Wildfire Black Box · San Diego County, California · 2003
Cedar Fire’s first night links rapid reported area growth, a changing clock and a fast-moving community emergency. Later records add thermal imagery and chaparral recovery research. This study follows those evidence layers while preserving the distinction between the whole incident, city damage and regional emergency operations.
Cedar Fire 2003: incident overview
CAL FIRE’s archive places the start on 25 October 2003 at 5:37 p.m. and lists full containment on 5 November at 6 a.m. Its jurisdiction field identifies Cleveland National Forest and the CDF San Diego Unit. These are retrieved archive fields, with a displayed update date of 23 September 2024. [1]
The City of San Diego’s historical account describes a signal fire lit by a lost hunter in Cleveland National Forest. The account provides the early area checkpoints used below. Its first checkpoint is labelled 5:30 p.m., so this record retains that reported checkpoint separately from CAL FIRE’s 5:37 start field. [2]
The analytical focus is how the incident crossed several boundaries: night into day, wildland into urban areas, offshore into onshore weather, and emergency reporting into retrospective science. Each transition changes the evidence needed to support a conclusion. A cause description establishes origin context; a dated area estimate describes extent; an image records a sensor’s view; a later ecological study measures a selected recovery outcome.
How many acres burned? Compare the source scopes first
| Source | Geographic or reporting scope | Selected figures | Evidence treatment |
|---|---|---|---|
| CAL FIRE archive [1] | Cedar incident | 273,246 acres | Retain the archive’s stated extent. |
| County after-action report [3] | Cedar-specific recap inside a multi-fire report | 273,246 acres; 2,820 structures destroyed; 15 fatalities | Use the Cedar row rather than the regional total. |
| City historical account [2] | Within San Diego city; described by the 28 October city containment statement | 28,676 acres; 335 structures destroyed; 71 damaged | Preserve the city boundary and reporting checkpoint. |
| Same city account [2] | Whole-county Cedar statement | 280,278 acres; 2,820 buildings; 15 deaths | Flag its area difference from the incident archive. |
| California State Parks [4] | Cedar summary with Cuyamaca Rancho context | 280,278 acres; 14 deaths including a firefighter; 2,232 homes | Preserve the stated area and unresolved fatality discrepancy. |
Our area check: 280,278 − 273,246 = 7,032 acres, about 2.6% of the CAL FIRE figure. The reviewed summaries do not identify a boundary version or calculation method that resolves that difference. We do not average the two figures or describe the gap as newly burned ground.
The county’s Cedar recap partitions the destroyed structures into 2,232 homes, 22 businesses and 566 outbuildings. Their sum is 2,820. It also separates 14 civilian fatalities and one firefighter fatality. These are source-defined categories; the building total is broader than the home count. [3]
Cal OES’s 2023 retrospective also reports 15 deaths. State Parks states 14 including a firefighter, and the NWS assessment’s Cedar summary also gives 14. Those values remain flagged against the explicitly partitioned county recap. We do not invent an explanation or treat a weather-services assessment as a comprehensive mortality audit. [4] [5] [6]
The city-scale and incident-scale containment descriptions have different footprints. An October city statement and a November whole-incident field need not mark the same milestone. Their coexistence is a scope question before it is a date conflict.
The first night: three reported area checkpoints
| Checkpoint | Reported area | Clock information retained |
|---|---|---|
| 25 October, 5:30 p.m. | 20 acres | The source supplies no explicit UTC offset for these checkpoint labels. |
| Midnight entering 26 October | 5,500 acres | The rollover date follows the source’s overnight narrative. |
| 26 October, 3 a.m. | 62,000 acres | The clock basis requires verification before an elapsed-time calculation. |
The same account describes the fire crossing into San Diego city at 6:30 a.m. on 26 October. These checkpoints document reported progression and community exposure. They are not a continuously surveyed perimeter history. [2]
For the midnight-to-3 a.m. pair, the reported area difference is 56,500 acres. Dividing by elapsed hours gives a mean change in reported area over that interval. A fire-front rate of spread instead requires distance between identified front positions and the corresponding elapsed time. Its units and spatial support are different.
Cal OES’s retrospective describes growth in acres per second; converting two acres per second gives 7,200 acres per hour. That dimensional conversion alone does not establish the summary’s averaging interval or show that such a rate held at every moment. Comparisons need the measurement definition and interval as well as a common unit. [6]
The overnight interval crosses the 2003 clock-change date
IANA’s historical rules for America/Los_Angeles place the autumn transition on the last Sunday of October in 2003. That date was 26 October: local civil time moved from daylight to standard time, repeating the 1 a.m. hour. [8]
The historical clock rule is established; the city narrative’s clock convention is not specified. We therefore calculate two conditional readings of the same midnight and 3 a.m. labels. They demonstrate sensitivity to the missing metadata. They are neither probabilities nor a reconstruction of which convention the original reporter used.
| Interpretation | Timestamp convention | Elapsed time | Mean reported-area change |
|---|---|---|---|
| A · Consistent fixed-offset clock | Illustrative UTC−07:00 at both endpoints | 3 hours | 18,833 acres/hour, rounded |
| B · Historical local civil clock | Midnight UTC−07:00; 3 a.m. UTC−08:00 | 4 hours | 14,125 acres/hour |
The four-hour interpretation yields a mean 25% below the three-hour interpretation. That difference comes entirely from the denominator. It is not an empirical uncertainty interval for the fire: uncertainty in reported area, rounding, observation time and perimeter methods has not been quantified here.
The accompanying code normalizes the local civil timestamps to UTC before subtraction. This makes the repeated hour explicit. It also retains the fixed-offset example independently, so the calculation can be reproduced without silently choosing a clock convention for the incident.
Why neither row is selected as the historical answer. Some original products explicitly label their times PDT or PST, while other narratives omit an offset. The thermal catalogue discussed below even retains a PDT label for its later 26 October acquisition window. Such labels must be preserved and investigated rather than automatically rewritten using civil-clock rules.
Weather changed the setting; warnings document a different timeline
The NWS service assessment, published in July 2004, describes prolonged dry conditions, very dry fuels and Santa Ana winds that peaked on 26 October. It then describes weakening offshore flow and a later onshore phase, with westerly winds driving eastward spread. This is a retrospective meteorological account of the regional event. [5]
For San Diego, the assessment records a Fire Weather Watch at 9 a.m. PDT on 23 October and an upgrade to a Red Flag Warning at 9 a.m. PDT on 24 October. These are documented forecast-product times. They establish advance weather messaging, rather than a household-level record of evacuation-warning receipt. [5]
The county after-action report focuses on Emergency Operations Center coordination during the Cedar, Paradise and Otay fires, with an activation window from 26 October to 1 November. Its scope statement specifically separates that review from fire and law-enforcement field-response analysis. [3]
To reconstruct an individual’s evacuation, an analyst would need a location-specific fire-arrival estimate, the relevant warning, evidence of receipt, a decision time and movement information. The weather-product chronology and an EOC review contribute context, but they do not supply that complete chain. This record therefore makes no evacuation-compliance estimate.
A thermal mosaic has a collection window
The Pacific Southwest Research Station FireMapper catalogue describes a Cedar thermal mosaic collected on 26 October between 11:34 and 14:40 PDT, at 8–12 micrometre wavelengths. It also provides repeat-pass sequences and notes approximate locations for its displayed geographic overlays. We reviewed the catalogue description, without reprocessing the underlying sensor data. [7]
Our duration check: the labelled acquisition window spans 186 minutes, or 3 hours 6 minutes, under the catalogue’s stated common clock basis. A mosaic assembled during that window can contain different locations observed at different times. Assigning every point the earliest timestamp would remove that temporal structure.
A front-displacement measurement needs the time of each relevant image frame or pass, the geometry of both front positions and a co-registration assessment. A whole-mosaic caption supplies context for collection; it does not guarantee that every location was imaged twice at the same interval.
Temperature-related imagery also has a different measurement target from a person’s radiant exposure. A reconstruction of exposure would need its own radiometric interpretation, geometry and atmospheric assumptions. We have not inferred heat flux, flame length or a building-ignition threshold from the displayed catalogue.
A 2019 airborne-imaging study is useful as a methods reference, but its Cedar data concern the 2016 Cedar Fire. The paper explicitly distinguishes that incident from Cedar 2003. Its numerical spread results are not imported into this record. [10]
Later knowledge: six-year chaparral recovery
Rachels and colleagues’ 2016 study specifically examines Cedar 2003. It compares selected chaparral sites representing Santa Ana wind-driven, Santa Ana backing and non-Santa Ana conditions, using pre-fire 2002 and post-fire 2009 imagery. The authors report lower shrub cover six years after the fire at the wind-driven study site than at the other two sites. [9]
The paper interprets recovery in relation to burn conditions, terrain and moisture availability. That is a selected-site landscape comparison, not an experiment isolating wind as the sole causal variable. Its interpolated MODIS progression surface is explicitly described as unsuitable for detailed fire-dynamics reconstruction. The study uses it for broader site selection. [9]
The ecological result adds a later evidence layer to the incident. It does not retroactively become a forecast available in October 2003, and it does not establish the present condition of every part of Cuyamaca Rancho or the Cedar footprint. A current recovery assessment would require current observations and a defined ecological outcome.
Known At, Fire DNA and reproducibility
| Layer | What it contributes | Time and scope retained |
|---|---|---|
| Incident and impact reporting | Reported dates, areas and damage categories | City, incident and multi-fire report boundaries stay separate. |
| Weather assessment | Forecast-product chronology and wind evolution | July 2004 retrospective with individually labelled forecast times. |
| Thermal catalogue | Sensor, acquisition-window and pass metadata | 26 October 2003 collection labels; original clock convention retained. |
| Recovery research | Selected chaparral sites and later shrub-cover comparison | 2002/2009 imagery, analysed in a 2016 publication. |
| Tecnobosque calculation | Area subtraction, clock sensitivity and duration checks | Original arithmetic prepared 13 October 2026; conditional assumptions visible. |
The Fire DNA supported here combines dry-fuel and wind context, rapid reported overnight growth, community exposure and a multi-stage evidence record. It contains no universal severity score. Forensic reconstruction remains bounded by the spatial and temporal resolution of each source.
Work performed: primary-source retrieval; visual verification of the county’s Cedar recap and NWS report title; review of selected weather and research sections; preservation of source-specific totals; original timestamp and arithmetic checks; and one original two-panel scientific graphic. The supplied CSVs and code reproduce those calculations.
Work still required for a full reconstruction: dated perimeter versions, original overnight clock metadata, frame-level acquisition records, native sensor data, address-level exposure and resident warning/movement evidence. We did not independently remap the incident, audit every page of the agency reports or reanalyse the ecological imagery.
Primary sources and review scope
- CAL FIRE Cedar archive. Current retrieved incident fields; displayed update 23 September 2024.
- City of San Diego Fire-Rescue: Cedar 2003. Early checkpoints, city outcomes and whole-county statement; overnight checkpoint offsets unspecified.
- San Diego County Firestorms 2003 after-action report. Scope, selected chronology and Cedar recap; printed page 8 recap visually verified; not a full-report audit.
- California State Parks: Cedar and Poomacha. Cedar section only; area and fatality discrepancies retained.
- NWS service assessment: Southern California Wildfires. Published July 2004; selected weather narrative and San Diego product chronology; title page visually verified.
- Cal OES Cedar retrospective. Published 3 November 2023; attributed impact summary and area-rate wording.
- Pacific Southwest Research Station FireMapper project catalogue. Original collection description and metadata reviewed; underlying imagery not independently processed.
- IANA time-zone database: North America. Historical U.S. rule for 1967–2006 and Los Angeles zone entry; retrieved 13 October 2026.
- Rachels et al. (2016): chaparral recovery after Cedar 2003. Abstract and selected site-selection/method sections reviewed; source imagery not acquired.
- Stow et al. (2019): uncertainty in airborne spread estimates. Methods and incident identity checked; Cedar 2016 numerical results excluded from this 2003 record.