Use a drone carrying a thermal-infrared camera to scan the peat surface for warmer-than-surroundings patches. Heat from a hidden smoldering fire can move upward and create a thermal anomaly even when an ordinary visible-light camera shows no flame or smoke. The anomaly is a lead for investigation—not, by itself, proof of the fire’s depth, size, or exact location.
Controlled studies have demonstrated this method, but they do not establish one universal altitude, sensor specification, accuracy figure, or real-time workflow for every peatland. A defensible operation combines thermal surveying, careful geolocation, offline or validated image analysis, and on-site confirmation.
As an Amazon Associate I earn from qualifying purchases.
What a drone is actually detecting
Underground peat fires are usually smoldering fronts beneath the visible surface. A thermal camera does not see through peat like an X-ray. Instead, it measures infrared radiation from the surface. Where subsurface heat reaches the ground, the surface can become warmer than adjacent peat and vegetation. That temperature contrast appears as a bright or otherwise anomalous region in a thermal image.
This distinction matters. A thermal hotspot indicates an unusual surface-temperature pattern associated with possible subsurface burning; it does not directly show the underground fire front. A cool, deep, weak, or heavily insulated fire may produce little detectable contrast, while a sun-warmed patch or another heat source can produce a false lead. Treat every aerial detection as a location to assess with additional evidence.
#1 Best Overall
- 🆕【 640x512@30 FPS】Autel EVO Lite 640T Enterprise has same t~h.ermal i.m.aging function as Autel EVO 2 dual 640t V3/Enterprise V3/RTK V3, but more cost-effective. It has a resolution of up to 640x512@30 fps. It also comes with a 9mm lens, 16X digital zoom, providing greater resolution than any c.a.mera in its class, enabling p.ilots to see deep detail in dark environments. 📣For ass.istance with packaging, accessories, usage, firmware updates, or any other issues, please rea.ch our te.ch su.pport team at 💌autelofficial @ outlook . com💌
- 🏆【1/2 Inch 48MP Visible Light C~a~mera】The 1/2 inch CMOS sensor and effective 48 million pixels, using to capture i.m.ages with a larger field of view within a relatively short shooting distance. It is equipped with a 3-axis stabilized gimbals with a high-precision power motor structure, which can ensure stable video recording when autel lite 640T is flying.
- 🏆【2026 AUTEL ENTERPRISE APP】The Autel Enterprise App goes beyond the Autel Explorer and Autel Sky apps with advanced features. AI target recognition and positioning enables the EVO Lite 640T enterprise to automatically detect and locate objects like people, vehicles, and boats—plotting them directly on the map. Designed for efficiency, the app offers a streamlined interface with customizable toolbars and integrates smart professional tools to maximize the industrial potential of your Autel EVO Lite Enterprise Series.
- 🏆【SKYLINK 2.0 & AUTEL SMART CONTROLLER SE V2 】Autel’s SkyLink 2.0 transmission system delivers enhanced stability and speed, supporting HD video up to 12 km. It automatically selects the optimal channel across 2.4GHz, 5.2GHz, and 5.8GHz bands for strong anti-interference performance. Included with the EVO Lite 640T Enterprise is the 6-inch Smart Controller SE V2, featuring 800 nits max brightness and a 1440×720 resolution display.
- 🏆【EXCELLENT PERFORMANCE】Weighing just 866g, the Autel EVO Lite 640T Enterprise is compact yet powerful. It delivers exceptional performance across diverse scenarios, enhancing efficiency for solo operations. Main features include: 40 min max fly time, 17m/s max speed, 12m/s wind resistance, 4000m max takeoff altitude.
The 2019 pilot study, “Thermal-Drones as a Safe and Reliable Method for Detecting Subterranean Peat Fires”, states: “In this pilot study we have shown that subterranean peat fires can be detected using a drone equipped with a thermal infrared camera.” The authors also describe the work as a pilot and call for further investigation in real-world detection and firefighting scenarios.
What the published demonstrations show
The strongest evidence comes from two studies conducted near Palangkaraya, Central Kalimantan, Indonesia. Both used controlled fires in degraded peatland, so their results establish feasibility under studied conditions rather than guaranteed performance in every peat type, climate, canopy condition, or fire depth.
| Study | Equipment and data | Flight conditions | What was demonstrated |
|---|---|---|---|
| 2019 controlled pilot | Drone-mounted thermal-infrared camera; visible and covered test fires; a simple temperature-threshold analysis and visual interpretation | Experimental detections between 20 and 100 m above ground level (AGL); modelling examined a possible range up to 250 m AGL under stated assumptions | Open and covered subterranean fires could be detected in thermal imagery |
| 2025 automated detection and geolocation study | DJI Mavic Dual Enterprise with a FLIR Lepton thermal sensor (native 160 × 120 pixels) and an RGB camera; 1,800 images from ten controlled experiments | Flights rose from 5 to 100 m AGL | Deep-learning segmentation and image-to-GNSS geolocation were demonstrated, with offline processing |
The 2019 altitude findings should not be read as a universal operating envelope. The paper’s synthetic observations considered detectability under its own camera, fire and modelling assumptions; a stated possible range to 250 m is not an operational guarantee. Likewise, the 2025 flights rising from 5 to 100 m describe that experiment, not the height that will work best at another site.
How the 2025 model performed
Odusina, Perkasa, Chalmers, Fergus, Longmore and Wich reported their results in Drones on 25 June 2025 in “Detection and Geolocation of Peat Fires Using Thermal Infrared Cameras on Drones.” Their proposed model used a ResNet-50 encoder with a U-Net-like segmentation architecture. On the paper’s unseen test set, it achieved 87.6% intersection over union (IoU) and 94.2% precision. A MobileNetV2 comparison model achieved 57.9% IoU under the same study conditions. On just over 350 images of real open peat fires, the paper reported 90% IoU.
| Model or evaluation | Dataset described in the paper | Reported result |
|---|---|---|
| Proposed ResNet-50/U-Net-like model | Unseen test set from the controlled experiments | 87.6% IoU |
| MobileNetV2 comparison | Same experimental evaluation context | 57.9% IoU |
| Proposed model | Unseen test-set evaluation | 94.2% precision |
| Proposed model | Just over 350 images of real open peat fires | 90% IoU |
IoU measures overlap between a predicted fire region and a labelled region; precision describes how often predicted fire pixels were correct in that evaluation. Neither metric is a universal probability that a hotspot at your site is a peat fire. Different peat, vegetation, weather, camera optics, flight heights and labelling practices can change performance substantially.
Choose the sensing system around the mission
Make thermal infrared the primary sensor
A visible-light camera alone is not the method demonstrated in these studies. Select a drone system with a thermal sensor capable of resolving surface-temperature patterns at the intended height. An RGB camera remains useful for visual context—roads, vegetation, smoke, drainage features and landmarks—but it does not replace thermal data for a covered fire.
Rank #2
- High-Resolution Thermal Imaging: This FPV thermal camera module is available in multiple resolutions (256, 384, 640) to deliver clear and detailed thermal vision, allowing for precise identification of heat signatures during flight.
- Compact 9mm Lens Design: Equipped with a compact 9mm lens, this micro imager offers a great balance between field of view and detection range, making it ideal for integration onto various FPV drone platforms.
- Standard CVBS Video Output: Features a CVBS interface, ensuring broad compatibility with a wide range of standard FPV video transmitters and displays commonly used in the drone ecosystem.
- Micro Size & Lightweight Build: The miniature and lightweight design minimizes the impact on your drone's weight, balance, and agility, facilitating easier installation and stable flight performance.
- Dedicated for FPV Drones: Specifically engineered for First-Person View (FPV) drone applications, this camera assists in scenarios such as search and rescue, infrastructure inspection, and nocturnal observation by visualizing thermal data.
The 2025 study used a thermal sensor with native 160 × 120-pixel resolution. A display may enlarge those images, but enlargement does not create additional ground detail. Ask for the native detector resolution, lens field of view and the ground footprint at each planned altitude.
Recommended Free Tools
Balance detail against area covered
At greater altitude, each pixel covers more ground. If a hot patch occupies only part of a pixel, its temperature signal is blended with the surrounding background. This is why a small or distant fire can become harder to distinguish as the drone climbs, even though a high flight covers more area in each image.
The practical choice is therefore a mission-specific trade-off: fly low enough for the suspected target to occupy useful pixels, while retaining safe clearance and enough coverage to search the area. The cited studies do not provide one universally optimal height or survey speed.
Account for endurance and payload
The 2025 paper notes that multirotor flights lasting roughly 20–45 minutes can limit the area covered in one mission. The authors suggest fixed-wing platforms as a possible wide-area alternative, but their peat-fire study did not provide a head-to-head evaluation of fixed-wing and multirotor systems. A fixed-wing aircraft also has different launch, landing, turning and low-speed imaging requirements, so platform selection must follow the terrain and payload rather than endurance alone.
Step-by-step field workflow
1. Define the search area and the decision you need to make
Mark the peatland sector to be screened and decide what a detection should trigger: a second, lower-altitude pass; a ground team’s inspection; a fire-management action; or a request for specialist assistance. This prevents a map full of unprioritized thermal anomalies from being mistaken for a completed diagnosis.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →2. Verify the aircraft, payload and operating permissions
- Use a thermal infrared camera; do not describe an RGB-only drone as a peat-fire detector.
- Confirm that the aircraft can safely carry and power the thermal payload for the planned mission.
- Check local flight rules, airspace constraints, land access and any incident-command requirements before launch.
- Review weather, wind, gusts, visibility, wildlife or bird activity, season, time of day, pilot skill and the data-processing setup. A 2024 Ukrainian-language study identifies these factors as relevant to unmanned detection of hidden ignition points and notes that wind and gusts can make control more difficult. See “Features and conditions of application of unmanned technologies for detecting hidden focuses of ignition in peat soils.”
3. Plan altitude from the target size and pixel footprint
Use the suspected fire size, sensor resolution and lens geometry to choose a starting height. The 2019 pilot reported detections from 20 to 100 m AGL, while the 2025 experiments climbed from 5 to 100 m. These values provide study context, not a rule that every operation should copy. If a suspected anomaly is small, a lower, safer pass may provide more useful pixels than a high overview.
Rank #3
- 🏆【AUTEL ROBOTICS USA AUTHORIZED】We are the longest and largest distributors of Autel Robotics. Over the years, we have maintained a close partnership with Autel, providing customers with high-quality products and excellent after-sales service. With extensive experience and strong technical expertise in this industry, we are able to offer customized solutions and professional technical support to our customers. All the UAVs we offer include FULL USA WA-RRANTY. 📣For ass.istance with packaging, accessories, usage, firmware updates, warranty repair or met any issue, please rea.ch our su.pport team at 💌autelofficial @ outlook . com💌
- 【THERMAL CAMERA - A NEW GENERATION OF THERMAL SENSOR】Compared to traditional inspection, thermal imaging can greatly improve efficiency, and reduce inspection time and c.osts. Autel EVO II Dual 640T V3 thermal imaging sensor has a resolution of up to 640x512@25 fps. It also comes standard with a 13mm lens, 16x digital zoom, and is capable of recognizing subjects at up to 100 m, provides greater resolution than any camera in its class, enabling pilots to see deep detail in dark environments.
- 【VISIBLE CAMERA - CAPTURE AND PRESERVE EVEN THE SMALLEST OF DETAILS】Autel 640T V3 drone equipped with a 4K visible-light camera with 84° wide FOV, 4 times lossless zoom, 16 times digital zoom and lots of amazing digital camera performance, EVO II Dual 640T V3 helps you capture every detail to assist in precise evidence capture.
- 【D-RI RANGES & 10+ TEMP MEASUREMENT MODES】Autel Evo ii dual 640T v3 thermal drone for hunting is packed with D-RI ranges function: 1)Detection: Detect if an object is present. 2)Recognition: See what type of object it is (i.e. person, vehicle, airplane).3)Identification: Determine subject affiliation and intent, helps you see a specific subject's identity, activity, or personal items. It also supports up to 10 thermal palettes/ central, spot, regional temperature measurement/ picture-in-picture display/ temperature alarm with isotherm and image enhancement/ f.ree infrared thermal analysis tool, enabling pilots to cope with a variety of complex shooting environments and mission scenarios.
- 【PRACTICAL AND CONVENIENT FLYING PERFORMANCE】The EVO II Dual 640T is equipped with 19 groups of sensors including 12 visual sensors, the main camera, ultrasound, IMUs and other sensors enable building of three-dimensional maps and path planning in real time. 38Mins flight time and No Fly Zone function give the pilot max flying possibilities.
4. Capture thermal images with flight context
Collect imagery at the planned heights and record the aircraft position, altitude, time, sensor configuration and flight conditions. The studies used imagery from multiple heights above experimental fires. Context allows analysts to distinguish a persistent spatial pattern from a one-frame artefact and to understand how the apparent size of a hotspot changes with altitude.
5. Inspect the thermal data for coherent anomalies
Begin with a manual review or a simple temperature-threshold method. The 2019 pilot detected fires by eye and with a rudimentary threshold algorithm. Look for a warmer region that is spatially coherent and related to the surrounding terrain, then compare it with the RGB view and nearby images. A threshold is a screening aid, not a universal fire-temperature cutoff: the surface can be affected by sun, moisture, vegetation, ash and other heat sources.
6. Apply automated segmentation only within its validated scope
The 2025 workflow used semi-automated annotations and deep-learning segmentation. Such a model can outline candidate fire regions consistently across many images, but its published scores come from the authors’ datasets. If you deploy a model elsewhere, validate it on imagery from that peatland, sensor and season before using its output to drive urgent decisions.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Automation in the 2025 paper was offline. The authors write: “Currently, model inference is conducted offline, meaning that the collected data must be downloaded and processed after the drone flight.” That delay can matter when crews need a rapid location. Do not present this implementation as a live, in-flight warning system.
7. Convert detections to map locations and preserve the uncertainty
The 2025 study describes mapping detected image regions to GNSS locations and provides a web application for offline image analysis. Use the resulting coordinates to guide a follow-up team, but keep the error budget visible. The authors did not model uncertainty from camera calibration or vehicle positioning. A coordinate should therefore be treated as an operational estimate, not a survey-grade boundary of the underground fire.
8. Verify from the ground and update the map
Send trained personnel to inspect prioritized anomalies using the procedures and protective equipment required for peat-fire work. Record which hotspots were confirmed, rejected or inaccessible, then use those observations to improve later thresholds or model validation. Aerial imagery helps locate possible fires; the cited studies do not show that an image classification alone establishes fire extent or replaces incident assessment.
Rank #4
- 👍【2026 NEW ARRIVAL, AUTEL EVO LITE 640T ENTERPRISE】Ready to elevate your mission capabilities? Weighing just 866g and featuring a compact foldable design (210×123×95mm), the Autel Robotics EVO Lite 640T Enterprise is easy to carry in any backpack.With a simple operation, it is ideal for efficient execution of missions by a single operator. 🩷Friendly NOTE: Autel Lite 640T Enterprise supports Remote ID. 📩 Need help? R.each out to us at: 📩📩AutelStoreOnline @ outlook . com📩📩
- 👍【640×512 @ 30Hz T~H.ERMAL】Autel Robotics EVO Lite 640T Enterprise comes with a high-resolution sensor that captures data at 640×512 and 30 Hz. It comes with a standard 9mm lens and offers 16x digital zoom, providing exceptional i~m.age clarity in its category. Capable of m-easuring t~emperatures from -20°C to 550°C with an accuracy of ±3°C, it ensures reliable detection of h~eat a.nomalies—even during critical operations.
- 👍【AI TARGET RECOGNITION】 Our AI-powered target recognition system offers the following capabilities:✅ 1) Self-learning system that improves object detection over time; ✅2) Enhanced recognition with wide-angle or infrared lenses; ✅3) Real-time projection of identified targets onto a digital map with support for 100+ objects, streamlining tasks such as search and rescue or su.rveillance operations.
- 👍【Long-Range 6K Transmission & SKYLINK 2.0】With Autel’s SkyLink 2.0 transmission system, Lite 640T enterprise provides 6K real-time video feed with a range of up to 12 km (FCC). Its 40-minute max fli.ght time ensures extended mission coverage without frequent battery swaps.
- 👍【What you will get ?】1* Autel EVO Lite 640T Enerprise (1*Battery, 2*Pair Propellers,1*Gimbal Holder), 1* Remote Controller SE V2, 1 x Power Adapter, 1 x USB-C to USB-A Data Cable, 2 x User Documentation.
How to interpret a thermal hotspot
Strong leads
- A warm region persists across adjacent frames while the aircraft moves.
- The anomaly has a shape and location consistent with a surface heat source rather than a transient sensor artefact.
- The thermal signal is supported by visual context such as smoke, scorched vegetation, cracks or other site evidence, while recognizing that none of these is required for a subsurface fire.
- A second pass at a different height shows the feature in a consistent ground location.
Reasons for caution
- A small fire can be diluted within a large ground pixel at higher altitude.
- Sun-heated soil, exposed dark surfaces, machinery, buildings or other sources can mimic a warm patch.
- Vegetation, peat moisture, fire depth, wind and changing weather can alter how heat reaches the surface.
- A model trained near Palangkaraya may not transfer directly to another peatland, sensor or season.
Use these observations to rank follow-up priorities, not to assign a false certainty. The most useful output is a map of candidate locations with acquisition details and an explicit note about positioning and classification limitations.
Geolocation: useful coordinates, incomplete certainty
Image-to-GNSS mapping can turn a highlighted thermal region into a point or polygon that a response team can navigate toward. Its practical value depends on the aircraft’s recorded position, camera geometry, altitude and the way the image footprint is projected onto the ground.
The 2025 method did not account for camera-calibration or vehicle-positioning uncertainty. That omission does not make the map useless; it means the map should not be read as an exact fire perimeter. Preserve the original image, the projected detection, the aircraft track and the assumptions used to create the coordinate. If a crew is approaching a suspected fire, provide an area of interest rather than implying centimetre-level accuracy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Weather, timing and flight-operation limits
Thermal contrast is not controlled by the fire alone. Time of day, recent sunlight, cloud, wind, humidity, vegetation and peat moisture can change the surface-temperature background. Wind and gusts also affect aircraft control and image sharpness. The 2024 study’s list of relevant factors includes platform and payload, weather, bird activity, season, time of day, pilot skill, air situation and data-processing systems.
Plan a repeatable acquisition window when possible, and document conditions with each flight. A single thermal frame taken under one set of conditions should not be compared casually with an image from a different time or sensor setting. If the operational question is whether a hotspot is persistent, repeat the survey under comparable conditions rather than relying on a visually dramatic but isolated frame.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow to compare candidate drone systems
The available evidence supports comparing systems by mission characteristics, not declaring one universal winner.
Best Value
- No Registration Needed - Under 249 g, this drone with camera for adults 4K does not require FAA registration or Remote ID for recreational use. Visit the FAA's official website for requirements related to drone registration and Remote ID. [1]
- 4K Ultra HD & 3-Axis Gimbal for Cinematic Quality Shooting - Capture stunning moments in any light—sunrises, sunsets, and night scenes with crystal-clear 4K resolution. Thanks to the 3-axis gimbal, you can achieve cinematic-quality footage with this advanced drone with camera.
- 38kph (Level 5) Wind Resistant - This drone for adults has a stable flight even in Level 5 winds. Brushless motors enhance power and allow takeoff at altitudes up to 4,000 meters.
- 10km Max HD Video Transmission-up to 10 km (32,800 feet) [2] of HD video transmission and has excellent anti-interference capabilities, giving you the ability to freely explore vast landscapes and see clearer.
- Due to platform compatibility issue, the DJI Fly app has been removed from Google Play. To ensure a better product usage experience, please log in to the DJI official website to download the latest version of DJI Fly.
| Comparison axis | Questions to ask | Why it affects peat-fire detection |
|---|---|---|
| Thermal resolution and optics | What is the native detector resolution, lens field of view and ground footprint at the planned heights? | Small hot areas can be diluted when they occupy only part of a pixel. |
| Coverage per flight | How much area can be safely surveyed before battery, weather or operating limits end the mission? | The 2025 paper notes approximately 20–45-minute multirotor durations as a coverage constraint. |
| Processing workflow | Are detections reviewed manually, processed after landing, or generated onboard? | Offline processing can delay response; the 2025 implementation was offline. |
| Geolocation quality | How are image regions projected to coordinates, and are camera and vehicle-position errors quantified? | Coordinates guide crews, but unmodelled errors can shift a mapped anomaly. |
| Operational suitability | Can the team operate safely in the expected wind, weather, wildlife, airspace and terrain conditions? | Flight stability and legal access are prerequisites for useful imagery. |
The 2025 study’s DJI Mavic Dual Enterprise and FLIR Lepton configuration is an example of a research setup, not an endorsement or a claim about current retail availability. The defensible equipment description for this use is a thermal camera drone whose sensor, optics, endurance and data workflow match the mission.
Common failure modes and corrective actions
The drone sees no obvious hotspot
Do not conclude that no underground fire exists. Recheck altitude and pixel footprint, inspect the weather and surface conditions, and consider whether the fire is too deep or weak to produce a measurable surface contrast. A lower, carefully controlled pass may resolve a target that was diluted at height, provided it is safe and permitted.
Every dark surface appears hot
Review the RGB context and compare adjacent frames. Sun-heated material and other non-fire sources can create anomalies. Use a threshold as a screening tool, then require spatial consistency and follow-up evidence before escalating.
The model reports many detections
Check whether the model was validated on imagery from the same sensor, peat type, season and operating conditions. The published IoU and precision figures are bounded by the 2025 study’s datasets. Re-run a manual review and prioritize detections that persist across frames or passes.
The map point does not match the visible feature
Inspect altitude, aircraft position, camera orientation and projection assumptions. The 2025 method did not include camera-calibration and vehicle-position uncertainty, so treat the coordinate as an estimate and search the surrounding area.
The team expects a live alarm
Set expectations before flight. The described 2025 workflow downloaded imagery and processed it after the drone flight. If rapid response is essential, the operation needs a separately validated onboard or near-real-time processing system; the cited study does not establish one.
Field checklist
- Thermal infrared payload selected; RGB used for context rather than as a substitute.
- Native thermal resolution and ground pixel footprint understood for every planned height.
- Search polygon, response threshold and follow-up responsibility agreed before launch.
- Weather, wind, gusts, season, time of day, wildlife activity and airspace reviewed.
- Aircraft position, altitude, time and sensor settings recorded with each image set.
- Thermal anomalies checked across neighboring frames and against visible-light context.
- Automated segmentation validated for the local sensor and environment, or clearly labelled as unvalidated.
- Offline processing time communicated to the response team.
- Geolocation outputs accompanied by their calibration and positioning limitations.
- Suspected hotspots passed to qualified personnel for ground verification and incident decisions.
What this method can—and cannot—promise
Drone thermal mapping can expose surface-temperature patterns associated with hidden peat fires without sending a person into every part of a hazardous or inaccessible area. The 2019 pilot demonstrated visual and threshold-based detection, and the 2025 study showed that segmentation and approximate geolocation can organize a larger image set.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Those results do not establish universal field accuracy, guaranteed detection at a particular altitude, real-time inference, precise fire perimeters or transfer to every peatland. The most reliable deployment treats the drone as a rapid reconnaissance and mapping tool: thermal imagery finds candidate areas, analysis ranks them, geolocation guides crews, and qualified follow-up confirms what is burning.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
