Esperanza Fire 2006: Thermal Imaging, Fire Models and Hindsight

Esperanza Fire 2006: incident overview

The fire began near Cabazon in Riverside County, California, on 26 October 2006. CAL FIRE’s archived page records a 1:12 a.m. start, 40,200 acres, 54 structures destroyed and an arson cause. Its displayed containment date is 1 November 2006 at 7:00 a.m.. [1]

The joint accident investigation’s executive summary records the first report at 1:11 a.m. Pacific Daylight Time and an Engine 57 burnover at approximately 7:15 a.m.. Five Forest Service firefighters suffered fatal burns while conducting structure protection near an isolated residence in Twin Pines. This record uses the agency report for that classification. [2]

The CAL FIRE webpage’s current template labels its five fatalities as civilian; that conflicts with the accident record’s identification of the firefighters. A template field should not silently override the investigated casualty category. The original page remains cited so readers can inspect the discrepancy.

The USDA Inspector General’s 30 November 2009 report gives the residential and outbuilding breakdown. [3] The sum 34 + 20 = 54 agrees numerically with the incident page, but 54 should not be described as 54 homes. Coen and Riggan’s later paper uses 30 October as the containment date. [4] No contemporaneous containment log has been reconciled in this review, so both dates stay attached to their sources.

What the airborne thermal imagery actually observed

Coen and Riggan (2014) analyzed the fire using FireMapper airborne thermal imagery and the CAWFE coupled atmosphere–fire model. Their six orthorectified maps cover observations from 11:17 a.m. to 9:42 p.m. PDT on 26 October. [4]

That window begins after the approximately 7:15 a.m. burnover described in the accident summary. Using that approximate incident time, the interval is about four hours: 4 hours 2 minutes by subtraction. The afternoon and evening maps are valuable evidence for the later fire, but they are not a direct thermal recording of the crew’s exposure at 7:15.

Tecnobosque interpretation: every map needs two clocks: when the sensor observed the fire and when an analyst later produced or interpreted the map. A later reconstruction of an earlier location is a derived result. It should not inherit the status of an observation simply because the analysis also uses measured imagery.

The same principle applies to a dramatic image without a timestamp. Before using it to support a sequence, establish its acquisition time, location, processing method and coverage. If those are missing, preserve the gap rather than assigning the image to the most consequential moment.

A model can reproduce important features and still miss others

The CAWFE simulation reproduced features including the fire’s broad direction and width, canyon-related spread and two heading regions. It also placed the fire at the early-morning fatality location. However, the authors report a mismatch: downhill head-fire spread under weak winds was slower in the simulation than observed. [4]

That combination matters. Agreement on overall shape is useful evidence, but it is not complete validation of every local arrival time, heat exposure or tactical counterfactual. A model’s successful features and unsuccessful features should be reported together.

A reader evaluating the reconstruction can ask:

• Which outputs were compared with actual observations, and at what times?
• Does the evaluation concern position, direction, width, speed or exposure?
• Was the comparison made at the crew’s location or across a broader map?
• Which discrepancies remain, and how could they affect the specific conclusion?
• Was the model run retrospectively with information unavailable during the incident?

These are review questions, not new findings about the crew. A reconstructed fire location does not by itself establish the warning a firefighter received, the choices available or whether a different decision would have changed the outcome.

Why a map-to-map average can hide a short fire run

The following example is invented for teaching. It does not use measured Esperanza distances, and it does not estimate the burnover’s speed. Assume the same identifiable section of a front is tracked along a defined direction for 30 minutes:

The overall rate is 40 m/min, while the final interval averages 90 m/min: 2.25 times the full-window average. The final interval’s rate is not an instantaneous maximum; shorter bursts could still be unresolved.

Adding the two rates and dividing by two gives 52.5 m/min, which is wrong for this full-window calculation because the intervals have different durations. The time-weighted calculation is (15 × 20 + 90 × 10) / 30 = 40 m/min.

There is also a geometry condition. The calculation assumes comparable positions of the same front along the stated direction. The most distant new spot fire is a different measurement from the continuous front’s advance. A perimeter difference that includes spotting, flank expansion and merging fronts should not automatically become a local head-fire speed.

Known At: preserve the order of evidence

The Green Sheet documents the five firefighter fatalities and identifies lookouts, communication, escape routes, safety zones and structure-protection triage among its review topics. [5] Its cover gives the incident date; that is not proof of the document’s publication date.

OIG identifies the SAIT factual report as dated 22 May 2007 and states that its own investigation found no criminal wrongdoing by Forest Service or CAL FIRE personnel. [3] That finding concerns those personnel in that investigation. It does not negate the arson cause or establish that every response-system question was resolved.

A later scientific paper adds another evidence object; it does not replace the accident record. Keep each source’s purpose and scope with the claim it supports.

Fire DNA: what makes this record useful

• Setting: Santa Ana conditions in mountainous Southern California.
• Human impact: five firefighter fatalities documented by the accident investigation.
• Observation issue: the thermal-map window begins after the burnover.
• Reconstruction issue: broad model agreement coexists with a reported spread mismatch.
• Data-quality issue: template casualty labels and containment dates require source checks.
• Teaching value: averages, local rates and spot-fire positions are different quantities.

The case supports careful comparison of observed and simulated fire progression. It does not support assigning blame from a single model output or projecting a retrospective map into what every responder knew at the time.

Continue with the historical fire archive or the engineering tools directory. Keep model assumptions visible when using a calculator to explore a historical case.

Sources and citation

TB-WBB-US-2006-ESPERANZA · Prepared 9 October 2026. Worked distances and rates are hypothetical; no original field study is claimed.

1. CAL FIRE — Esperanza Fire archived incident page. Start, acreage, structures and cause; template-label discrepancy and containment date retained explicitly.
2. Esperanza Accident Investigation Factual Report — Executive Summary. Primary joint-agency text reproduced by Colorado Firecamp; host is not the issuing authority.
3. USDA Office of Inspector General — Esperanza Fire Fatality Investigation. 30 November 2009; primary agency report preserved by the Wildland Fire Lessons Learned Center.
4. Coen, J. L. and Riggan, P. J. — Simulation and thermal imaging of the 2006 Esperanza Wildfire. International Journal of Wildland Fire 23, 755–770; online 9 July 2014, DOI 10.1071/WF12194.
5. CDF Green Sheet — Esperanza Fire firefighter burnover. Primary preliminary summary preserved by the Municipality of Anchorage; incident date on cover.