Wildfire Black Box · Oregon · Source record and raster audit
A burn map and a river graph answer different questions. Holiday Farm connects a soil-severity assessment with a monitoring network disrupted by the fire. This record adds a reproducible count of the official raster’s classes, then explains why neither a map category nor a missing live reading is an observed flood measurement.
What happened in the Holiday Farm Fire?
The Forest Service’s 30 October 2020 BAER summary describes an approximately 7:45 p.m. start on 7 September 2020, west of McKenzie Bridge, during strong easterly winds. Its incident-area figure is 173,439 acres as of 12 October 2020. The assessment examines post-fire risks to values on Forest Service lands. [1]
BAER means Burned Area Emergency Response. The report’s purpose is emergency assessment after the fire, rather than determining every original ignition or documenting every subsequent storm. This review makes no ignition-cause finding or present-day closure claim.
What does the soil burn severity map represent?
The BAER imagery program describes satellite-derived BARC as an input used to guide field assessment and development of a final soil burn severity product. Satellite change classification and the resulting field-adjusted map occupy different stages of that process. [4]
The downloaded Holiday Farm package identifies Sentinel-2 imagery dated 4 September and 29 September 2020, field validation by the BAER team, and metadata publication on 13 November 2020. The embedded grid is NAD83 / UTM zone 10N, EPSG:26910, with 20 × 20 metre cells. [3]
Keep the object being measured explicit. A categorical soil-severity product is not a thermal camera measurement of fire temperature, a map of home losses or a direct record of how fast flames moved. Its classes describe mapped effects under the product’s method.
The catalogue labels the download final SBS, while the supplied metadata retain preliminary/draft limitations. This audit preserves that wording and identifies the exact files reviewed; it does not confer a new validation status on the map.
Original calculation: how much mapped area lies in each class?
Tecnobosque raster audit, performed 12 October 2026. We counted the native cells in HolidayFarm_SBS_final.tif, verified the class colors against the accompanying legend, and selected class codes 1–4. The table uses the complete downloaded classified footprint, without an ownership mask.
| Soil severity class | Cells | Area, hectares | Share of classified cells |
|---|---|---|---|
| Unburned to very low | 75,666 | 3,026.64 | 4.3% |
| Low | 427,277 | 17,091.08 | 24.4% |
| Moderate | 1,084,897 | 43,395.88 | 62.0% |
| High | 162,930 | 6,517.20 | 9.3% |
| Total classified | 1,750,770 | 70,030.80 | 100.0% |
Each native cell represents approximately 400 square metres in the projected grid: 20 × 20. Dividing by 10,000 gives 0.04 hectares per cell. The moderate and high classes together contain 1,247,827 cells, or 71.3% of the classified cells.
These are new calculations from an existing agency product, not figures quoted from an agency summary. Areas use complete grid cells and no terrain correction. The displayed decimals describe the arithmetic; they do not establish field accuracy to a hundredth of a hectare.
This calculation also does not replace the official incident acreage. The selected raster footprint, a mapped perimeter and a dated incident-area estimate can have different boundaries and inclusion rules. Reconciliation would require those objects and their versions to be compared explicitly.
Why background cells change the apparent percentage
The rectangular raster contains 3,286,960 cells. Of those, 1,536,190 carry code 127, outside the four severity classes. We exclude those cells rather than interpreting them as unburned ground.
There is a practical trap in this file: the reader exposes no declared NoData value, and its automatic valid-data mask includes the whole rectangle. Selecting the four verified classes explicitly is therefore essential for this calculation. A software mask should not override the dataset’s legend without review.
NoData or background is not another severity class. Including it changes the population being described. Conversely, removing genuine unburned-to-very-low cells would also change the denominator: those cells belong to this classified product and remain in the calculation.
The package includes the unmodified source ZIP, file hashes, an audit script and numerical results. Pixel counts were checked independently using two readers. Neither check measures the accuracy of the original classification on the ground.
Why the BAER report’s 63% is not this audit’s denominator
The October BAER summary reports that 63% of mapped areas on Forest Service-managed lands had moderate or high soil burn severity. The management-domain qualifier is part of the finding. [1]
Our 71.3% concerns the complete classified raster we downloaded. It is not an ownership-filtered result, and its metadata identify a later product version. This audit therefore neither reproduces nor disproves the 63% statement. It supplies a different, explicitly defined calculation.
To test the earlier statement, obtain the ownership boundary and mapped product used for that report, establish how boundary cells were counted, and rerun the calculation for that domain. Do not infer the reason for a difference from the two percentages alone.
Equally, combining moderate and high classes does not make every included cell high severity. A summary must retain the grouping used in its numerator. The table’s high-only share is 9.3%, a separate quantity.
What happened to the McKenzie River monitoring record?
The USGS account dated 23 September 2020 describes damage to power and communications equipment and interruptions to public data transfer, including at sites outside the fire footprint. Its McKenzie River near Vida hydrograph identifies disrupted discharge monitoring from 8–16 September. [2]
The same account describes turbidity data stored in a sonde and retrieved on 17 September. Local measurement and public transmission therefore need to remain separate evidence states. Later retrieval can make an earlier observation available without making it publicly available at the time it was measured. [2]
| Object | What it records | What it does not establish |
|---|---|---|
| Interrupted public data feed | Missing or delayed availability in that channel | Zero river flow or zero environmental change |
| Stored observation retrieved later | Measurement time and a later recovery time | Contemporaneous public access to that measurement |
| Post-fire model output | A result under specified inputs and assumptions | A measured outcome of a particular storm |
A gap should retain its reason when known. Replacing missing discharge with zero would assert a physical condition that the communication failure does not establish. Filling the interval by interpolation introduces an analytical assumption and must be identified as such.
Are the predicted streamflow increases observed flood results?
The BAER summary reports modeled post-fire peak-streamflow responses of 1.2–3.2 times pre-fire levels, depending on the analyzed drainage’s moderate/high severity proportion. It also states that watershed response depends on storm events. [1]
That range is a prediction within the assessment, not a measurement extracted from the disrupted Vida graph. Validating a prediction needs a matching storm, drainage, flow definition and appropriate baseline. A ratio from one catchment cannot automatically be assigned to another monitoring site.
A useful comparison would retain the model’s inputs and version, the observation’s timestamp and quality flags, and any recovery of locally stored records. Missing observations remain a limitation of the comparison until their status is resolved; they should not be converted into evidence that the model was right or wrong.
Known At: measured, transmitted, assessed and published
| Checkpoint | Evidence object | Boundary |
|---|---|---|
| 7 September 2020 | Start described in the BAER report | Retrospective incident chronology |
| 8–16 September | Vida monitoring interruption | Availability gap, not zero discharge |
| 17 September | Stored sonde data retrieved | Retrieval time differs from measurement time |
| 23 September | USGS public account | Contemporary network-restoration evidence |
| 29 September | Post-fire satellite image | Acquisition date in the map metadata |
| 30 October | BAER summary | Post-fire assessment and model outputs |
| 13 November | Map metadata publication | Not the satellite acquisition date |
| 12 October 2026 | Tecnobosque raster audit | New calculation of an existing historical product |
Fire DNA and the limits of this record
- Mapping: a field-adjusted soil product with explicit classes and dates.
- Arithmetic: native cell counts, a defined mask and projected area.
- Monitoring: measurement, storage, transmission and retrieval are distinct.
- Prediction: modeled responses require observations appropriate to their scope.
- Remaining gaps: no ownership-subset reconstruction or matched storm-validation study.
No severity ranking or TB-ECI score is assigned. This analysis does not estimate current flood risk, erosion tonnage, drinking-water safety or household loss. No new field survey or class-accuracy validation was performed.
Primary sources and reproducibility
- Willamette National Forest: Holiday Farm Fire BAER Summary. USDA Forest Service, 30 October 2020. Original agency URL; the full primary document was examined through a publicly preserved PDF copy. The mirror is not the issuing authority.
- USGS: USGS Responds to Oregon Wildfires. Oregon Water Science Center, 23 September 2020. Monitoring disruption, Vida discharge graph and locally stored turbidity observations.
- Holiday Farm official BAER product record and its soil burn severity package. USDA Forest Service GTAC dataset distributed through the USGS portal; supplied metadata, raster, auxiliary palette and legend reviewed.
- BAER imagery support: Background, Products and Applications. Role of satellite BARC and field-adjusted soil products.
The audit preserves the source archive, SHA-256 hashes, reader versions, code and results. It does not resample the raster, apply an ownership mask or interpolate monitoring data. Class counts agree between independent Rasterio and Pillow reads; grid-area and pixel totals reconcile. The chart depicts computed quantities, not an official agency graphic.
Cite: Tecnobosque, Holiday Farm Fire 2020, record TB-WBB-US-2020-HOLIDAY-FARM, version 1.0, evidence review and raster audit 12 October 2026.