/programs-surveys/acs

License

• Code (.py files and the code in the notebooks) is licensed under the MIT License.
• Notebook outputs (tables, figures, maps and printed data) are derived from the Vibrant Planet California structure graph data and are therefore licensed under CC BY-NC-SA 4.0, not MIT. They may be shared and adapted for noncommercial purposes only, with the attribution above, under the same license. Outputs that use Open Climate Risk data also carry the Open Climate Risk attribution above.
• The datasets themselves remain under their original licenses listed above.