Foodenvr. Food Environment Accessibility Measures

An open-source R workflow for computing food environment accessibility measures based on drive times at the household and census tract level, and evaluating how population representation method (census tract centroid, population-weighted centroid, household parcel point) affects measured access to food retailers. The current workflow is for the city of Los Angeles but this workflow is configurable for any U.S. city or county.

---

Prerequisites

• R 4.4.0
• Java 21 (required by r5r; see rJavaEnv for installation)
• RAM: ≥ 16 GB recommended for driving-time computation; 16 GB+ for parcel-level runs
• R packages: installed automatically via 0_Libraries.R, but key dependencies are r5r, sf, tigris, data.table, tidyverse

---

Quick Start (LA County)

config.R — edit these values before running anything else:

• base_path: where you are storing your raw data
• processed_path: where you are storing intermediate data and outputs
• Java memory guidance: set this to roughly 75% of available RAM. Running at 12G requires ~16G total; reduce to 6G for 8G machines and decrease CHUNK_SIZE_CAR in config.R to compensate.

# Step 1 — Download raw data
source("1_data_collection.R")

# Step 2 — Clean food POI and assign categories
source("1_data_cleaning.R")

# Step 3 — Compute accessibility measures (~hours; requires Java + OSM network)
source("2_gen_measures.R")

# Step 4 — Aggregate and merge measures
source("2_summarize_measures.R")

# Step 5 — Compare methods and generate results
rmarkdown::render("3_prelim_results.Rmd")

Each script begins with source("0_Libraries.R"), which loads packages, sets paths, and sources all helpers.

Java memory guidance: set this to roughly 75% of available RAM. Running at 12G requires ~16G total; reduce to 6G for 8G machines and increase CHUNK_SIZE_CAR in config.R to compensate.

---

Adapting to a New City or County

All location-specific parameters live in config.R - edit these values before running anything else:

STUDY_STATE   <- "CA"
STUDY_COUNTY  <- "Los Angeles"
STUDY_CITY    <- "Los Angeles"    # must match tigris::places() NAME field
STUDY_YEAR    <- 2020
OSM_LOCATION  <- "socal"          # geofabrik slug; run osmextract::oe_match()
                                  # to find the right value for your region

Everything else is derived from these values automatically. To see derived variables, go to helper/universal_variables.R

Data you must supply yourself

Two constructs do not have complete public data and either must be provided manually or downloaded from incomplete public databases:

Input	LA County source	Alternatives for other regions
Residential parcel points	LA County Tax Assessor GDB	County assessor data; OpenAddresses; Census LODES/WAC
Food POI	Data Axle (proprietary)	SNAP retailer data (public; downloaded automatically if you set FOOD_POI_SOURCE = "snap" in config.R

Set HOUSEHOLDS_PATH in config.R to point to your parcel file. It must be a point layer with columns: id (integer), GEOID_<year> (character), UseType, EXCLUDE. If parcel data is unavailable, you can skip the parcel-level computation and run only the census tract centroid and population-weighted centroid methods.

---

Data sources

Data	Source	Access	Script
Census tracts + blocks	Census TIGER/Line via tigris	Public	1_data_collection.R
Street network (OSM)	Geofabrik via osmextract	Public	1_data_collection.R
SNAP retailer locations	USDA via ArcGIS REST	Public	1_data_collection.R
Food POI (primary)	Data Axle	Proprietary	1_data_cleaning.R
Residential parcels	County assessor	Must supply	2_gen_measures.R

---

Pipeline

Script	Section	Description
0_Libraries.R	—	Load packages, set paths, source helpers
1_data_collection.R	A	Download census, OSM, SNAP, and health data
1_data_cleaning.R	B	Clean Data Axle POI; assign food categories via NAICS
2_gen_measures.R	C	Compute drive-time accessibility with r5r (chunked)
2_summarize_measures.R	D	Merge chunks; aggregate parcels to census tract level
3_prelim_results.Rmd	E	Compare methods; produce tables and maps

Food retailer categories

POIs are classified into seven categories mapped via NAICS codes:

Code	Category
CNV	Convenience store
FF	Fast food
GRC	Grocery
RR	Restaurant
SMK	Supermarket
SPF	Specialty food
Not.included	Excluded retailers

Note: SNAP retailer data do not have restaurants or fast food restaurants as a category since most states do not authorize them as SNAP-eligible food retailers (with a few exceptions).

Population representation methods

Measures are produced for three origin point types:

• ct_cent — unweighted census tract centroids
• ct_wtcent — population-weighted centroids (using 2020 census block counts)
• parcel — individual residential parcels (household-level ground truth)

Drive-time cutoffs can be specified by the user but generally are at 5, 10, 15 minute drive time increments. Walkability may also be calculated by setting mode="WALK" in the function call compute_accessibility from the file 2_gen_measures.R

---

Outputs

Final outputs are written to processed_path/LAC_cleaned/:

File	Contents
ct_driving_times.csv	All three methods merged at census tract level
parcel_drivingdt.csv	Household-level measures (long format)
parcel_driving_all.csv	Household measures merged with census tract data
dt_household_ct.csv	Parcel measures aggregated to census tract (mean, median, SD, CV)

Column naming convention: {network}_{category}_{cutoff}_{method} (e.g., driving_SMK_15_ct_wtcent).

---

Citation

If you use this pipeline, please cite:

Zhang AW, Cai Y, Macdonald B, Shah P, Espinoza J, Wilson J (Under review). Foodenvr: Quantifying error and bias in food environment measures using an open source workflow.

---

License

Foodenvr is open source and licensed under the MIT License. OpenStreetMap's Open Database License requires that derivative works provide proper attribution.