Wildfire Black Box · California · Historical case
Oakland–Berkeley in 1991 shows how a vegetation fire can become a community fire. Wind-driven spotting, burning buildings and disrupted water delivery interacted across the hills. The record becomes more useful when incident reports, early reconnaissance and later fire models retain their separate roles.
Oakland Hills fire 1991: what happened?
The US Fire Administration’s East Bay Hills Fire report describes a brush fire on 19 October that rekindled on 20 October 1991 under strong Diablo winds. Its summary leaves the cause of the original fire undetermined. The rekindle mechanism and the initial ignition cause are therefore separate claims. [1]
Known as the Tunnel Fire and the Oakland–Berkeley Hills fire, the incident overwhelmed initial suppression and developed across a populated urban interface. The USFA summary reports 25 deaths, 150 injuries, 3,354 structures destroyed and approximately 1,500 acres. These are that report’s categories and estimates, not a harmonized dataset assembled from every later account. [1]
| Source | Reported quantity | Interpretation boundary |
|---|---|---|
| USFA technical report, summary | 3,354 structures destroyed | Source-defined structure total |
| NISTIR 4724, November 1991 | 2,449 single-family residences | Residential subcategory |
| NISTIR 4724 | 437 apartments and condominiums; 3 rental units | Additional residential categories |
| NISTIR 4724 | 2,889 residences destroyed | Not the same label as all structures |
NIST’s preliminary report lists those residential figures. [2] The category sum is 2,449 + 437 + 3 = 2,889. Subtracting this from the USFA structure total gives 465, but that difference does not prove that exactly 465 nonresidential buildings were destroyed. The sources differ in category and reporting context. Without a reconciled inventory, the subtraction is arithmetic rather than a validated property count.
For comparisons with another disaster, select an explicitly matching category first. A housing-unit count, a building count and a structure count should not occupy the same chart column without a documented conversion.
Why burning houses matter to fire spread
A 1997 study by Woycheese, Pagni and Liepmann examined firebrand lofting from large fires, specifically including burning structures. Its NIST-hosted abstract identifies brands deposited hundreds of meters ahead of the front as a major Oakland spread mechanism and models a burning house using a buoyant-plume representation. [3]
This means the analytical boundary cannot end at the vegetation line. Once buildings burn, their contribution must be considered when reconstructing subsequent exposure. It does not mean every downwind house received the same brand load or that every ignition can be assigned to one roof.
NIST’s November 1991 reconnaissance emphasized high winds, vegetation near buildings and combustible exterior materials. It did not find wood framing itself to be a significant driver of spread in the investigated multi-family construction context. [2]
Tecnobosque interpretation: distinguish a building’s load-bearing frame from the exterior components exposed first. A conclusion about framing cannot be turned into a claim that the entire building envelope was resistant. Likewise, evidence that brands traveled hundreds of meters is not a tested maximum travel distance or a guaranteed protective setback.
A useful reconstruction asks four address-specific questions: what was burning nearby, what exposure reached the building, which component ignited first, and what defensive action occurred? A regional loss total cannot answer those questions for a particular property.
Water delivery depended on power, pipes and compatible equipment
The 27 February 1992 OES multi-agency review reports nine pressure zones affected by the fire. It describes power failures, reservoirs losing water through ruptured household service lines, and pumping stations without auxiliary power. It also identifies a compatibility problem between Oakland’s three-inch hydrant outlets and the two-and-a-half-inch connections commonly used by responding equipment. [4]
These are different constraints. An adapter helps connect an engine; it cannot refill a depleted reservoir. Backup pumping can support refill; it cannot by itself eliminate water losses through damaged connections. Adequate stored volume is also different from adequate pressure and flow at a particular hydrant.
The review describes a fire-flow reserve based on 1,500 US gallons per minute for two hours. [4] The following calculation illustrates the volume represented by that stated design basis. It does not recreate a specific Oakland reservoir’s performance.
| Assumption | Calculation | Result |
|---|---|---|
| 1,500 US gpm for 120 minutes | 1,500 × 120 | 180,000 US gallons |
| Same usable volume, 3,000 US gpm demand | 180,000 / 3,000 | 60 minutes |
| Same usable volume, 4,500 US gpm demand | 180,000 / 4,500 | 40 minutes |
These idealized durations assume no refill, constant demand and fully usable storage. They exclude changing pressure, hydraulic losses, inaccessible volume and damaged connections. The example explains why a reserve designed for one demand cannot automatically support several simultaneous demands for the same duration.
What the later fire-induced wind simulation tested
Trelles and Pagni modeled the incident with the Baum–McCaffrey mass-fire approach. Their study represented 38 fires at 11:45 a.m. and 259 at noon, assigning each modeled house a heat-release rate of 50 MW. It produced a maximum horizontal induced velocity of approximately 13 m/s for the noon case and examined its relationship with the changing spread pattern. These are model inputs and outputs, not direct measurements at every burning house. [5]
The assigned source-power totals are 38 × 50 = 1,900 MW and 259 × 50 = 12,950 MW, or 1.90 GW and 12.95 GW. Their ratio is about 6.82. That ratio belongs to the simplified source inventory; it is not a measured amplification of local wind speed or a fireline-intensity value.
Nor should the modeled 13 m/s induced velocity simply be added to a weather-station wind reading. Direction, location, elevation and the model’s flow-field definition matter. A scalar sum would erase those distinctions.
NIST’s reconnaissance also cautioned that airport weather could differ from winds in the hills because of terrain and elevation. [2] For reconstruction, keep the station’s actual location and observation time with its data. A nearby instrument is contextual evidence until its representativeness for the specific site is established.
Known At: how the evidence developed
| Stage | Evidence object | What it can support |
|---|---|---|
| 19–20 October 1991 | Incident events later described by agency reports | Sequence of brush fire, rekindle and major spread; historical reconstruction |
| November 1991 | NISTIR 4724 preliminary reconnaissance | Early site observations and construction-focused conclusions |
| 27 February 1992 | OES multi-agency operational review | Response-system issues, water dependencies and recommendations |
| Later research | Fire-induced wind and brand-lofting studies | Mechanism analysis under stated model assumptions |
The NIST PDF’s title page and bibliographic sheet identify November 1991, although its website landing page displays a January date. This record uses the document’s own date. [2] A document about an October fire must not be treated as incident-time knowledge merely because a catalogue has an earlier placeholder date.
The table tracks the development of the public evidence. It does not establish which warnings particular residents received, nor the exact information available to each command position. Those questions require timestamped communications and recipient-specific evidence.
Fire DNA: the case’s analytical signature
- Ignition history: distinguish the original cause from the subsequent rekindle.
- Spread mechanism: examine wind-driven spotting and additional sources created by burning buildings.
- Infrastructure: treat power, stored water, damaged service lines and equipment compatibility as separate dependencies.
- Loss accounting: preserve source-defined residential and structure categories.
- Model evidence: keep assumed heat release and simulated flow separate from observed weather.
- Remaining gap: these reports do not provide a reconciled, causal explanation of every property loss.
Oakland is useful for asking how a community fire changes the problem faced by responders and buildings. It does not demonstrate that any single intervention would have prevented the disaster. Counterfactual claims require an explicit method and evidence beyond listing the problems identified afterward.
Primary sources and citation
Record TB-WBB-US-1991-OAKLAND · Prepared 9 October 2026. Agency findings, published models and Tecnobosque’s worked arithmetic are labeled separately.
- US Fire Administration — The East Bay Hills Fire, USFA-TR-060. Overview and key-issues summary; cover identifies the October 1991 incident.
- Steckler, Evans and Snell — NISTIR 4724. November 1991; preliminary study of the Oakland Hills fire and multi-family wood-frame construction. Document date takes precedence over catalogue placeholder.
- Woycheese, Pagni and Liepmann — Brand Lofting Above Large-Scale Fires. 1997; this record uses the NIST-hosted abstract for its mechanism summary.
- California OES — The East Bay Hills Fire: A Multi-Agency Review. 27 February 1992; water availability and equipment compatibility, printed pages 27–29.
- Trelles and Pagni — Fire-induced Winds in the 20 October 1991 Oakland Hills Fire. Fifth International Symposium on Fire Safety Science, pages 911–922; modeled sources and wind-field interpretation.