Forty-Five Days to an Authoritative E911 Address Layer
The GIS coordinator for a city of 182,000 residents has 45 days to deliver one authoritative address point layer to a new computer-aided dispatch (CAD) system. Three candidate sources exist. The county E911 file holds 71,400 points last field-verified in 2016, with a stated positional accuracy of plus or minus 5 m. The utility billing master holds 69,800 address strings with no coordinates, and about 9 percent of its records carry unit or building identifiers. Parcel centroids from the assessor roll cover 74,100 polygons and sit within about 0.3 m of the parcel fabric. A trial match found 4,300 duplicates across sources, 12 percent of records lacking ZIP+4, and 1,900 billing addresses with no parcel because they fall on unplatted land. Dispatch requires every structure to route to a driveway or building entrance, not to a parcel center. The vendor's conflation tool reports a 92 percent automatic match rate and offers to push the result straight to production.
1. The vendor proposes accepting the 92 percent automatic match and pushing it to production. Evaluate that proposal and define source-precedence rules for geometry versus attributes.
Read the worked answer
Accepting 92 percent would put roughly 5,900 unmatched records into production along with every duplicate and every bad ZIP+4 inside the matched set. The failures cluster in exactly the records dispatch cares about most: new construction, unplatted rural addresses, and multi-unit buildings.
Recency is not accuracy. The billing file updates monthly but is self-reported and never field-verified. The E911 file is stale but was survey-checked. Parcel centroids are geometrically excellent but semantically poor, because a parcel is not a mailing address.
Split authority by what each source actually knows:
- Geometry: parcel fabric where the parcel is platted and the structure sits close to the centroid; structure centroid from orthoimagery or field GPS where it does not.
- Address attributes (street name, number, ZIP+4): E911, reconciled against USPS address validation.
- Discovery: billing is the only source reaching the 1,900 unplatted structures, so use it to generate candidates, never to confirm them.
Publish the vendor's matched set only at a 'provisional' confidence tier. Anything conflicting or unplatted goes to a field-verification queue with a named owner and a sign-off date.
2. Dispatch's entrance requirement conflicts with parcel-centroid geometry. Describe how to change the geometry model and how to quantify and report residual error.
Read the worked answer
A parcel centroid can be 40 m from the front door on a large lot and worse on flag lots, sending units to the wrong side of a fence line. Move to an access point: derive structure footprints from leaf-off orthoimagery (or the city building layer if one exists), then place the routing point at the driveway-road junction or main entrance, preserving sub-address identifiers for apartments and suites. Field-verify the high-risk classes first: rural, multi-unit, new construction, and anything on a private road.
Quantify error rather than asserting it. Draw a random stratified sample of at least 60 to 100 points checked against survey-grade or sub-meter GPS, then report RMSE and 95 percent circular error separately for each point class (field-verified, imagery-derived, centroid-derived). A single blanket accuracy figure hides the class that will fail.
Also confirm the datum and realization behind the E911 plus-or-minus 5 m claim. A legacy HARN versus NAD83(2011) difference can exceed a metre, and the vendor's stated accuracy may not apply to your control network at all.
3. What should happen to the 1,900 unplatted addresses, and what metadata and lineage fields must ship with the layer?
Read the worked answer
Treat them as the highest-risk tier, not as records to drop. Unplatted land usually means long private driveways, shared access easements, and street names that are not in the MSAG, which is precisely where a wrong point costs minutes.
Create provisional points from current imagery, attach the billing address string, and route each one to the county 911 addressing authority for official assignment, recording the request date. Do not silently delete them; a silent delete destroys the audit trail and guarantees the same duplicates reappear at the next refresh.
Minimum fields to ship: source (E911, billing, parcel, field), derivation method, capture date, horizontal accuracy estimate, verification status (verified, provisional, unverified), verified-by, last-edit date, and a persistent unique ID. Add FGDC/CSS-style metadata covering lineage, the conflation rules applied, known exclusions, and the refresh cycle. Push rejected matches into a review table rather than discarding them so the next analyst can see what was decided and why.