Environmental monitoring with IoT sensors

A room temperature sensor mounted above a radiator measures the warm plume from the radiator. It can read several degrees high for a whole heating season, and nothing in the data shows it: the radio delivers every report on time and the trend looks smooth. Most environmental monitoring faults are of this kind. The link works and the measurement does not represent the thing it is meant to.

Start with the decision the data supports. A comfort complaint, ventilation control, cold-chain protection, condensation risk and outdoor exposure each need a different variable, position and response time.

Point schedule

Write a point schedule before you choose hardware, with one row for each sensor. Two example rows:

The CO₂ figures are from the Sensirion SCD40 datasheet. The temperature figures are for the EpiSensor TES waterproof probe. Add columns for the owner, the calibration reference and a photograph.

The two columns people leave out are the question and the position. Without the question, nobody can decide later whether a 15-minute average is acceptable. Without the position and a photograph, nobody can see that the sensor sits in a supply jet.

Record the medium each sensor measures. Air, surface and liquid temperatures are different measurements with different probes. Relative humidity is measured. Dew point is calculated from temperature and humidity, so it carries the error of both.

An NDIR sensor measures CO₂ directly. An "equivalent CO₂" (eCO₂) value is inferred from a metal-oxide gas sensor's response to volatile organic compounds. It rises with cleaning products, cooking and perfume, follows occupancy only loosely, and cannot be compared with the HSE figure below. A TVOC reading from a metal-oxide sensor is a response to a gas mixture, scaled to a reference gas that each manufacturer chooses, so two brands report different numbers in the same room.

Use the specification of the finished instrument in its housing. Sensirion's SHT3x humidity element has a 63 % response time of 8 s at 1 m/s airflow. The same maker's SCD4x CO₂ module has a humidity element built in, and specifies 90 s for humidity. The housing and the airflow around it set the response.

The Device Directory's temperature and humidity and air quality sections list each third-party sensor's documented readings, units and ranges, with the manufacturer's sources.

Sensor placement

A sensor reports the conditions at its own position. Keep it out of supply-air jets, direct sun and doorways. Keep it away from radiators, equipment and the heat of its own electronics, unless one of those is the thing you measure.

For workplace CO₂, HSE recommends an NDIR monitor at head height, more than 50 cm from people, and away from windows, doors and air supply openings. HSE also asks for several readings through the day while the room is occupied. It says larger spaces usually need more than one sampling location. Try several positions with a portable monitor before you fix the permanent one.

Each common placement fault leaves a signature in the trend:

Carbon dioxide and ventilation

HSE says CO₂ consistently above 1,500 ppm in an occupied room indicates poor ventilation, and that you should act to improve it. That figure is a ventilation indicator, not a universal safety limit. The workplace exposure limit for CO₂ in HSE's EH40 is 5,000 ppm as an 8-hour average. It also gives a scale: 1,000 ppm is equivalent to about 10 litres of outdoor air per second per person. Outdoor air is now just over 420 ppm.

CO₂ tracks ventilation only where people are the main source and the air is mixed. HSE lists the cases where it does not work: rooms with air cleaning units, which remove contaminants but not CO₂; processes that produce CO₂; spaces with inconsistent occupancy; and large open areas where mixing is uncertain. ASHRAE's June 2026 brief adds that the CO₂ differential set points in Standard 62.1 are for demand-control ventilation, and are not indicators of indoor air quality.

Most low-cost NDIR modules correct their own drift with automatic baseline correction. By default, Sensirion's SCD4x assumes that it sees outdoor air at 400 ppm for more than 3 minutes in every week of operation, and uses the lowest reading as that baseline. A room that never ventilates down to outdoor level breaks the assumption: a 24-hour control room, a sealed store or a laboratory with a CO₂ source. The correction pulls the baseline up to the room's own minimum, and the sensor reads progressively low.

To detect this, look at the overnight or weekend minimum. In an office that is ventilated when empty, the minimum settles near outdoor level every week. In a space that never does, turn automatic correction off and recalibrate against a reference instrument instead.

Temperature: air, surface and immersed

Each medium needs its own probe and mounting. A probe in free air measures air, responds slowly in still air, and picks up radiation from warm or cold surfaces nearby. A pipe probe must sit flat on the pipe under the insulation, or it reads somewhere between the pipe and the room. A probe in a thermowell measures the fluid, but the well adds thermal mass and slows the response.

The EpiSensor TES probe temperature sensor uses a 10k NTC thermistor in every probe. The waterproof probes, on a short lead or a 1.5 m or 2.5 m cable, suit air and liquids in cold rooms, fridges and tanks. The steel pipe probe has a flattened stainless tip for flow and return pipes. Accuracy is ±0.2 °C across the normal range (−20 °C to 70 °C for the waterproof probe) and ±0.5 °C from −40 °C to 105 °C. A TES-22 reads two probes on separate channels, so one device can record the flow and return on the same pipe pair.

An air probe in a chiller sees every door opening. Some sites put the probe in a small bottle of glycol so that it follows the temperature of the stock instead of the air. If you do this, write it in the point schedule, because the same 8 °C alarm then means something different.

Process temperature points often use a Pt100 or Pt1000 resistance thermometer with a 4-20 mA transmitter. An IEC 60751 class B Pt100 is within ±0.3 °C at 0 °C, and the tolerance widens with temperature; the Pt100 and Pt1000 calculator gives each class. The transmitter connects through a ZIO analogue signal sensor. The analogue signal commissioning guide covers scaling and fault currents.

Humidity, dew point and condensation

Condensation and mould happen at surfaces. A room humidity reading answers only half the question. Calculate the dew point of the room air, then compare it with the temperature of the coldest surface.

A room at 20 °C and 60 % RH has a dew point of 12.0 °C:

A cold corner at 14 °C in that room is 2 K above the dew point, so no water forms there. The air against that surface is at about 88 % RH:

ISO 13788 uses a monthly mean surface humidity of 80 % as its design limit for mould growth. In this room, a surface below 15.4 °C exceeds that limit. The corner is at risk long before it is visibly wet.

To monitor this, put a contact probe on the cold spot and a temperature and humidity sensor in the room air. An Edge calculated measurement can then derive surface RH from the three readings with an expression. A ±2 % RH sensor error moves the dew point by about ±0.5 K at room conditions, so check the humidity sensor before you trust a margin of 1 K.

Capacitive humidity sensors shift after long periods of high humidity. Sensirion's SHT3x datasheet gives an offset of +3 % RH after 60 hours above 80 % RH, which recovers slowly once conditions return to normal. In a cold store, a wash-down area or an outdoor enclosure, expect readings to run high after wet periods.

Outdoor points

An outdoor point needs its own reason for its location. A sensor beside a road, an exhaust or a loading bay measures that source. A background point must be away from all of them. EPA's siting guide puts outdoor air sensors about 1 to 2 m (3 to 6 ft) above the ground, with free airflow, away from buildings, fences, vegetation, exhausts and dusty roads.

For air temperature, the Met Office measures at 1.25 m inside a Stevenson screen. The screen shields the thermometer from direct sun and rain, and air moves through its louvres. Without a ventilated radiation shield, a sensor in sun reads the temperature of its own housing.

Check the enclosure rating and the operating range against the site. The TES enclosure is IP67 and its probes are IP68, and it operates from −30 °C to 55 °C; the IP rating checker explains the codes. A correction fitted at one site does not transfer to another. Keep the original readings and the correction you applied.

Calibration and drift

A calibration record names the instrument, the reference or certificate, the range and conditions, the uncertainty and the date. Metrological traceability is an unbroken, documented chain of calibrations back to a reference. It tells you how well the value is known. You still decide whether that is good enough for the decision.

Two sensors side by side show relative offset, drift and lag. They can agree and both be wrong, so accuracy needs a traceable reference. For outdoor air sensors, EPA's siting guide describes collocation with a regulatory monitor, within 20 m of it. Practical field checks:

• Temperature probe: an ice-point bath of crushed ice and water, at 0 °C.
• CO₂: a calibrated reference instrument beside the sensor, or a reading outdoors against about 420 ppm.
• Humidity: a calibrated reference hygrometer, after both have settled at the same position.

Set the recheck interval from the drift in the datasheet and from the cost of a wrong value. Sensirion gives the SHT3x a typical humidity drift below 0.25 % RH per year. It gives the SCD4x an additional CO₂ drift of ±(5 ppm + 0.5 % of reading) per year after the first five years. Recheck after any move or change to the installation.

Response time, sampling and averaging

Four durations set what a trend can show:

1. the sensor's response time;
2. the local sampling interval;
3. the radio reporting interval; and
4. the display or alarm averaging interval.

The sensor's response time is the lower limit. Reporting faster does not show a change the sensor has not yet seen. An NDIR CO₂ module with a 63 % response time of 60 s gains little from reports every 10 s.

Worked example: a TES probe in a chiller is set to report every 15 minutes, the interval behind its battery estimate of up to ten years. A door propped open for 5 minutes can fall between two reports and never appear. A compressor failure appears, but the first report above 8 °C can arrive up to 15 minutes after the air crosses 8 °C. A 15-minute hold time adds 15 minutes more. Decide whether 30 minutes is acceptable for the stock. If it is not, shorten the reporting interval and accept the shorter battery life: a shorter interval, a colder site or a weak signal all shorten it.

Measure the delay end to end with a known step. Move the probe from room air into an ice-point bath and time how long the dashboard takes to show 63 % of the change.

Data quality and stale data

Keep the raw readings and their timestamps. Flag suspect values; never overwrite them or fill a gap with invented data. Check units, time zones and sudden jumps that the physics does not allow.

A stale point has stopped reporting. A stuck point keeps reporting the same value. They have different causes. A stale point is usually a battery or radio fault. A stuck point is usually a sensor fault, and its link looks healthy. Set a stale rule for each point, for example three missed reports: 45 minutes at a 15-minute interval. Flag a reading that has not changed at 0.01 °C resolution for several hours in a space that should vary.

Edge shows each device's online or offline state and its last report time. The online state is the Zigbee bridge's view of the device, and a cached value can still be old. Compare the last report time with the expected reporting interval.

Calculated values need the same care. By default, an Edge calculated measurement uses the last known value of a stale input. For a surface humidity or a room average, set the stale action to suppress the output or to exclude the input, so that a dead sensor does not keep feeding the result.

For averages, set a completeness rule and show when a period fails it. EU air quality law gives a defensible default. Directive 2008/50/EC, Annex XI, requires 75 % valid data for an hourly value (45 minutes), and 18 of 24 hourly values for a daily mean.

Commission the link and the measurement as two separate checks. The link check uses link quality and missed reports in Edge's network health view. The measurement check compares the reading with a reference at the sensor. A strong link delivers a wrong reading as reliably as a correct one.

Alerts and control

Write each alert with its threshold, hold time, reset, recipient and action. For the chiller:

8 °C is the legal maximum for chilled food in England, Wales and Northern Ireland. The Food Standards Agency advises setting fridges at 5 °C or below to allow for fluctuation. Without the hold time, every delivery through the door raises an alarm, and staff learn to ignore it. Without the 1 K reset band, a reading that hovers at 8.0 °C sets and clears the alarm on alternate reports.

Test both alerts. Warm the probe for a real excursion, then remove the battery for a communications failure. An alert starts an investigation: a surface humidity alert means check for mould risk, and a CO₂ alert means check the ventilation.

If a sensor drives a fan, a damper or a valve, decide what the plant does when the value goes stale. Hold the last output, go to a fixed safe position such as minimum fresh air, or return to local control. Test that case at commissioning by stopping the sensor.

Commissioning and handover

Three records are often missing a year later:

• a photograph of each mounted sensor, which shows the radiator or the sunny window that nobody wrote down;
• any offset or scaling set in the configuration, because a replacement sensor will not have it; and
• the clock source for each device, because a timestamp error looks like a response delay.

Keep the point schedule, calibration certificates and alert test results with them.

Apply a known input at each sensor and follow it to the dashboard. Breathe near the CO₂ sensor and watch the reading rise. Put a temperature probe in an ice-point bath. Confirm that the value and its timestamp reach Edge, the stored history and the dashboard.

Review the point schedule when the room use, occupancy, ventilation, mounting, firmware or alarm purpose changes. Continue with the indoor environmental application plan, or build a system.