Track 5 (Food Access & Resource Navigation)
Where will food be needed next?
Some coding helps
The problem
Food banks in the Houston area send food to many partner agencies and pantries. When a hurricane, a freeze or a wave of job losses hits, need can jump fast, and not always in the usual places. Planners have to decide where to send extra food and mobile pantries before the full numbers come in. Build an early-warning tool that shows where need is likely to rise next month, and test it honestly on past months.
Who it's for
Food bank staff who plan where food and mobile pantries go each month.
What to build
The minimum is a target, not a rule. Judges score every team with the same five criteria.
Build at minimum
An early-warning view for 8 counties: Harris and the 7 counties that border it (Brazoria, Chambers, Fort Bend, Galveston, Liberty, Montgomery and Waller). Every team uses the same 8, so judges can compare. It shows:
- which counties are most likely to see need go up next month, using public signals (public numbers like weekly unemployment claims, SNAP cases and disaster declarations);
- a test on past months. Pick a month, use only data from before it, make your prediction, then check what really happened. Do this for every month from January 2023 to August 2025;
- how your tool compares with a simple guess: "next month will be the same as last month." Show how often you beat it. If you don't, say so. That still counts;
- how sure the tool is: a range for each county, based on how far off it was in your test.
What you check against: the change in SNAP cases. It's the only public monthly count for each county, so it's what you can test. It isn't the same as need (see Good to know).
Stretch goals
- Suggest where to send 10 extra mobile pantry visits across the 8 counties, and explain each one in plain language.
- Look closely at Hurricane Beryl (July 2024). Did your tool flag the right counties in time?
- Add the newer months (September 2025 to now) from the SNAP dashboard.
- A "what if a hurricane hits next week" scenario.
- Plan simple delivery routes from a public building you pick, like a county courthouse. Don't use a real food bank or pantry site.
The data
Public data only, with no personal information. Links checked October 25, 2026.
| Dataset | What's in it | Official source |
|---|---|---|
| Texas SNAP cases by county, monthly | SNAP (food stamps) cases and eligible people by county. One Excel file per month (.xls or .xlsx), January 2022 to August 2025. From September 2025, an interactive dashboard instead (the latest month downloads as CSV). Older years by request to HHSC. Public (Texas Health and Human Services) | Texas HHSC SNAP statistics (opens in a new tab) |
| Weekly unemployment claims by county | New unemployment claims each week for every Texas county, January 2019 to about two weeks ago. One Excel file, with a row per week and a column per county. Public (Texas Workforce Commission) | Texas LMI downloads (choose "UI Claims") (opens in a new tab) |
| FEMA disaster declarations | One row per county per declared disaster, back to 1953 (counties from 1964). About 70,500 rows in all. API and CSV. Public (FEMA) | OpenFEMA dataset page (opens in a new tab) · API (opens in a new tab) |
| County unemployment, monthly | Labor force and unemployment by county, latest 14 months, in one ZIP file. Monthly county data back to 1990 through the BLS database tools or the free BLS API (no key needed). October 2025 is missing because of the government shutdown. Public (US Bureau of Labor Statistics) | BLS Local Area Unemployment tables (opens in a new tab) |
| SNAP households by census tract | Share of households getting SNAP, by census tract (small areas of a few thousand people), 5-year estimates (table S2201). Public (US Census Bureau) | Census table S2201 (opens in a new tab) |
| Food access by census tract | Distance to large food stores, by 2010 census tract. Excel or ZIP. Formerly the Food Access Research Atlas, now called the 2019 Large Retailer Access Map. The same page has a newer 2025 map of stores that take SNAP, on 2020 tracts. Public (USDA Economic Research Service) | USDA ERS download page (opens in a new tab) |
| County food insecurity estimates | Food insecurity by county for 2024, from Feeding America's Map the Meal Gap 2026 interactive map. The full dataset needs a request form. Check Feeding America's terms before you reuse it | Map the Meal Gap by county (opens in a new tab) · Interactive map (opens in a new tab) |
| County and tract boundaries | TIGER/Line shapefiles, for joining the data to a map. Public (US Census Bureau) | Census TIGER/Line files (opens in a new tab) |
| Texas road network | OpenStreetMap roads, from the Geofabrik Texas extract (about 700 MB). Only needed if you run your own route planner. ODbL license: credit "© OpenStreetMap contributors" | Geofabrik Texas extract (opens in a new tab) · License (opens in a new tab) |
| Routing engine | OSRM, a free, open-source route planner that runs on OpenStreetMap data. Its free demo server needs no setup: send no more than one request per second. BSD license | OSRM (opens in a new tab) |
Start here
- No code? Download 12 monthly SNAP files, put the 8 counties in Google Sheets, and chart how cases changed month by month. Then add weekly unemployment claims for the same counties and look for jumps.
- New to Python? See open a Colab notebook and load a CSV.
Good to know
About the data
- Real food bank shipments, routes and pantry stock aren't public, so you test against SNAP cases instead.
- SNAP counts people who get benefits, not people who need food. Rule changes in 2025 made fewer people eligible. So SNAP cases can go down while need goes up, because people who lose SNAP often turn to food banks. Don't treat a drop in SNAP as a drop in need. Show which counties your tool might be missing, and say this on your "AI we used and how we checked it" slide.
- Weekly unemployment claims are the freshest signal here, about two weeks behind. In Harris County they about tripled the week after Hurricane Beryl.
- In your test, only use data that existed before the month you predict. Treat Census and Feeding America numbers as fixed background, not as monthly signals.
- If you make up pantry locations, trucks or stock, label them "made up" everywhere they appear.
- Small counties have small numbers, and a jump of a few hundred cases can look huge. Show how sure your tool is.
- The SNAP data is 44 monthly Excel files with different file names, and the dashboard months come in a different layout. Plan time to combine them.
- Scripts: the HHSC site blocks scripts that pretend to be a web browser. Plain Python or curl works.
- The Census API now needs a free key. Without one, use the download button on data.census.gov.
- Use only public data. No personal information.
Planning routes
Two datasets from the food help challenge may help: households without a car (Census table B08201) (opens in a new tab) and stores that take SNAP (USDA) (opens in a new tab).
If you use OpenStreetMap roads, credit "© OpenStreetMap contributors".
How it's judged
Same five criteria as every track. A strong demo shows your test on past months next to the "same as last month" guess, a range for each county (not just one number), and one county where your tool could be wrong, and why. The numbers come from the data, not from a chatbot. How judging works