Buffer, Travel Matrix, or Isochrones? Drive-time analysis in Snowflake.

by Vladi ‐ 2 min read

The same 15-minute catchment question answered three ways in Snowflake SQL.

I ran it three ways on the same Snowflake data: what percentage of buildings in Orange County, California are beyond a 15-minute drive of any Whole Foods Market?

The straight-line buffer

Assume a drive speed, divide by a circuity factor, draw a circle. At 48 km/h and a factor of 1.3 that is 615 m/min, so 15 minutes is a 9.2 km radius, then ST_DISTANCE to the nearest store.

Result: 21.8% outside.

Straight-line buffer. Green circles are the 9.2 km buffers; hexes are buildings outside them, colored by drive time and sized by building count.

View map and data

The precomputed H3 travel matrix

Real routing, precomputed into an origin-destination matrix on H3 cells. I used a California matrix of 2.2 billion rows at resolution 7.

Result: around 48% outside, range 44–57%.

The range is the catch, and it applies to any precomputed matrix. The table is a cross product of 34,929 origin cells and 65,128 destination cells, and neither set covers every cell. In Orange County only 5 of the 12 store cells were origins, so most buildings got measured against 5 stores instead of 12. Check cell coverage on both sides of the join before trusting a matrix.

H3 travel matrix. The green reach is dissolved from H3 cells, so it follows the road network instead of a radius.

View map and data

→ Snowflake on precomputing H3 travel-time matrices

TravelTime isochrones

TravelTime publishes a native app on Snowflake Marketplace that calls their isochrone API from SQL, so you get true drive-time polygons without leaving the warehouse:

SELECT TO_GEOGRAPHY(t.GEOMETRY) AS g
FROM wf w,
     TABLE(TRAVELTIME.V2.TIME_MAP_FAST(
       w.lat, w.lon, 900, 'driving', 'many_to_one', 'weekday_morning', TRUE)) t

Twelve calls dissolved with ST_UNION_AGG into one reach polygon of 1,220 km². Inside and outside are decided per building against it, so the boundary is exact rather than snapped to a grid.

Result: 34.5% outside.

TravelTime isochrones. Genuinely network shaped: freeway fingers, interior holes, and a detached lobe around the Los Angeles store.

View map and data

→ TravelTime on Snowflake Marketplace

→ TravelTime isochrone API docs

How the three compare

Method	Buildings beyond 15 min	Share
Straight-line buffer	181,535	21.8%
H3 travel matrix (precomputed routing)	~398,000	~48%
TravelTime isochrones	286,678	34.5%

All of it runs on Overture Maps via Snowflake Marketplace: 831,475 buildings in Orange County and 12 stores.

Run it yourself

All three maps were built by Claude with the GeoSQL skill and rendered in Dekart, both open source and running against your own warehouse.

pip install geosql && geosql

github.com/dekart-xyz/geosql