Bushfire or Wildfire detection using Home Assistant

Periodically, we get requests for interesting projects that do not fit the normal definition of a fire safety system as described in the National Construction Code or Australian Standards.

A little while ago, we received one of these requests, and it sparked my interest.  The request was for an inexpensive sensor necessary to detect ember attack in a bushfire (wildfire) prone area.  While I am not specifically familiar with this type of sensor for home use, it did get me thinking!

Traditional fire detection systems and smoke alarm systems are used indoors, and unlikely to operate reliably (without nuisance activations) outside, so I had to think laterally.

Before I get into my thoughts on a possible, inexpensive solution, I should explain what an ember attack is.

An ember attack in a bushfire is when strong winds carry burning pieces of plant material (embers) ahead of the main fire.  These embers can land on or around your home, potentially starting new fires well away from the main fire front. The risk is that these embers can get into small gaps, like vents and gutters, and ignite flammable materials such as leaf litter, making ember attack a major cause of house fires in bushfire-prone areas.

In our connected world, there are many ways to ingest (receive) an alert from the local weather service, such as the Bureau of Meteorology here in Australia.  These services often provide free data access to their data via an API or Application Programming Interface.  The API will provide information like temperature, humidity, solar radiation, wind speed, wind direction, rain, warnings and other real-time and forecast information.

Another option is to purchase a relatively inexpensive local weather station that can be interfaced locally with a platform like Home Assistant.  I personally use an Ecowitt HP2553 Wi-Fi Weather Station. This system provides an incredible amount of real-time data including temperature, humidity, solar radiation, UV index, rainfall, wind speed, wind direction and atmospheric pressure.
The weather data provided by the Government, commercial weather data provider or your own local weather station can be used to help predict days and times when the conditions for a fire or ember attack are more likely.  This can then be combined with other information as part of some conditional logic to trigger some action.  For example this data might be helpful from someone living in a bushfire prone area in Victoria;

INPUTS (WHEN THIS HAPPENS)

• Temperature: ≥ 35`C
• Humidity: ≤ 40%
• Wind Speed: >≥ 35 kp/h
• Wind Direction: NE, N or NW
• Barometric Pressure: Increases by 2% in less than 15 minutes

OUTPUTS (THEN DO THIS)

• Action: Send a Notification “Possible fire conditions, take additional precautions”

In this example, we are drawing on data (inputs) to make a conditional action that informs the receiver of a notification of conditions similar to those that may indicate the possibility of ember attack in the event of a fire.  Of course you should develop and test your own data that is indicative of the precursor conditions of a fire in your area.

In Australia we also use a regional fire danger rating system, where the data is publicly accessible.  The rating system is a blunt instrument insofar as it is broad in its categorization of the fire danger and location and is based on four levels of preparedness;

• ◾ Moderate (Green) - Plan & Prepare;
• ◾ High (Yellow) - Be ready to act;
• ◾ Extreme (Orange) - Take action now to protect life and property;
• ◾ Catastrophic (Red) - For your survival, leave bushfire risk areas.

Combined with other information (inputs) this system could be very useful as part of a notification system for people planning their response to fire danger.  In addition to other inputs such as weather, warnings and alerts, the fire danger rating could form part of an integrated early warning system.  An example might look like this;

INPUTS (WHEN THIS HAPPENS)

• Temperature: ≥ 35`C
• Humidity: ≤ 40%
• Wind Speed: ≥ 35 kp/h
• Wind Direction: NE, N or NW
• A fire has been detected within 5km north of my location
• Fire Danger Rating ≥ Extreme

OUTPUTS (THEN DO THIS)

• Action: Send a Notification “A fire has been detected in your area, leave immediately!”

In this example, we are drawing on three inputs; weather, local alerts and the fire danger rating. Once again, by increasing the information available we can improve the model, helping us to predict the possibility of imminent ember attack.

Many of the following IoT sensor devices are inexpensive and easily accessible online or from your local big-box hardware store. They connect with a platform like Home Assistant via services like WIFI, Bluetooth and radio frequencies, typically on the 433MHz band.

There are also ”Smart Home” protocols that overlay on top of things like WIFI and Ethernet such as Matter, developed by Apple, Google, Amazon, and others to enable the easy integration of these devices.

So far we have covered publicly available data sources, but with platforms like Home Assistant, we can draw on an extensive array of IoT sensors like temperature, humidity, moisture, illumination, carbon monoxide, and airborne particulates (PM1, PM25, PM10), time, location, proximity and manual inputs. We can also combine data from these sensors or use the change in sensor data over time as part of our prediction model.

• Temperature: While temperature by itself may not be useful, when combined with other data or looked at over time could be useful in refining our model.  Compatible temperature sensors can be purchased inexpensively (Less than $25 AUD) either online or at some specialty stores.
• Humidity: Like temperature may not be useful by itself, however the periods of low humidity, combined with other factors are a useful tool as part of our predictive model.  More often than not, a humidity sensor is combined with a temperature sensor in the same device.
• Barometric (Atmospheric) Pressure: Based on my research, there is a correlation between a quick increase of barometric pressure and the incidence of a storm.  This data could be useful to help predict lightning, increase wind speed, change in wind direction or other phenomena that may be useful in developing a predicative model.
• Moisture: The data from a soil moisture sensor could be used as part of a predictive model, although it might be better to get a range of data from multiple sensors.  The measurement of soil moisture is typically shown as a percentage (%) where a low number indicates dryness and a high number indicates a greater level of soil saturation.
• Illumination: The level of illumination (light), measured in lux (lx) may be useful in one of two ways (1) to indicate the presence of clouds (or smoke) or the time of day. This data may be used as part of the conditional logic based on other criteria, to determine if an action should occur.  For example you may wish to have different outputs based on the light level (day or night). Light sensors that can be interfaced with a platform like Home Assistant are readily available and may be purchased for less than $30.00 AUD.
• Carbon Monoxide: Carbon monoxide measured in parts per million (ppm) and is one of the by-products of combustion. Carbon monoxide can be an extremely useful sensor in our model, particularly when used as a comparative sensor over time and along with other input signals.  For example, typical atmospheric carbon monoxide (CO) levels typically range from 100 ppb (parts per billion) to 10 ppm (parts per million). Conditional logic could be developed to monitor normal levels of carbon monoxide over time, and where the level increases an alert signal could be generated.  The level of carbon monoxide may also form part of our predictive model.

• Airborne Particulates: Airborne particulates refers to visible and invisible airborne particles that could indicate the presence of smoke.  These particulates are generally measured according to their size as follows;

  • PM1: Ultrafine Particles - Such as VOCs, ultra-fine contaminants, viruses, bacteria;
  • PM25: Fine Particles - Such as dust, pollen, mould, pollution;
  • PM10: Course Particles - Such as combustion (smoke) and pollution particles. PM10 Particulates are visible to the human eye and are present in a bushfire.

  An inexpensive indoor sensor that can detect both PM25 and PM10 particulates can be purchased for a moderate price for less than $100.00 AUD.  A photo-electric smoke detector is a type of PM25/PM10 sensor that uses the principle of light scattering to deliver a binary (on/off) signal when smoke is detected.

• Time: While not immediately evident, time is a useful sensor in our arsenal. Time can be used for a range of things such as determining the relative position of the sun based on our location, or in combination with other sensor data as the basis for comparison.
• Location: Location can be derived from multiple sources such as town, postcode, IP address, or more precisely latitude and longitude. Location is extremely useful when combined with other data as part of our conditional logic.  The more accurate the location, the better it is for our predicative model.  If you don’t own a GPS, you can estimate your location using Google maps.
• Proximity: For the purposes of this article, we are using proximity as a measure of the distance of a person (i.e. their location) relative to some other input, for example the location of a fire.  When combined with a compatible smart phone or smart watch, proximity may be used in a multitude of ways to help prevent people from moving closer to a fire, or as an alert of an individual who is in danger.
• Manual Inputs: While not fully automated, a manual input such as “binary input”, “scale” or text could form part of the warning, action or response system.  For example, a trigger input could be used to initiate an alert, activate a warning system or turn on a fire pump and sprinkler system.

There are so many inexpensive sensors and inputs available to a platform like Home Assistant, used in combination with conditional logic, it’s possible to develop a model to predict the risk or onset of a fire in your local area.  While it may not be up to a commercial system, it could be useful enough to heighten your attention with sufficient time to escape safely.

An infrared (IR) sensor for fire detection works by detecting the electromagnetic radiation emitted by a fire. All objects emit IR radiation, but fire produces a distinct intensity and pattern in specific IR wavelengths. An infrared fire sensor receives this IR energy, causing a change in its electrical properties. This change is then converted into an electrical signal corresponding to the intensity of the IR radiation and the potential of fire.

Infrared radiation exists beyond the red end of the visible light spectrum, meaning our eyes cannot see it. Just like visible light has different colors corresponding to different wavelengths, IR also has a range of wavelengths. Fire emits strongly in certain mid- and far-infrared bands. Selecting the correct sensor that uses filters to be sensitive to these specific wavelengths of IR will help it distinguish fire from other heat sources that might emit different IR with a different signature (wavelength).

At the time of writing this article, and with respect to an interface to Home Assistant there were some hobbyist infrared sensors that could be used, but they should be considered a hobbyist solution at best and should not be used as a primary method of fire detection.  The intended way these sensors work is the sensor circuitry analyzes the received signal, looking for characteristics like intensity and rapid fluctuations (flickering flames). If the signal matches the expected pattern of a fire, the sensor could trigger an alarm output. This output can be a simple digital signal (on/off) or an analog signal proportional to the detected IR intensity.

To integrate with Home Assistant, the sensor's output is typically connected to a microcontroller (like an ESP32). This microcontroller reads the sensor's state and communicates the signal to Home Assistant. Home Assistant can then use this information to automate alerts and actions.

Inexpensive video fire, smoke and ember detection is an emerging and very exciting type of sensor that combines a video signal processing (computer vision) with machine learning (ML) or artificial intelligence (AI) to identify smoke, fire and embers.  While this is slightly out of the reach of an inexperienced home enthusiast, it is not far away.

In very simple terms, these sensors work processing a video feed and then comparing that signal with parameters that appear like fire, smoke and embers.  The resultant signal is then returned with a probability score. This means that the result could be a low probability or a high probability of smoke or embers.  The speed and accuracy of this signal is determined by a range of factors such as distance, size of the fire, image quality, processing speed of the processing computer and the underlying machine learning model.

A quick search of the internet will reveal plenty of emerging examples (April 2025) of cameras and technologies that will evolve and become less expensive and more consumer (enthusiast) friendly over time.

If you're willing to invest the time, an experienced developer with some knowledge of machine learning, a powerful single-board computer or accelerator could develop a sensor for less than a few hundred dollars. There are already public Artificial Intelligence (AI) models available that an experienced person could use to develop a video sensor that could be interfaced with a platform like Home Assistant.  How this could be achieved is outside of the scope of this article.

I have included an extensive list of links to hardware, software and other resources at the bottom of this article, to give interested readers a starting point for further research.

The easiest way to get started with Home Assistant is to purchase a single-board computer called the “ Home Assistant Green” along with a USB dongle called the Home Assistant Connect ZBT-1  that adds support for integrating devices using accessories using Zigbee and Thread.  The Home Assistant Green ensures a fast and smooth Home Assistant user experience.

If you have a little more experience and twice the price, you could try the Home Assistant Yellow. The primary difference is the Home Assistant offers a few additional features (and complexity) that the Home Assistant Green does not offer.  But if you're just getting started, try the Home Assistant Green first. (I use a Home Assistant Yellow).

Lastly, you can also try a very small (low power) device called a Raspberry Pi.  The Raspberry Pi is a very small computer (slightly larger than a pack of playing cards). There are many reasons why you might choose this path, but if you're just getting started and you're not super computer savvy (and a little nerdy), then again start with the Home Assistant Green.