Drones for Forest Fire Management in Northern Ontario

System Design & Feasibility Document

Project codename: BOREAL-UAS Author: Jerry Status: Concept exploration / feasibility design Related projects: RH-FC1 flight controller (technology feeder)

1. Executive Summary

Northern Ontario’s boreal forest presents one of the world’s hardest wildfire management problems: an area larger than France, mostly roadless, sparse population, lightning-dominated ignition, and a fire season that is lengthening and intensifying. Ontario’s Aviation, Forest Fire and Emergency Services (AFFES) fights this with a fleet of manned aircraft (CL-415 waterbombers, bird dogs, helicopters), ground crews, and increasingly satellite/AI detection.

The honest thesis of this document: drones will not replace waterbombers — the physics of direct suppression at scale is unfavorable — but there are four missions where uncrewed systems are decisively better than the status quo, and Northern Ontario’s geography is ideal for them under Canada’s new (2025/2026) BVLOS regulations:

1. Night-time IR fire mapping — manned aircraft largely stand down at night; fires don’t. This is the highest-value, most achievable mission.
2. Post-lightning early detection patrol — persistent, cheap surveillance of lightning strike corridors before smoke columns are visible from far away.
3. Mop-up hotspot verification — replacing crews walking fire perimeters with hand-held IR, one of the most labor-intensive phases.
4. Cargo resupply & comms relay for remote fire crews.

Direct suppression is treated separately (§4) with a physics reality check: it is viable only for spot-fire and structure-protection niches, not fire fronts.

The document defines a three-tier system architecture, the regulatory pathway (Level 1 Complex BVLOS + SFOC for fire-zone ops), the technical design of each tier, and a build/validation roadmap that starts from your existing flight controller and embedded Rust work.

2. Problem Context: Northern Ontario’s Fire Regime

2.1 The operating environment

• Scale: Ontario’s fire region spans ~90 M ha; the Northwest and Northeast fire regions are overwhelmingly roadless boreal forest (spruce, jack pine — high-intensity crown fire fuels).
• Ignition: roughly half of fires (and most burned area) are lightning-caused, arriving in spatial clusters after convective storms — often dozens of new starts in a 24–72 h window across areas the size of European countries.
• Detection today: public reports near communities; aerial detection patrols; lightning detection networks driving patrol routing; satellites (NASA FIRMS MODIS/VIIRS ~375 m resolution, several passes/day; Canada’s WildFireSat program purpose-built for afternoon fire monitoring, launching later this decade).
• Suppression doctrine: full response near communities/values; monitored (“appropriate response”) fires in the remote north. Initial attack speed is everything — a fire hit at 0.1 ha with a helicopter crew is trivial; the same fire at 1,000 ha two days later may be unstoppable until rain.
• AFFES assets: a fleet of 28 aircraft (20 fixed-wing including CL-415s, 8 helicopters), positioned dynamically on forecast fire hazard.

2.2 The capability gaps drones can actually fill

2.3 Constraints unique to this environment (design drivers)

• No cellular coverage across most of the operating area → C2 link cannot assume LTE. Satellite (Starlink-class) or long-range point-to-point RF required.
• Smoke: EO cameras degrade badly; LWIR thermal sees through most smoke — thermal is the primary sensor, not the secondary.
• Fire-generated weather: convection columns produce severe turbulence, downdrafts, and embers to high altitude; standoff and altitude discipline are flight-safety requirements, not preferences.
• Airspace deconfliction is THE safety issue: an unauthorized drone near a fire grounds all manned aircraft (this has repeatedly halted waterbomber ops in Canada). Flying near a forest fire is prohibited airspace by regulation (CARs restrict aircraft near forest fire areas / NOTAM’d fire zones). Any system here operates inside the fire organization with AFFES air attack coordination, or it operates far away from active incidents (detection patrol in pre-fire areas). There is no third mode.
• Cold starts, icing shoulder seasons, hot summer density altitude, dust and ember ingestion, chargingin the field from generators.
• Distances: a “nearby” fire can be 150 km from the nearest road. Endurance and range dominate the airframe trade space — this pushes Tier 1 away from multirotors toward fixed-wing VTOL.

3. Mission Analysis & Value Ranking

Scoring: value to fire management × technical feasibility × regulatory feasibility (1–5 each).

Program strategy that falls out of this table: enter through M3 (mop-up), build trust and flight hours, expand to M1 (night mapping), then M2 (BVLOS detection network) as the flagship. M4–M6 are options once the operator certificate, safety case, and AFFES relationship exist. M7 is research, not a product.

4. Direct Suppression: Physics Reality Check

This section exists because “firefighting drones” usually means “drones dropping water,” and the numbers must be faced before designing anything.

Energy scale. A modest 100 ha crown fire releases heat on the order of 10^13–10^14 J over its burn period; fireline intensity in boreal crown fire reaches 10,000–100,000 kW per meter of front. Suppressing flame front with water requires roughly 1–4 L/s per meter of active front (classic fireline figures) delivered continuously.

Delivery comparison:

A heavy multirotor delivers two orders of magnitude less water than one CL-415. Ten of them still deliver an order of magnitude less, with far more airspace complexity. Against an active boreal fire front, multirotor water drops are irrelevant.

Where the physics DOES close:

1. Incipient spot fires (< a few m²): tens of liters of water/retardant or suppressant balls genuinely can extinguish an ember-ignition, if you arrive within minutes. This couples naturally with M2 detection — the interceptor concept: detect a lightning holdover or ember spot at < 1 m² and kill it before it grows. High-risk R&D, potentially transformative, unproven at operational scale.
2. Structure protection: pre-wetting/gel application on cabins and remote values ahead of a front, where a 50 L drone making repeated trips from a lake over hours is doing a job crews can’t safely stay for.
3. Night suppression support: uncrewed aircraft (uncrewed helicopters in the 250+ kg class) can keep dropping when manned aviation stands down — this is the serious end of suppression UAS and is a fleet-scale investment, out of scope for a small program but worth tracking.

Design decision: BOREAL-UAS treats suppression as M7 research payload on the Tier 3 platform only; the program’s value case never depends on it.

5. System Architecture: Three Tiers

┌────────────────────────────────────────────────────────────────────┐
│                      BOREAL-UAS SYSTEM OF SYSTEMS                  │
│                                                                    │
│  TIER 1 "SENTINEL"        TIER 2 "MAPPER"        TIER 3 "MULE"     │
│  Fixed-wing VTOL          Quad/hex multirotor    Heavy-lift        │
│  detection patrol         tactical IR mapping    multirotor        │
│  25 kg class, 4–8 h       7–15 kg, 45–60 min     ~150 kg MTOW      │
│  BVLOS remote patrol      on-incident, night     cargo/ignition/   │
│                           perimeter + mop-up     research payloads │
│         │                        │                      │          │
│         └──────────┬─────────────┴──────────────────────┘          │
│                    ▼                                               │
│           GROUND SEGMENT: mobile GCS (truck/trailer),              │
│           Starlink backhaul, edge compute (detection AI),          │
│           charging/generator, ADS-B in + FLARM/RemoteID            │
│                    ▼                                               │
│           DATA SEGMENT: georeferenced IR/EO products →             │
│           GIS layers (GeoTIFF perimeters, hotspot points) →        │
│           fire management systems / incident command               │
└────────────────────────────────────────────────────────────────────┘

5.1 Tier 2 “MAPPER” — build this first

The tactical multirotor is the entry product (missions M3, M1). It is also the platform your RH-FC1 flight controller work feeds directly.

Airframe

• Configuration: quad (7”) for mop-up variant; X8 coax octo (15”) for the night mapping variant (redundancy over an incident matters — motor-out capability).
• MTOW 7–15 kg, weather-sealed (IP54 target), operating range −10 °C to +45 °C.
• Endurance: 45–60 min with payload (Li-ion packs, not LiPo, for energy density; heated battery compartment for shoulder-season ops).

Sensor payload (the heart of the system)

• LWIR radiometric thermal: 640×512 radiometric core (FLIR Boson+ class or equivalent), calibrated temperature output — radiometric is non-negotiable: you need actual temperatures to classify residual heat vs sun-warmed rock.
• EO camera: 20 MP+ mapping camera, mechanical/global shutter preferred for photogrammetry.
• 3-axis gyro-stabilized gimbal; nadir mapping mode + operator-steerable mode.
• Georeferencing: PPK GNSS (u-blox ZED-F9P class) + camera trigger timestamping → hotspot coordinates good to < 2 m without ground control points. A hotspot report a crew can’t navigate to is worthless; this drives the precision requirement.
• Laser rangefinder (optional v2) for target geolocation from oblique angles.

Autonomy & software

• Grid/corridor survey planning with terrain following (SRTM/HRDEM elevation).
• Onboard hotspot detection: threshold + blob detection on radiometric frames in real time; hotspots streamed to GCS as points with temp + confidence while the flight continues (crews get actionable data before landing).
• Products auto-generated on landing: orthomosaic (EO), thermal ortho, hotspot shapefile/KML, perimeter polygon (isotherm extraction from thermal ortho).
• Flight stack: PX4 or ArduPilot for the operational aircraft (certification and interlock maturity), with RH-FC1/Rust stack on the R&D airframe — dogfood your own FC where the risk is acceptable, fly the proven stack over incidents.

Fire-environment hardening

• Ember/dust filtration on cooling paths; no open ESC heatsinks facing down.
• Thermal ceiling logic: abort/climb on sustained ambient > limit (fire column proximity indicator via rate-of-climb anomaly + air temp sensor).
• Return-to-home logic that accounts for wind drift and never transits the fire column; geofenced “no-fly volumes” over active fire set by the operator.

5.2 Tier 1 “SENTINEL” — BVLOS detection patrol

Mission profile: after a lightning event, patrol assigned strike-cluster corridors at 100–120 m AGL (regulatory ceiling for lower-risk BVLOS), 300–500 km per sortie, dual EO/IR, detect smoke plumes and thermal anomalies, report detections with coordinates + imagery over satellite backhaul in near-real-time.

Airframe: fixed-wing VTOL (quad-plane) — vertical launch/recovery from clearings/barge/truck, wing-borne cruise for endurance. ~25 kg MTOW class, 4–8 h endurance, cruise 70–90 km/h. (Build vs buy analysis in §8 — likely buy/partner for Tier 1 airframe, build the payload + autonomy.)

Detection stack:

• Wide-FOV EO for smoke plume detection (CV model: smoke segmentation — plumes are visible far beyond thermal range in daytime).
• Nadir/forward LWIR for sub-canopy thermal anomalies on close passes.
• Edge inference onboard (Orin-class module): only detections + thumbnails go over the constrained satellite link; full data offloaded on landing.
• Fusion with tasking inputs: lightning network strike locations, FIRMS/CWFIS hotspots, fire weather indices → patrol routes auto-generated to maximize probability of detection per flight hour.

C2 & links: primary RF C2 (900 MHz/2.4 GHz long-range) within ~40 km of GCS; Starlink Mini (or equivalent) onboard for over-the-horizon telemetry/detection reporting; full lost-link behavior tree (loiter → climb-in-place NO (fire column risk) → preplanned lost-link route home at survey altitude → flight termination zones).

Airspace: ADS-B in (detect and avoid input), conspicuity out per whatever the operation’s authorization requires; strict NOTAM/fire-restriction geofencing compiled into the mission before launch; operations coordinated with the provincial fire center so patrol areas never overlap active air operations.

5.3 Tier 3 “MULE” — heavy lift (later phase)

• ~150 kg MTOW multirotor or uncrewed helo, 30–50 kg payload.
• Payloads: cargo pod (pumps/hose/fuel/food to crews), aerial ignition sphere dispenser (M6, with agency partnership only), water/gel module for M7 research.
• Almost certainly procure the platform; the program’s IP is the mission system, not a heavy airframe. Revisit at Phase 3.

5.4 Ground & data segment

• Mobile GCS: pickup/trailer — mast antennas, Starlink, 5 kW generator + charge station, rugged workstation with edge GPU, weather station.
• Data pipeline: flights → products (GeoTIFF/KML/shapefile) → synced to a simple web map (self-hosted, offline-capable) → export in whatever format the incident/AFFES wants. Don’t build a platform; build clean layers that drop into their existing GIS. Interop beats features.
• Fleet ops software: mission logs, airworthiness/maintenance tracking, pilot currency records — required for the RPOC operator certificate anyway; build it as boring structured data from day 1.

6. Regulatory & Airspace Pathway (Canada, 2026)

This is the section that determines whether the program is real. The regulatory landscape shifted materially in 2025–2026 and it now favors exactly this use case — but fire-zone operations remain the hard part.

6.1 The new BVLOS framework

• Canada introduced a Level 1 Complex Operations pilot certificate for lower-risk BVLOS: age 18+, the advanced online exam, at least 20 hours of ground school, and a flight review; certification doesn’t expire but requires recurrent training every two years. [TC amendments]
• Lower-risk BVLOS is confined to uncontrolled airspace, below 122 m (400 ft), at least 1 km from populated areas, with pilots required to verify population density of the operating area in advance — and organizations conducting BVLOS need an RPAS Operator Certificate (RPOC) demonstrating policies and procedures matched to their operation.
• The second phase of the new rules came into force November 4, 2025, enabling the lower-risk BVLOS operations plus EVLOS with a visual observer and medium-drone VLOS in controlled airspace, without case-by-case SFOCs.
• Higher-risk operations (dense areas, complex airspace, and categories outside the lower-risk definitions) still require an SFOC-RPAS.
• Note for platform selection: L1C-category operations require the manufacturer to hold a Pre-Validated Declaration against Standard 922 — meaning Tier 1 BVLOS aircraft must be a platform whose manufacturer has made the required declaration (or you become the “manufacturer” and carry that burden — a strong argument for buy-over-build on the Tier 1 airframe).

Implication: M2 (remote detection patrol over unpopulated boreal forest, under 122 m, uncontrolled airspace) is almost the canonical case the lower-risk BVLOS category was written for. Remote Northern Ontario population density is effectively zero over vast areas.

6.2 Fire-zone operations (the hard part)

• Airspace over/near active forest fires is restricted to fire-suppression aircraft (CARs forest-fire restriction + NOTAM’d zones). Unauthorized drones near fires have repeatedly grounded suppression aircraft in Canada; this is the single fastest way to make the program radioactive.
• Therefore M1/M3 on active incidents require: authorization tied to the incident (SFOC and/or operations under the fire agency’s umbrella), named coordination procedures with air attack supervision, defined altitude blocks and time windows (e.g., drone IR mapping block at night when manned aircraft are down — which is also when the mission is most valuable), and positive check-in/check-out with the fire’s air ops.
• Practical path: don’t be an outsider requesting access — be a contracted service provider inside AFFES’s aviation plan, the way helicopter and bird dog contractors are. That means: RPOC, insurance, SMS (safety management system), trained pilots, and a services agreement or participation in a provincial innovation/pilot program.
• Bill C-15 (royal assent March 2026) tightened the security framework around drones — including prohibitions on interfering with lawful drone operations and counter-drone authorities — underscoring that operating near emergency ops without authorization is treated severely.

6.3 Certification & organization checklist

6.4 Stakeholders

• AFFES / MNRF: the customer and the airspace gatekeeper. Early engagement (innovation office / fire science staff) before building anything bespoke.
• Transport Canada: certificates above.
• CIFFC / Natural Resources Canada: national coordination, fire-science programs (WildFireSat era = appetite for detection-layer innovation).
• First Nations communities: much of the operating area; FireSmart and community-protection programs are both a duty to engage and a real early market (community-perimeter detection patrols, structure-protection mapping).
• Federal wildfire funding: Ottawa is investing in early-warning technology explicitly combining satellites, cameras, drones, thermal imaging and AI for near-real-time detection — a live funding channel for exactly this program.

7. Key Technical Design Problems (deep dives)

7.1 Detecting a small fire under canopy

The hardest sensing problem. A smoldering lightning holdover under closed canopy has: little visible smoke initially, thermal signature occluded by crown cover, and sub-pixel size for satellites. Approach:

• Fly lower (100–120 m AGL BVLOS ceiling is actually good here — small ground sample distance: a 640-px LWIR with 32° HFOV at 120 m gives ~9 cm/px).
• Off-nadir look angles catch canopy gaps; plan crossing grid patterns.
• Detection = radiometric anomaly (ΔT vs local background, not absolute threshold — sun-heated rock defeats absolute thresholds) + persistence across frames + EO smoke segmentation as independent channel.
• Train/validate the smoke model on boreal-specific imagery (public wildfire smoke datasets exist but are biased to US fuels/terrain; plan a data collection season).
• False-positive economics: every false detection can trigger a costly patrol or crew dispatch. Design the reporting UI around confidence + evidence (thumbnail chips, revisit-confirmation pass) rather than binary alarms.

7.2 Geolocation accuracy without ground control

Hotspot coordinates must be crew-navigable (< 5 m). Chain: PPK GNSS (cm-level trajectory) + hardware camera sync (µs timestamping) + gimbal encoder angles + terrain model intersection. Radiometric LWIR optics have distortion; calibrate. Budget error stack and verify against surveyed heat sources (charcoal beds) in testing. This is a solvable engineering problem but only if designed in — it’s the difference between a science demo and an operational tool.

7.3 C2 and data over no-infrastructure terrain

• ≤ 40 km from GCS: long-range RF (900 MHz mesh; terrain-shadow planning with viewshed analysis in mission planner).
• Beyond: satellite onboard (Starlink-class flat panel on Tier 1; mass/power budget ~1.5 kg / 40–60 W — significant but feasible on a 25 kg VTOL).
• Design rule: the aircraft must complete its mission autonomously with ZERO link — links are for tasking updates and detection reporting, never for flight-critical control. Lost-link trees rehearsed and geofenced.

7.4 Cold/hot weather energy

• Li-ion packs: heated enclosures (self-heat pre-takeoff), insulation, state-of -charge derating below 0 °C; charging doctrine off generators with proper BMS telemetry logged to fleet software.
• Summer: density altitude + high ambient derate multirotor hover margins; require ≥ 30% hover thrust margin at 35 °C for Tier 2.

7.5 Airworthiness of the flight stack

Over an incident with manned aircraft nearby, the FC must be boring and proven: PX4/ArduPilot with parameter discipline, redundant EKF sources, dual GNSS on Tier 1. The RH-FC1 Rust stack earns its way on via the R&D airframe → mop-up-only variant → (long-term) a certified-by-declaration pathway if it ever makes commercial sense. Keep the boundary honest: safety case first, pride of authorship second.

8. Build vs Buy Analysis

The defensible IP is the mission system: boreal-tuned detection, lightning- cluster tasking, radiometric products crews can navigate to, and an operator organization certified to fly it BVLOS. The airframes are commodities.

9. Program Roadmap

Phase 0 — Groundwork (3–4 months, low cost)

• Advanced RPAS cert; begin Level 1 Complex ground school; scope RPOC.
• Meet AFFES fire-science/innovation contacts + 1–2 Northern municipalities / First Nations FireSmart programs; find where a pilot project could attach.
• Buy one thermal-equipped enterprise drone; fly practice hotspot surveys on legal proxies (campfire beds, controlled burns with permits, biomass piles).
• Deliverable: capability demo package — sample thermal ortho + hotspot layer from a controlled burn, accuracy report.

Phase 1 — Mop-up service (fire season 1)

• Tier 2 v1 (bought platform + custom pipeline): PPK georef, hotspot autodetect, standard product formats. Target: contracted/piloted mop-up surveys on real incidents in the mop-up phase (lowest airspace risk — fire is mostly out, manned traffic light).
• Measure: hotspots found vs ground-truth crew sweep; hours saved.

Phase 2 — Night mapping (season 2)

• SFOC + incident-integration procedures; night ops training; X8 mapper build.
• Fly night IR perimeters on assigned incidents during manned-aviation stand-down windows; deliver morning perimeter products to incident command.

Phase 3 — BVLOS detection pilot (season 2–3)

• RPOC + L1C pilots + PVD Tier 1 platform; integrate payload + Starlink + edge detection; lightning-tasking software.
• Pilot: one patrol corridor near a value-protection zone; measure detection latency vs existing methods; publish results (credibility flywheel).

Phase 4 — Scale decisions (season 3+)

• Expand corridors / add Tier 3 cargo / ignition partnership / interceptor research (M7) depending on Phase 1–3 evidence and funding.

Cost order-of-magnitude (CAD)

10. Risk Register

11. What This Program Is, In One Paragraph

A Northern-Ontario-specific uncrewed sensing service that gives fire managers three things they don’t reliably have today — night perimeters, fast post-lightning detection, and cheap mop-up verification — built on commodity airframes, differentiated by a boreal-tuned radiometric detection stack and crew-navigable data products, operated by a certified BVLOS organization that lives inside the fire agency’s airspace rules rather than around them. Suppression is a research payload, not the pitch. The 2025–26 regulatory changes made the flagship mission (remote BVLOS detection) routinely legal for the first time; the window to build operating history is now.