Behind the numbers

Every number on this site, and how we got it.

Each claim below shows its source, the steps we took, the result, what a skeptic might ask, and how to re-run it yourself. Where a number comes from our own model, it says so in the first line. Public records are linked to the public record, never to our own analysis.

Version 1, 4 November 2026. When a number on the site changes, this page changes with it.

Claim 1
Public record Homepage hero, footnote 1

Near Kelowna, 7 in 10 wildfires are human-caused.

Near Kelowna, 7 in 10 wildfires are human-caused. (Footnote 1: Of the 140 wildfires recorded within 15 km of downtown Kelowna from 2010 to 2025, 97 were human-caused, 24 were caused by lightning and 19 have an unknown cause.)

Where the data comes from

BC Wildfire Service, Fire Incident Locations (Historical), a provincial dataset on the BC Data Catalogue. Every record is one incident the service attended, with its location, year, type and cause as the service recorded it. We did not change any cause.

How we counted, step by step

  1. Centre point: downtown Kelowna, 49.888 N, 119.496 W (near Bernard Avenue and Water Street).
  2. Radius: 15 km. This covers Kelowna, West Kelowna, Westbank First Nation, Lake Country and Peachland, the area the first 1,552 sensors would cover. The farthest fire returned is 14.9 km out.
  3. Period: fire years 2010 to 2025, the sixteen most recent complete seasons.
  4. Record type: only records of type “Fire”. The dataset also holds nuisance fires (mostly abandoned campfires logged at 0.01 ha), smoke chases that turned out not to be fires, training, prescribed burns and duplicates. We count fires, not reports.
  5. Cause: the FIRE_CAUSE field as recorded by the service: Person, Lightning or Unknown.
  6. Count: divide Person by the total.

Result

Cause (as recorded by BCWS) Fires Share of all 140 Share of the 121 with a known cause
Person 97 69.3% 80.2%
Lightning 24 17.1% 19.8%
Unknown 19 13.6%
Total 140 100%

97 of 140 is 69.3%, which rounds to 7 in 10. We did not use the 80% figure (human-caused among fires with a known cause) because it assumes the unknown fires split the same way as the known ones. 7 in 10 is the conservative reading.

By year, 2010 to 2025
Year Person Lightning Unknown Total
2010 8 1 1 10
2011 7 2 1 10
2012 11 4 2 17
2013 4 0 1 5
2014 5 4 0 9
2015 4 0 2 6
2016 3 2 0 5
2017 5 0 1 6
2018 7 5 1 13
2019 4 1 3 8
2020 1 0 1 2
2021 9 0 0 9
2022 7 2 0 9
2023 7 1 3 11
2024 1 2 2 5
2025 14 0 1 15

What if the circle, the years or the filter were drawn differently?

A fair challenge is that 7 in 10 depends on choices we made. So we ran the same count nine ways: three radii, three periods, with and without the nuisance-fire records. The shaded row is the one on the site.

Radius Years Records counted Person Lightning Unknown Total Person share Known cause only
10 km 2010 to 2025 Fire only 39 12 11 62 62.9% 76.5%
10 km 2010 to 2025 Fire + Nuisance Fire 51 12 23 86 59.3% 81.0%
10 km 2000 to 2025 Fire only 66 20 19 105 62.9% 76.7%
10 km 2000 to 2025 Fire + Nuisance Fire 78 20 57 155 50.3% 79.6%
10 km 2016 to 2025 Fire only 29 5 6 40 72.5% 85.3%
10 km 2016 to 2025 Fire + Nuisance Fire 41 5 12 58 70.7% 89.1%
15 km 2010 to 2025 Fire only 97 24 19 140 69.3% 80.2%
15 km 2010 to 2025 Fire + Nuisance Fire 141 24 75 240 58.8% 85.5%
15 km 2000 to 2025 Fire only 179 55 35 269 66.5% 76.5%
15 km 2000 to 2025 Fire + Nuisance Fire 223 55 204 482 46.3% 80.2%
15 km 2016 to 2025 Fire only 58 13 12 83 69.9% 81.7%
15 km 2016 to 2025 Fire + Nuisance Fire 102 13 39 154 66.2% 88.7%
25 km 2010 to 2025 Fire only 192 74 35 301 63.8% 72.2%
25 km 2010 to 2025 Fire + Nuisance Fire 318 74 215 607 52.4% 81.1%
25 km 2000 to 2025 Fire only 344 167 56 567 60.7% 67.3%
25 km 2000 to 2025 Fire + Nuisance Fire 470 167 551 1188 39.6% 73.8%
25 km 2016 to 2025 Fire only 116 52 18 186 62.4% 69.0%
25 km 2016 to 2025 Fire + Nuisance Fire 242 52 82 376 64.4% 82.3%

Counting fires only, the human-caused share is between 60% and 73% in every window. It drops below 60% only when nuisance-fire records are included, and the Unknown column shows why: those records are mostly abandoned campfires with no cause recorded, so they inflate the total without adding known lightning fires. Among records with a known cause, the human share is 67% to 89% everywhere. 7 in 10 sits in the middle of all of this.

What a skeptic might ask

“You cherry-picked 15 km and 2010.” Nine windows are shown above. Fire-only counts give 60% to 73% in all of them. We chose the area the first sensors cover and the most recent sixteen seasons.
“The unknown causes could all be lightning.” Then the human share would still be 97 of 140, 69%. Person is counted directly, so the share does not depend on how the unknowns are assigned.
“Province-wide it is 40%, so Kelowna cannot be 70%.” The provincial figure is dominated by remote lightning fires in the north. Near a city, people are the main source of ignition. The site states both figures side by side for exactly this reason.
“One of your records is marked DUPLICATE.” Correct. K52283 (October 2025) carries “DUPLICATE” in its incident name but is typed Fire with cause Person in the catalogue, so the query returns it. Removing it gives 96 of 139, which is 69.1%, still 7 in 10. We leave the query result as it is and disclose the record here.
“Which 97 fires?” All 140 are listed below with the BCWS fire number, date, cause, location and size.
Show all 140 fires (BCWS records, type Fire, within 15 km of downtown Kelowna, 2010 to 2025; shaded rows are the 97 person-caused)

BC Wildfire Service records of type Fire within 15 km of downtown Kelowna, fire years 2010 to 2025, retrieved 4 November 2026. Sorted by date. Shaded rows are the 97 person-caused fires. Row 140 (K52283) carries “DUPLICATE” in its recorded name; see the skeptic answers in claim 1.

# Fire no. Date Cause Name or location (as recorded) Size (ha)
1 K50032 2010-04-24 Person Behind Wilson Landing fire hall 0.1
2 K50160 2010-07-15 Person Lakeshore Road Kelowna 9.76
3 K50233 2010-07-24 Person Bear Creek 0.009
4 K50237 2010-07-25 Unknown Bear Creek / Sailview Bay / 13km S of Bear Creek 0.9
5 K50395 2010-08-06 Person KLO Creek 0.009
6 K50422 2010-08-10 Person KLO Creek 0.009
7 K50425 2010-08-10 Person KLO Creek #2 0.009
8 K50430 2010-08-11 Person KLO Creek #3 0.009
9 K50465 2010-08-18 Person Knox Mountain 0.009
10 K50577 2010-09-19 Lightning Bear Creek Prov Park 0.009
11 K50005 2011-04-11 Person Dilworth Mount 1.5
12 K50077 2011-06-03 Person W of Okanagan Lake, Bear Creek and Blue Grouse Rds 0.009
13 K40119 2011-07-01 Person NE of Kelowna Airport 0.2
14 K40125 2011-07-02 Person 1km off Postill Lake Road 0.009
15 K50124 2011-07-03 Person 2446 Reece Road - W Kel 2
16 K50322 2011-08-09 Lightning Myra Bellevue Prov Park 0.009
17 K50303 2011-08-10 Lightning Mount Hayman 0.009
18 K50458 2011-08-21 Person W of W Kelowna 0.009
19 K50603 2011-09-05 Unknown Bear Creek Park 38.8
20 K50605 2011-09-06 Person 400 block Westside Road 0.009
21 K50015 2012-04-10 Unknown Myra Bellvue 0.009
22 K50065 2012-05-13 Person Blue Grouse Mountain 0.009
23 K50114 2012-06-17 Person Gellatly Rd. and Carrington Rd. 0.009
24 K50140 2012-06-30 Lightning Myra-Bellevue Park 0.009
25 K50165 2012-07-09 Lightning June Springs Rd. 0.009
26 K50167 2012-07-09 Lightning Blue Grouse Mt. 0.2
27 K50351 2012-07-27 Lightning Hill above Wilson Landing (West Kelowna) / Wilsons Landing 27
28 K50515 2012-08-10 Unknown Myra Creek 0.3
29 K50532 2012-08-12 Person Bear Creek Rd 0.3
30 K50572 2012-08-18 Person 3 km Chute Lake Rd 0.009
31 K50840 2012-09-03 Person Bear Creek 0.009
32 K50912 2012-09-20 Person Westside Rd 0.6
33 K50914 2012-09-21 Person Black Knight Mtn 41
34 K50941 2012-09-28 Person Carrot Mountain 0.009
35 K50974 2012-10-05 Person The Boulderfields - Kelowna 0.009
36 K50975 2012-10-05 Person Myra Canyon 0.009
37 K50988 2012-10-09 Person Smith Creek 0.009
38 K40035 2013-05-01 Person Glenmore + Carson Rd - Lake Country 0.3
39 K50105 2013-06-12 Person Gellatly Rd 0.6
40 K50221 2013-07-20 Person Kalamoir Park 0.6
41 K40258 2013-07-25 Person 2km N of Kelowna Airport 3.1
42 K50306 2013-07-30 Unknown Scenic Road 0.4
43 K40022 2014-04-08 Person Okanagan Centre 0.5
44 K50093 2014-05-28 Lightning Shanks Road 0.009
45 K50180 2014-07-02 Lightning Peterson Rd and Bear Main 0.009
46 K50183 2014-07-02 Lightning 2km Lost Lake Trail 0.009
47 K50192 2014-07-02 Lightning Traders Cove 0.1
48 K50249 2014-07-13 Person Smith Creek 0.3
49 K50273 2014-07-15 Person Mount Boucherie 7.5
50 K50293 2014-07-17 Person Smith Creek 345
51 K50435 2014-08-05 Person Mt Boucherie 0.2
52 K50059 2015-05-13 Person 1 km Myra Canyon FSR 0.009
53 K50139 2015-06-16 Person 1 km Myra FSR 0.009
54 K50161 2015-06-26 Unknown Rose Valley Dam 0.5
55 K50389 2015-07-19 Unknown Bear Creek Park 2.5
56 K50722 2015-08-27 Person 5km Gillard FSR 0.009
57 K50774 2015-10-18 Person 4km Scotty Creek Road 0.009
58 K50109 2016-06-17 Person Spland Road 1.5
59 K50132 2016-06-27 Lightning Farmers Road 0.009
60 K50324 2016-08-08 Lightning Powers Creek 0.009
61 K50403 2016-08-21 Person Bear Creek 60.34
62 K50478 2016-09-25 Person Towers Ranch Golf Course 0.009
63 K50314 2017-06-13 Person Blue Grouse 2
64 K51448 2017-07-30 Person Knox Mountain 0.7
65 K51771 2017-08-17 Unknown Webber Road 0.009
66 K51839 2017-08-22 Person Black Mountain 2.1
67 K52158 2017-09-18 Person Powers Creek 0.009
68 K52163 2017-09-18 Person Glenrosa Telemark Trail 0.1
69 K50406 2018-05-26 Person Mission Hill 3.5
70 K50432 2018-05-27 Person 5km on Gillard FSR 0.009
71 K50460 2018-06-02 Person 1.5 km Gillard FSR 0.009
72 K50548 2018-06-17 Unknown Carre Road 0.009
73 K51070 2018-07-06 Person Telemark 0.009
74 K51133 2018-07-13 Person 4km Up Bear Main FSR 0.009
75 K51252 2018-07-17 Lightning Carrot Mountain / Carrot Mtn 1 0.1
76 K51253 2018-07-17 Lightning Soul Train Summit 0.009
77 K51285 2018-07-17 Lightning Carrot Mtn 3 0.009
78 K52469 2018-07-17 Lightning Carrot Mountain 0.2
79 K51953 2018-07-18 Lightning Mount Swite 0.1
80 K51369 2018-07-20 Person Knox Mtn 0.2
81 K51497 2018-07-27 Person Yates Rd 4.8
82 K50097 2019-04-21 Unknown Bear Main / 3 km Bear Main FSR 0.009
83 K50181 2019-05-05 Person Top of Blue Grouse Mountain 0.009
84 K50344 2019-05-23 Person Blue Grouse Mountain Rd 0.009
85 K50657 2019-06-18 Unknown Powers Creek 0.6
86 K50661 2019-06-18 Person Carrington / Carrington Road 0.1
87 K50667 2019-06-18 Person Mt. Swite / Mt. Swite 0.009
88 K50980 2019-07-18 Lightning Medicine Creek 0.009
89 K51753 2019-09-08 Unknown Gillard Creek FSR 0.009
90 K50034 2020-04-10 Person E of Law Creek 0.009
91 K50848 2020-08-02 Unknown Rose Vally Dam Fire / Rose Valley 3
92 K50444 2021-05-01 Person olk / Old Tolko MIll site 0.1
93 K50480 2021-05-09 Person Westside Road 0.2
94 K50911 2021-06-25 Person LOUIE DRIVE / Louie Drive 0.009
95 K51176 2021-07-01 Person McDougal Creek 0.009
96 K51709 2021-07-10 Person Teltron / Snowsell Creek 0.2
97 K51984 2021-07-16 Person Bear Creek 0.3
98 K52716 2021-08-24 Person Rose Valley Lake 2.67
99 K52723 2021-08-24 Person 7km off the Gillard 0.1
100 K52846 2021-09-20 Person HIDDEN CREEK / Hidden Creek 0.4
101 K50647 2022-07-19 Person Keefe Creek 2.8
102 K51454 2022-08-14 Person Robert Lake 0.25
103 K52332 2022-09-08 Person Blue Grouse Creek 0.1
104 K52450 2022-09-17 Lightning CARROT MOUNTAIN / Carrot Mountain 0.01
105 K52651 2022-10-10 Person June Springs / Klo Creek 1.5
106 K52676 2022-10-10 Person Hidden Creek FSR 0.009
107 K52759 2022-10-10 Person HAYMAN CREEK / Hayman Creek 2.04
108 K52661 2022-10-11 Lightning McDougall Creek / Mcdougall Creek 2.2
109 K52793 2022-10-25 Person Lambly Creek 0.009
110 K50375 2023-05-18 Lightning 2.5 Km off Bear Main FSR 0.009
111 K50607 2023-06-03 Person Gillard FSR 5 Km 0.009
112 K50729 2023-06-11 Person 3 km Bear Creek FSR / 3.5 Km Bear Creek 0.7
113 K51040 2023-07-01 Person Knox Mountain 6.3
114 K51745 2023-07-15 Person McDougall Creek 0.009
115 K52451 2023-08-03 Person 5.5 Gillard FSR 0.009
116 K52767 2023-08-15 Unknown McDougall Creek 13970.4
117 K52772 2023-08-16 Person Dilworth Creek 0.009
118 K52808 2023-08-17 Unknown WALROY LAKE / Walroy Lake 733
119 K52809 2023-08-17 Unknown Porcupine Crescent 0.009
120 K53475 2023-11-30 Person Prather Creek 0.009
121 K50247 2024-05-06 Unknown K50247 / Lambly Creek 0.1
122 K50436 2024-06-05 Lightning K50436 / North East of Lambly Creek 0.009
123 K50478 2024-06-11 Person K50478 / McDougall Creek Road 0.009
124 K50516 2024-06-17 Lightning K50516 / Deeper Creek 0.009
125 K51828 2024-08-05 Unknown K51828 / Scenic Canyon 0.03
126 K50148 2025-04-25 Person K50148 / Carre Road 0.1
127 K50196 2025-05-01 Person K50196 / Bertram Creek 0.009
128 K50505 2025-06-05 Person K50505 / Mill Creek 0.5
129 K50859 2025-07-06 Person Whelan Creek 4.9
130 K51515 2025-08-13 Person K51515 / Powers Creek 0.1
131 K51519 2025-08-13 Person K51519 / Smith Creek 0.15
132 K51589 2025-08-21 Person K51589 / Hydraulic Creek 0.02
133 K52075 2025-09-10 Unknown K52075 / Baldylocks Boulder 0.009
134 K52090 2025-09-12 Person K52090 / 2km Jackpine FSR 0.1
135 K52166 2025-09-21 Person K52166 / 6km up Gillard Creek 0.009
136 K52259 2025-10-13 Person K52259 / Bear Main 0km 0.64
137 K52272 2025-10-13 Person K52272 / Priest Creek 0.009
138 K52274 2025-10-15 Person K52274 / South of Bear Creek 0.2
139 K52277 2025-10-16 Person K52277 / Priest Creek 0.009
140 K52283 2025-10-17 Person K52283 / DUPLICATE -- K52274 -SOUTH OF BEAR CREEK 0.1

Re-run it yourself

This address returns the count of person-caused fires in the site window (97). Change FIRE_CAUSE to Lightning or Unknown for the other counts, or remove the cause clause for the total (140). Replace resultType=hits with outputFormat=json to get the full list.

Open the live query (97 person-caused fires) ↗

Sources for this claim

↑ Back to the list of claims
Claim 2
Public record Homepage, footnote 1

Across BC, human activity causes about 40% of wildfires.

Across BC, human activity causes about 40% of wildfires (BC Wildfire Service, What causes wildfire).

Verification

The Province of British Columbia’s page “What causes wildfire”, checked on 4 November 2026, states that lightning causes approximately 60% of wildfires every season and human activity approximately 40%. It gives these as approximate figures with no multi-year average. Our wording, “about 40%”, matches the page. We make no further claim from it.

Why both numbers are on the site

A reader who knows the provincial figure would otherwise think 7 in 10 is a mistake. Showing both makes the point: near a town, the fires that start are mostly started by people, and those are the fires that start at the edge of town, where a sensor can be.

Sources for this claim

↑ Back to the list of claims
Claim 3
Model result Homepage black box, footnote 2

1,552 sensors. The first wall. The layout is a model.

1,552 sensors. The first wall. The layout is a model: locations may change as we build and test. (Footnote 2: The 1,552-sensor grid is our own design based on BC Wildfire Service fire records, 1950 to 2025. It has not yet been tested in the field.)

This is a model result. The number is the output of a placement model we built in October 2026 (the Okanagan Fire Atlas, version 1). It is not a count of anything that exists yet.

Inputs, all public

Layer Publisher Used for
PROT_HISTORICAL_INCIDENTS_SP BC Data Catalogue (BC Wildfire Service) Where fires started, 1950 to 2025, with cause and size
PROT_HISTORICAL_FIRE_POLYS_SP BC Data Catalogue Mapped fire perimeters (734 in the study area)
PROT_FUEL_TYPE_SP BC Data Catalogue Which land is burnable; non-fuel and water are excluded
PROT_WUI_HMN_INTRFCE_BUFFR_SP BC Data Catalogue The provincial wildland-urban interface buffer
TRIM_EBM_BUILDINGS BC Data Catalogue Building footprints, to measure distance from each grid cell to the nearest building
Kelowna Airport weather, 1990 to 2025 Environment and Climate Change Canada Wind rose by day and night for the detection probabilities

How the grid was built, step by step

  1. Study area: a circle of 25 km radius around downtown Kelowna.
  2. Burnable land: every cell of the provincial fuel map inside the circle that is not non-fuel or water. Orchards, fields and built-up land are coded non-fuel and are left out (see the caveats).
  3. The grid: a hexagonal grid with 500 m between neighbouring sensors over all burnable land. In our smoke model, 500 m is the spacing at which a sensor straight downwind sees a fire of about 7.5 m² on a typical afternoon, and the full grid detects 2 in 3 historical ignitions within 20 minutes. The full grid is 6,942 sensors.
  4. Phasing by distance to buildings: each cell is tagged by its distance to the nearest mapped building. Phase 1 is every sensor within 1 km of a building.
  5. Count Phase 1: 1,552 sensors. That is “the first wall”. At CA$300 a node, about CA$466,000 of hardware.
  6. Why that band first: in the same fire records, the first kilometre around buildings holds 47% of all ignitions since 2000 and 67% of the fires that grew past 1 ha.
Phase Distance to nearest building Sensors Hardware at $300 Share of ignitions 2000 to 2025
1 Within 1 km 1,552 about CA$466k 47%
2 1 to 2 km 1,338 about CA$401k 25%
3 2 to 5 km 2,896 about CA$869k about 25%, half lightning
4 Beyond 5 km, inside the circle 1,156 about CA$347k 3%
Total 6,942 about CA$2.1M 100%

Caveats we state ourselves

  • The building layer (TRIM) is old and thin inside dense suburbs, so “distance to a building” in practice reads as distance to the rural and suburban fringe. That is the edge of town, which is what we want, but the exact count would move with a better building layer.
  • Orchard and agricultural land is coded non-burnable, so the grid has no nodes there. Two fires that reached homes, Rose Valley 2009 and Nighthawk Road 2017, started 0.9 and 1.4 km from the nearest node for this reason. A few hundred extra sensors on the orchard fringe would close that gap; they are not in the 1,552.
  • The grid is uniform. The real build will follow hosts, access and terrain, which is why the site says locations may change.
  • Nothing about the grid has been tested in the field.

5 sensors built, so 1,552 minus 5 is the 1,547 “to go” in the black box.

↑ Back to the list of claims
Claim 4
Model result Homepage, Fact 03, footnote 3

In our model, it raised the alarm on 8 of 11 local fires within 6 minutes.

In our model, it raised the alarm on 8 of 11 local fires within 6 minutes. We replayed the 15 largest Central Okanagan fires since 2003, with a typical summer-afternoon wind from Kelowna Airport records and a sensor every 500 metres. Eleven of them started inside our study area. (Footnote 3: our own smoke model, not a field result; we checked 28 Central Okanagan wildfires since 2003 and modelled the 15 largest.)

This is a model result. The fires are real and documented. The alarm times are what our smoke model computes for a sensor placed by the grid in claim 3.

Step 1: the fires we checked

In October 2026 we built a case file for every notable Central Okanagan fire since 2003 that we could document from the public record (BC Wildfire Service, Castanet, Global News BC, CBC, Kelowna Daily Courier, KelownaNow, iNFOnews, Penticton Herald, City of Vernon, Central Okanagan Emergency Operations, the Firestorm 2003 review). For each: date and time of discovery, cause, ignition location, who found it, size at first report, weather, final size, homes lost.

Show the 27 fires in the per-fire tables, with wind at the ignition hour
Fire Ignition date and hour Final size Wind at ignition hour (m/s)
Okanagan Mountain Park 2003-08-16, 03h 25,912 ha 0.0 (still)
Terrace Mountain 2009-07-18, 17h 9,300 ha 5.6
Glenrosa 2009-07-18, 14h 400 ha 4.7
Rose Valley 2009-07-18, 19h 200 ha 1.9
Trepanier / Peachland 2012-09-09, 15h 200 ha 3.6
Smith Creek 2014-07-17, 13h 340 ha 4.7
Knox Mountain 2015-06-21, 20h 3 ha 4.2
Westside Road 2015-07-21, 15h 460 ha 6.1
Little White Mountain 2015-08-27, 15h 194 ha 5.8
Nighthawk Rd / Okanagan Centre 2017-07-15, 22h 55 ha, 8 homes 3.1
Knox Mountain 2017-07-30, 14h 0.75 ha 2.5
Philpott Road / Joe Rich 2017-08-24, 13h 465 ha 5.8
Finlay Creek 2017-09-02, 17h 2,224 ha 7.2
Mount Eneas 2018-07-17, 20h 1,790 ha 5.8
Goode's Creek 2018-07-19, 15h 1,370 ha 2.2
Christie Mountain (Penticton) 2020-08-18, 14h 2,122 ha 0.0 (still)
White Rock Lake 2021-07-13, 15h 83,047 ha 2.2
Brenda Creek 2021-07-14, 15h 824 ha 2.2
Mount Law 2021-08-15, 17h 976 ha 9.7
Knox Mountain 2023-07-01, 14h 6.5 ha 9.7
Dilworth rail trail 2023-07-20, 19h 0.05 ha 1.1
McDougall Creek 2023-08-15, 17h 13,970 ha 3.1
Walroy Lake (ember start) 2023-08-17, 22h 769 ha 3.6
McCurdy Rd 2024-06-14, 10h 0.05 ha 8.9
Glenrosa / Hwy 97 2024-08-18, 17h 0.1 ha 9.4
Whelan Creek 2025-07-06, 16h 6 ha 5.8
Munro Lake / Peachland 2025-09-25, 15h 310 ha 8.3
A note for the record. The case-file set was 28 fires; the per-fire tables we kept list 27. We could not retrace the 28th by name, so until the research notes confirm it, read “28 checked” as “27 checked”. This is the only number on the site we could not fully retrace.

Step 2: the 15 largest, replayed in the model

The 15 largest by final size were placed in the model. For each, the ignition point was estimated from the news and agency descriptions (to a few hundred metres in town, 1 to 3 km in the backcountry), the nearest sensor in the 500 m grid was found, and the model computed the first minute at which that sensor, if straight downwind, would see a rise of 25 µg/m³ in fine particles above background. The wind for every fire was the same: a typical summer-afternoon wind of 3.5 m/s from the Kelowna Airport record. Shaded rows count toward 8 of 11.

Fire Year How it was actually found Size at first report Final size Nearest grid sensor Model alert if downwind Counts?
Okanagan Mountain Park 2003 Lightning at about 2 am, witnesses saw the strike not recorded 25,912 ha, 239 homes 210 m 4.5 min Yes
Glenrosa 2009 Mill forester from the road, about 2:30 pm not recorded 400 ha, 4 homes 189 m 4.1 min Yes
Terrace Mountain 2009 Not recorded, about 5:30 pm not recorded 9,300 ha outside the 25 km circle No (outside)
Rose Valley 2009 Not recorded, about 7:30 pm not recorded 200 ha 917 m (orchard cells, no node) 16 min No (too far)
Trepanier 2012 Not recorded, about 3 pm 200 ha by 7 pm 200 ha, 4 homes 283 m 5.7 min Yes
Smith Creek 2014 911 call as a grass fire, 1:20 pm about 2 ha 340 ha 208 m 4.4 min Yes
Nighthawk Rd / Okanagan Centre 2017 Not recorded, evening, arson not recorded 55 ha, 8 homes 1,354 m (orchard cells, no node) 24 min No (too far)
Philpott Rd / Joe Rich 2017 Not recorded, about 1:30 pm 300 ha same day 465 ha 242 m 5.0 min Yes
Finlay Creek 2017 Public photos to media, about 5 pm 10 ha 2,224 ha outside the 25 km circle No (outside)
White Rock Lake 2021 Not recorded; crew on site 30 min after report 10 to 100 ha 83,047 ha, about 90 homes outside the 25 km circle No (outside)
Mount Law 2021 Passing driver saw candling, 5:22 pm 40 ha by 9 pm 976 ha, 1 home 230 m 4.8 min Yes
Knox Mountain 2023 Canada Day crowds via BCWS app, about 2 pm 0.5 ha 6.5 ha 222 m 4.7 min Yes
McDougall Creek 2023 Report of a small puff of smoke, 5:59 pm under 1 ha 13,970 ha, about 190 homes 236 m 4.9 min Yes
Whelan Creek 2025 Not recorded, about 4 pm 4 to 6 ha 6 ha 832 m 15 min No (too far)
Munro Lake / Peachland 2025 Not recorded, afternoon about 70 ha 310 ha outside the 25 km circle No (outside)

Step 3: the 11 inside the study area

Four of the 15 started outside the 25 km circle where the grid is drawn: Terrace Mountain 2009, Finlay Creek 2017, White Rock Lake 2021 and Munro Lake 2025. They are not counted either way. That leaves 11.

Step 4: the 8 that would have raised the alarm

Eight of the 11 had a grid sensor within 300 m of the estimated ignition point and a modelled alert between 4.1 and 5.7 minutes: Glenrosa 2009 (189 m, 4.1 min), Smith Creek 2014 (208 m, 4.4), Okanagan Mountain Park 2003 (210 m, 4.5), Knox Mountain 2023 (222 m, 4.7), Mount Law 2021 (230 m, 4.8), McDougall Creek 2023 (236 m, 4.9), Philpott Road 2017 (242 m, 5.0), Trepanier 2012 (283 m, 5.7). The slowest is 5.7 minutes, which the site rounds to “within 6 minutes”.

The three that fail: Rose Valley 2009 (917 m, 16 min) and Nighthawk Road 2017 (1,354 m, 24 min) started in orchard land that the fuel map codes non-burnable, so the grid has no node there; Whelan Creek 2025 (832 m, 15 min) started on the rural fringe east of the airport, between cells. These three are why claim 3 recommends extra nodes on the orchard fringe.

The model and its settings

Setting Value Why
Wind 3.5 m/s, “typical afternoon” The Kelowna Airport fire-season afternoon mean is 4.1 m/s; 3.5 is slightly conservative
Stability Pasquill-Gifford B to C Sunny afternoon, strong mixing
Detection threshold +25 µg/m³ above a rolling baseline Kelowna summer background is 2 to 6 µg/m³; the sensor resolves 1 µg/m³ with a stated tolerance of ±10
Emission factor 15 g PM2.5 per kg of fuel (flaming) Published range 6 to 25 g/kg for conifer and grass
Fire growth FBP curves, C-7 pine and O-1 grass The two fuel types that cover 96% of Kelowna vegetated land
Sensor Straight downwind, 2.5 m above ground Best case for direction; the grid, not one sensor, handles other directions
Terrain Flat, steady wind, uniform fuel A simplification; slopes over 30% can double spread rates

The model is a Gaussian plume with Briggs rural spread and buoyant rise, fed by a fire that grows on Canadian Forest Fire Behaviour Prediction curves. Its full documentation, including the equation, the six Kelowna weather cases and the assumptions a reviewer should know, will be published with the code.

A second run, with each fire’s own wind

A fair objection to Step 2 is that one afternoon wind does not fit a fire that started at 3 am. So we also ran all 27 fires with the wind recorded at Kelowna Airport in each fire’s own ignition hour and the matching stability and growth curve. Result: 23 of 27 would have alerted a sensor 500 m downwind within 10 minutes. The four exceptions are the calm cases, where a 500 m sensor takes 12 to 41 minutes. That is the basis for the footnote sentence “fires that start at night or in calm air would be detected later”.

Show the 27-fire run (alert at 100, 200 and 500 m; shaded rows take more than 10 minutes at 500 m)

Shaded rows take more than 10 minutes at 500 m; all four are calm or near-calm cases.

Fire Ignition date, hour Final size Wind (m/s) Alert at 100 m Alert at 200 m Alert at 500 m
Okanagan Mountain Park 2003-08-16, 03h 25,912 ha 0.0 (still) 3.5 min 6.8 min 16.8 min
Terrace Mountain 2009-07-18, 17h 9,300 ha 5.6 2.3 min 3.8 min 7.3 min
Glenrosa 2009-07-18, 14h 400 ha 4.7 2.8 min 4.5 min 9.5 min
Rose Valley 2009-07-18, 19h 200 ha 1.9 3.3 min 5.5 min 12.3 min
Trepanier / Peachland 2012-09-09, 15h 200 ha 3.6 2.8 min 4.5 min 9.3 min
Smith Creek 2014-07-17, 13h 340 ha 4.7 2.8 min 4.5 min 9.5 min
Knox Mountain 2015-06-21, 20h 3 ha 4.2 2.0 min 3.0 min 6.0 min
Westside Road 2015-07-21, 15h 460 ha 6.1 2.0 min 3.0 min 5.8 min
Little White Mountain 2015-08-27, 15h 194 ha 5.8 2.3 min 3.8 min 7.5 min
Nighthawk Rd / Okanagan Centre 2017-07-15, 22h 55 ha, 8 homes 3.1 2.3 min 3.8 min 7.3 min
Knox Mountain 2017-07-30, 14h 0.75 ha 2.5 2.5 min 4.5 min 9.5 min
Philpott Road / Joe Rich 2017-08-24, 13h 465 ha 5.8 2.3 min 3.8 min 7.5 min
Finlay Creek 2017-09-02, 17h 2,224 ha 7.2 1.0 min 1.8 min 3.3 min
Mount Eneas 2018-07-17, 20h 1,790 ha 5.8 2.0 min 3.0 min 5.8 min
Goode's Creek 2018-07-19, 15h 1,370 ha 2.2 2.5 min 4.5 min 9.5 min
Christie Mountain (Penticton) 2020-08-18, 14h 2,122 ha 0.0 (still) 3.5 min 11.8 min 41.3 min
White Rock Lake 2021-07-13, 15h 83,047 ha 2.2 2.5 min 4.5 min 9.5 min
Brenda Creek 2021-07-14, 15h 824 ha 2.2 2.5 min 4.5 min 9.5 min
Mount Law 2021-08-15, 17h 976 ha 9.7 1.0 min 1.8 min 3.0 min
Knox Mountain 2023-07-01, 14h 6.5 ha 9.7 1.0 min 1.8 min 3.0 min
Dilworth rail trail 2023-07-20, 19h 0.05 ha 1.1 5.0 min 10.3 min 26.5 min
McDougall Creek 2023-08-15, 17h 13,970 ha 3.1 2.5 min 4.3 min 9.3 min
Walroy Lake (ember start) 2023-08-17, 22h 769 ha 3.6 2.0 min 3.8 min 7.0 min
McCurdy Rd 2024-06-14, 10h 0.05 ha 8.9 1.0 min 1.8 min 3.0 min
Glenrosa / Hwy 97 2024-08-18, 17h 0.1 ha 9.4 1.0 min 1.8 min 3.0 min
Whelan Creek 2025-07-06, 16h 6 ha 5.8 2.3 min 3.8 min 7.5 min
Munro Lake / Peachland 2025-09-25, 15h 310 ha 8.3 1.0 min 1.8 min 3.3 min

Correction history

On 1 and 2 November 2026 we had the site’s numbers audited against the model and the primary data. Three figures were corrected: “9 of the 11 big fires” became 8 of 11 (a recount of the table); “we modelled 28 real Okanagan fires” became 15 modelled, 28 checked; “replayed with that hour’s wind” became “with a typical summer-afternoon wind”, because one wind was used for every fire in the 15-fire run.

What a skeptic might ask

“A model is not evidence.” Agreed, and the site says “a model result, not a field result” under the number. The model is published with its equations and assumptions, and the burn tests to check it are the next step.
“Your ignition points are guesses.” They are estimates from the public record, to a few hundred metres in town. On a 500 m grid, moving a point by 200 m rarely changes the nearest-sensor distance by more than that, and the 4 to 6 minute band absorbs it.
“Straight downwind is the best case.” Yes. That is why it is a grid: whichever way the wind blows, some sensor is downwind. The model also gives the probability of detection within 20 and 60 minutes over the real wind rose (56% and 94% for Smith Creek).
“The airport is 10 to 30 km from these fires.” True. The 3.5 m/s afternoon wind is the regional value, not the slope wind. Slope winds on a summer afternoon are usually stronger, which would shorten the times.
“Has a sensor like this been tested on a real fire?” Ours has, once, at two metres (claim 8). No one, including commercial vendors, publishes a detection-probability-by-distance curve from field burns. A witnessed burn programme at 25 to 300 m is in the plan, and it is what will replace this section.
↑ Back to the list of claims
Claim 5
Record plus model Homepage, Fact 03, footnote 4

Smith Creek 2014: 1:20 pm, about 2 hectares, 75 tennis courts, 4.5 minutes.

Smith Creek, 2014, was called in as a grass fire at 1:20 pm and was about 2 hectares (about 75 tennis courts) by the time the emergency centre opened. In the model, the nearest sensor would have seen it at four and a half minutes, when it was smaller than a tennis court.

The record

Fact Source What it says
Called in as a grass fire at about 1:20 pm, 17 July 2014 iNFOnews, 17 July 2014 The fire was called in at about 1:20 pm as a grass fire above Smith Creek Road, West Kelowna
About 2 hectares when the Emergency Operations Centre was activated ArmchairMayor.ca, 17 July 2014 The Central Okanagan Emergency Operations Centre activated as the fire reached about two hectares; evacuations followed
Final size 340 ha; about 2,500 evacuated BC Wildfire Service records; news record West Kelowna, 2014

The arithmetic

  • 1 hectare is 10,000 m², so 2 hectares is 20,000 m².
  • A doubles tennis court is 23.77 m by 10.97 m, which is 260.8 m².
  • 20,000 divided by 260.8 is 76.7 courts. The site says “about 75”, rounded down.

The model result

In the 500 m grid the nearest sensor to the estimated ignition point above Dixie Road is 208 m away. With the typical afternoon wind, the model alerts that sensor at 4.4 minutes, which the site rounds to four and a half. How big is the fire at 4.4 minutes? That depends on which growth curve applies, so here are all three plausible ones. Area grows roughly with the square of time, so each value is the 5-minute area in the FBP table scaled by (4.4 ÷ 5)².

Growth curve (FBP, 12 km/h wind) Area at 5 min Area at 4.4 min Tennis courts (261 m²)
C-7 pine and fir, afternoon 9 m² about 7 m² 0.03
Matted grass, afternoon 26 m² about 20 m² 0.08
Standing cured grass, afternoon 135 m² about 105 m² 0.4

Under every curve the fire is well under one tennis court at the moment of the modelled alert, against about 75 courts when the emergency centre opened. The comparison holds whichever fuel the fire was burning in.

What a skeptic might ask

“The fire could have been burning for a while before 1:20 pm.” Possibly, and the record does not say when it ignited. That cuts in our favour: the sensor clock starts at ignition, and the 911 clock starts when someone happens to see it. The gap between the two is exactly what the sensor is for.
↑ Back to the list of claims
Claim 6
Physics plus record Homepage, Fact 02

How fast and how wide the smoke travels.

How fast depends on the wind: 500 metres takes a few minutes on a summer afternoon, and about a minute in the winds that drive wildfires. The smoke also spreads and grows into a cloud as it travels. Two metres from a fire it is as wide as a doorway; 500 metres away it is a cloud about 200 metres across.

The wind numbers

Kelowna Airport hourly observations, May to September, 2019 to 2025 (25,644 observations, Environment and Climate Change Canada station 48369): summer mean wind 2.6 m/s, afternoon mean 4.1 m/s, night mean 1.6 m/s, 12% of hours calm. Every destructive run in the record happened in 7 to 10 m/s sustained valley wind with gusts of 14 to 19 m/s on the slopes: Okanagan Mountain Park on 22 August 2003, White Rock Lake on 15 August 2021, McDougall Creek on 17 August 2023.

Condition Wind Time for smoke to travel 500 m Site wording
Model “typical afternoon” 3.5 m/s 2.4 minutes a few minutes
Kelowna fire-season afternoon mean 4.1 m/s 2.0 minutes a few minutes
Wildfire-driving wind, low end 7 m/s 71 seconds about a minute
Wildfire-driving wind, high end 10 m/s 50 seconds about a minute
Summer mean, all hours 2.6 m/s 3.2 minutes still a few minutes

Travel time is distance divided by wind speed. The model alert times in claim 4 are longer (4 to 6 minutes at 200 to 300 m) because they also include the time the fire needs to grow and make enough smoke, and the 10-minute averaging at the sensor.

The width numbers

In a Gaussian plume the sideways spread is set by the stability of the air. The Briggs rural coefficients give the standard deviation of the plume width (sigma-y) at any distance. A practical “width of the cloud” is four sigma-y, the band within which concentration is at least one seventh of the centreline value. Shaded rows are the model’s typical afternoon.

Stability class Typical conditions Sigma-y at 500 m Width at 500 m Width at 2 m
A Very unstable, hot sunny, light wind 107 m 429 m 1.8 m
B Unstable, sunny afternoon 78 m 312 m 1.3 m
C Slightly unstable, sunny with breeze 54 m 215 m 0.9 m
D Neutral, overcast or windy 39 m 156 m 0.6 m
E Slightly stable, evening 29 m 117 m 0.5 m
F Stable, clear calm night 20 m 78 m 0.3 m

On a sunny afternoon with a breeze (class B to C) the cloud at 500 m is 215 to 312 m across. “About 200 metres” is the conservative end of that. At 2 m from the fire the same arithmetic gives 0.9 to 1.3 m, the width of a doorway. In stable evening air the cloud is narrower, about 120 m at 500 m, which is one more reason evening fires are harder for a grid.

What a skeptic might ask

“Smoke rises, it does not drift along the ground.” Both happen. The hot core rises on its own buoyancy, and the model assumes only 15% of flaming emissions and 50% of smouldering emissions stay near the ground. The site’s sentence is about the part that stays low, which is what a 2.5 m sensor reads, and the 4 to 6 minute results already carry that 15% assumption. It is the single assumption the field burns most need to measure.
↑ Back to the list of claims
Claim 7
Project record Homepage, Fact 01

What the sensor reads and how it reports.

One sensor counts smoke particles in three sizes; a second chip reads the gases a fire gives off, plus temperature and humidity. A fire moves all of these together, dust moves just one, and a small model on the board tells them apart. It sends its alert by satellite and charges itself from the sun.

The parts, as fitted

Claim Part What it does How to verify
Counts particles in three sizes Bosch BMV080 Reports PM1, PM2.5 and PM10 in µg/m³, 1 µg/m³ resolution Bosch datasheet; the readings in the field test video
Reads gases, temperature, humidity Bosch BME688 (SparkFun breakout) Gas resistance in kilo-ohms, temperature, relative humidity, pressure Bosch datasheet; the gas reading in the field test video
A small model on the board Arduino Nano ESP32 running the project firmware Builds the data packet; the detection rule compares the pattern across readings, not one number Firmware on GitHub
Sends by satellite Blues Notecard with satellite service Sends the packet without WiFi or cell coverage Notecard documentation; the three-minute arrival in the field test
Charges from the sun Solar panel and battery Keeps the board powered without a mains connection The device walkthrough video on the homepage

Every part is shown and named in the device walkthrough video on the homepage (2:12). The custom circuit board that carries them was designed by the team, and the first five were manufactured in 2026.

One wording note

“A small model on the board tells them apart” describes the detection rule we are building toward: a rolling baseline per node, a pre-alert at +10 µg/m³, an alert at +25 µg/m³ held for 60 seconds, confirmed by the PM1 to PM2.5 ratio, a gas rise, a neighbour node or persistence beyond three minutes. Its false-alarm rate has not yet been measured over a season; that measurement (alarms per node-day, with causes) is on the list of things to publish.

↑ Back to the list of claims
Claim 8
Project record Homepage, Fact 01 video

The field test readings.

First field test, Okanagan Lake, Kelowna, fall 2026. One sensor, one small controlled fire, two metres away. Before the fire: PM1 2, PM2.5 4, PM10 4 µg/m³, gas 54.76 kΩ. After the fire: PM1 19, PM2.5 42, PM10 58 µg/m³, gas 28.25 kΩ. Fire detected.

Where the numbers come from

They are read directly off the screen in the field test video on the homepage (1:26). The video is uncut from the lighting of the fire to the readings arriving.

Video time What is shown Reading
0:01 Baseline before the fire, on the laptop PM1 2, PM2.5 4, PM10 4 µg/m³; gas 54.76 kΩ
0:27 Kaan lights a small test fire two metres from the sensor
0:34 The sensor collects ten minutes of readings and sends them by satellite
0:57 Three minutes after sending, the readings appear on screen
1:11 Readings after the fire PM1 19, PM2.5 42, PM10 58 µg/m³; gas 28.25 kΩ

PM2.5 rose by 38 µg/m³, above the model’s +25 µg/m³ alert threshold. The gas resistance fell from 54.76 to 28.25 kΩ; a fall in resistance on the BME688 means more combustion gases in the air, the second signal the detection rule looks for.

What this test is and is not

  • It is one real sensor, one real fire, outdoors, with the readings travelling by satellite and arriving in three minutes. It shows the chain works end to end.
  • It is not a test of distance. The sensor was two metres from the fire. The 500 m layout is the next test.
  • It is not a false-alarm test. No barbecue, wood stove or dust comparison was run that day.
  • The background of 2 to 4 µg/m³ before the fire matches the Kelowna clean-air background used in the model (2 to 6 µg/m³).
↑ Back to the list of claims
Claim 9
Project record Homepage black box

5 built, 2 online, 3 waiting for a host, 1,547 to go.

5 built. 2 online. 3 waiting for a host. 1,547 to go.

Project records as of 4 November 2026. Five sensors have been built on the custom board. Two are installed and reporting in Kelowna; their readings are public through the hive. Three are built and waiting for someone to host them. 1,552 minus 5 is 1,547. The hive is the live source for “online”; when a sensor goes up or down, the number on the site follows it.

Other dates and spans on the site

Wording Meaning Check
“75 years of Okanagan wildfires”, “1950 to 2025” The span of the BC Wildfire Service historical incident layer used for the grid 2025 minus 1950 is 75
“since 2003” The first year of the case-file set; Okanagan Mountain Park 2003 is the modern reference fire Claim 4
“after the 2023 fire season” When the project started; the McDougall Creek fire was August 2023 Project record
↑ Back to the list of claims
What we do not know yet

The weak points, listed as carefully as the strong ones.

  1. No field anchor for the 4 to 6 minute figure. It comes from a plume model. The only vendor field test we found in the literature took 15 to 25 minutes to reach sensors 33 to 56 m away, with a different sensor and threshold. Ten witnessed burns under 10 m² with a node at 200 m alerting inside 6 minutes would settle it.
  2. Night and calm air. On a still night the smoke front reaches about 34 m in 10 minutes and 120 m in an hour. A 500 m grid is a daytime instrument; the 150 m spacing for trailheads and gullies in the model is the answer, and it is not in the 1,552.
  3. Ignition points are estimates, to a few hundred metres in town and 1 to 3 km in the backcountry.
  4. Airport wind is regional. The airport is 10 to 30 km from most of the fires and in the valley trough.
  5. 28 versus 27 fires checked. See claim 4.
  6. The near-ground smoke fraction (15% flaming) is a judgement parameter that controls nearly all stable-night detection. It is the first thing the burn programme should measure.
  7. False alarms have not yet been measured over a season.
  8. The building layer is old and thin in suburbs, so the 1,552 would move with a better layer.