USDA Food Access Research Atlas

USDA Food Access Research Atlas data is the USDA Economic Research Service's census-tract picture of how far Americans live from food retailers: 84,119 tracts per table across the 50 states plus DC, each scored for access to SNAP-authorized stores at 0.5, 1, 10 and 20 mile thresholds, measured both as straight-line distance and as road-network driving distance. Every tract carries the low-income/low-access (LILA) flags that formally mark a food desert, beside poverty rates, median family income, carless households, SNAP participation and full race and ethnicity counts. The unit of analysis is the census tract, keyed on the tract GEOID - the same key Census demographic tables use, so every access measure joins onto population baselines without a geocoding project.

Two editions ship together. The 2025 SNAP-authorized Retailer Access Map, released July 2026, measures every authorized outlet from the June 2025 STARS store directory and adds a veteran-population focus. The 2019 Large Retailer Access Map keeps the supermarket-and-supercenter view from a merged 45,233-store directory.

Get a sample of this dataset to see real tract rows with the complete field dictionary before you commit.

Two views of the same extract. The first is General Tract Characteristics - one row per census tract - shown here for Autauga County, Alabama. The second is a slice of the variable-lookup file that maps every field code to its long name, joined alongside whichever distance table you take.

TRACT CHARACTERISTICS - one row per census tract
| CENSUSTRACT20 | STATE   | COUNTY20       | URBAN | POP2020 | POVERTYRATE | MEDIANFAMILYINCOME | TRACTSNAP |
| ------------- | ------- | -------------- | ----- | ------- | ----------- | ------------------ | --------- |
| 1001020100    | Alabama | Autauga County | 1     | 1775    | 17.4        | 76979              | 100       |
| 1001020200    | Alabama | Autauga County | 1     | 2055    | 9.5         | 62331              | 143       |

VARIABLE LOOKUP - joined to every access measure
| FIELD                     | LONG NAME                                                                    |
| ------------------------- | ---------------------------------------------------------------------------- |
| SD_SRAM_LILATracts_1And10 | Low income and low access tract at 1 mile urban / 10 miles rural             |
| SD_SRAM_lapop1share       | Share of tract population beyond 1 mile from an authorized store, straight-line |

The package holds three tract tables of 84,119 rows apiece - characteristics, straight-line distance, driving distance - each paired with its own lookup file and a Read Me. Pivot on the tract key and every measure becomes a column-complete record.

Identity rides first: CensusTract20 doubles as the universal join key across all three tables, flanked by state, county and an urban-rural flag. Demographic context follows - POP2020, occupied housing units, group-quarters counts, then PovertyRate, MedianFamilyIncome, TractLOWI, children, seniors, seven race-and-ethnicity counts, TractHUNV for carless households, TractSNAP for benefit-receiving households and TractVeteran, new with the 2025 edition. The access block then grades every threshold: four SD_SRAM_LILATracts_* variants, the SD_SRAM_HUNVFlag for carless households beyond half a mile, low-access flags at 1 and 10 miles, and lapop, lalowi and lasnap exposure counts with their population shares - mirrored by DD_SRAM_* equivalents computed over the road network.

The dictionary covers the core row schema plus worked access fields; all 41 documented columns arrive mapped in exactly this shape, and any narrower slice you scope ships as additional fields on request.

Four thresholds, two routing methods, one flag family:

1. 0.5 mile (urban) - the walkable-store test behind the SD_SRAM_LILATracts_halfAnd10 variant and the vehicle-based flag, which marks tracts with 100 or more carless households beyond half a mile.
2. 1 mile (urban) / 10 miles (rural) - the headline LILA pairing, SD_SRAM_LILATracts_1And10, and the standard low-access cut behind SD_SRAM_LATracts1 and SD_SRAM_LATracts10.
3. 1 mile (urban) / 20 miles (rural) - the wide-country variant for rural reach, SD_SRAM_LILATracts_1And20.
4. 20 miles (driving) - the network-distance extreme, DD_SRAM_LATracts20, computed over road segments rather than straight lines.

Straight-line measures come from centroid geometry; driving measures route over a commercial street network, which is why the two families disagree in terrain and suburbia alike. Exposure shares travel with each count, so a lapop1share of 93.35 reads immediately as a tract almost wholly beyond a mile, without needing the denominator.

• Geography: all census tracts across the 50 states plus Washington, DC - 84,119 tracts on 2020 polygons for the 2025 edition; the 2019 edition sits on 2010 polygons. Territories are excluded.
• Temporal: the 2025 edition layers the June 2025 STARS store directory over 2020 Decennial Census counts, LandScan USA 2020 population and ACS 2020-24 income, vehicle and SNAP variables, released July 2026; the 2019 edition carries 2019 store estimates with 2014-18 ACS context; archived vintages run 2015, 2010 and the 2006 Food Desert Locator.
• Granularity: census tract, roughly 84,000 rows per table, one row per tract in each of the three files, with distance-based access measured at the 0.5, 1, 10 and 20 mile thresholds.

Your call, not ours: API, files, or your warehouse. Daily, weekly, or hourly.

Every delivery carries the complete field dictionary above, the sample rows, and the coverage profile - mapped and ready, with no integration archaeology on your side.

1. Market researchers scoring any metro's neighborhood structure at once - store access, income, carlessness and benefit participation in one keyed table of 84,119 tracts.
2. Retail and e-commerce operators screening expansion territories with LILA flags and exposure shares before committing a site-selection model.
3. Data scientists joining 41 documented access fields onto an internal tract master, keyed on the GEOID, with every code mapped back to a long name and definition.
4. Developers and builders wiring a normalized tract-GEOID feed into pipelines that already speak Census geography.
5. Journalists and academics citing USDA Economic Research Service tract figures for food-desert reporting and teaching, with a lookup file behind every flag.

Notes and adjacent reading:

• USDA Food Environment Atlas - the county-level sibling: the same access question one census geography wider, with 304 indicators for roughly 3,100 counties.
• SHED Public Use Data Files (2013-2025) - household-level financial well-being to pair with the Atlas's place-level structure.
• FDIC National Survey of Unbanked and Underbanked Households - household banking status when the question turns from store access to financial access.

Browse the whole vertical on the Consumer Finance data hub, the ranked shortlist of the best consumer finance datasets or the best food retail datasets.