Disaster response · Spatial analysis · After-action
Camp Fire
Where the Warning Chain Broke
A geospatial after-action analysis of the 2018 Camp Fire that reconstructs the minute-level timeline, locates the failure at each link of the warning-to-action chain, and converts the causes into recommendations.
01Operational question
The Camp Fire ignited at 06:29 on 8 November 2018 near Pulga in Butte County from failed transmission equipment during a red flag warning. It burned 153,336 acres over 17 days, destroyed about 18,800 structures, killed 85 people, and forced 52,000 to evacuate. The question for the after-action analysis: where did the path from recognizing danger to helping people act break down, and what should change before the next fire?
02Why the problem was difficult
- Compressed decision space. Ninety-three minutes from ignition to a town-wide order, and Skyway gridlocked 58 minutes later; every response function operated inside that interval.
- Compounding drivers. Cured fine fuels, Jarbo Gap winds of 50–70 mph carrying embers a mile ahead of the front, a ridge-top wildland-urban interface community with narrow winding roads, dead-end streets, and one principal evacuation route.
- Warning is a chain. An issued order is one link; residents also had to receive it, understand it, and have a viable way to act, and a different failure affected each link.
- Evidence is scattered. Reconstructing a minute-level timeline required reconciling official reports, investigations, peer-reviewed research, agency sources, and more than 30 published accounts.
- Systems failed as services. The data was good and the delivery was fragile: cellular and fiber infrastructure burned early, and the emergency operations center lost phone and internet.
03My role
I authored the 20-page geospatial after-action analysis: timeline reconstruction, plan-versus-execution comparison, the two original maps, the technology contribution-and-limitation assessment, and the BLUF recommendations.
04Data architecture
The evidence base has four tiers: official records (the Butte County after-action report and corrective action plan; the Cal Fire incident information and investigation report), peer-reviewed research on no-notice evacuation timing, fast-moving evacuation simulation, and displacement, agency material, and journalism. Two original maps carry the spatial argument.
| Map | Platform | Content | Analytical purpose |
|---|---|---|---|
| Study area map | ArcGIS Online web map | Butte County boundary, burn area and perimeter, county wildland fire protection facilities by type | Places the burn area in relation to response infrastructure and the communities of Paradise, Concow, Magalia, and Pulga |
| Multivariate map | ArcGIS Pro layout at 1:783,811 | Share of households without a vehicle by census unit, draped on hillshade terrain, with the fire perimeter | Relates evacuation dependence to terrain and fire extent; shows where carless households sat inside or beside the perimeter |
Vehicle access is a direct measure of the ability to self-evacuate, terrain governs both fire spread and road geometry, and the perimeter marks exposure. Overlaying the three turns a general statement about vulnerable residents into a located one.
05Methodology
Structured after-action analysis: reconstruct what happened, compare it with what the plans assumed, isolate the causes of the gap, and convert the causes into recommendations. Five steps: establish the environmental and meteorological setting; build a timestamped timeline from ignition through containment; evaluate each response function, notification, evacuation, emergency operations center coordination, mutual aid, and public information; assess the geospatial technologies used; deliver recommendations.
Evacuation orders went to Pulga and Concow at 07:13, to eastern Paradise at 07:46, and to all of Paradise at 08:02, while spread reached about 80 acres per minute at peak.
Table view
06Validation
- Plan versus execution. Each response function was compared with the assumption the plan made about it: phased zone-based evacuation, opt-in notification, single-route capacity.
- Evidence triangulation. Timeline entries were reconciled across official records, research, agency sources, and journalism before use.
- Located vulnerability. The multivariate map tested the general claim about immobile residents against census units inside and beside the perimeter.
- Technology assessed on contribution and limitation. GIS common operational picture tools, MODIS and VIIRS thermal detection, commercial imagery, and UAS were each judged on what they contributed and what limited them; connectivity, update latency, and user training were the limiting factors.
07Output
- Twenty-page geospatial after-action report with minute-level timeline, warning-chain findings, two original maps, and technology assessment.
- BLUF recommendations to public officials: layered warning on a primary, alternate, contingency, and emergency structure with opt-out Wireless Emergency Alerts as primary; interagency geospatial integration with data-sharing agreements signed before the incident; community preparedness.
08Result
- An opt-in alert system with 30 to 40 percent enrollment rode on cellular and fiber infrastructure that burned early; about 7,000 of 52,000 evacuees received a CodeRED alert, and integration faults defeated simultaneous alerts.
- Zone-based wording without place names broke the understanding link for residents who did not know their zone.
- One principal route gridlocked within an hour of the town-wide order; households without a vehicle sat inside and beside the perimeter.
- Geospatial systems failed as services when connectivity failed, which sets the design position: architect for degraded communications, offline operation, and agreements in place before the incident.
09Tradeoffs and limitations
- Retrospective evidence. The timeline rests on published records; some intervals remain contested across sources.
- Census resolution. Vehicle access is aggregated by census unit, which smooths within-unit variation.
- Scope. The analysis evaluates warning and evacuation; it does not derive burn severity from imagery, which would extend it.