Reading Temperature and Humidity Sensors Reliably

Why Your Sensor Reading Might Be Fibbing

You have wired up a temperature and humidity sensor, uploaded your sketch, and the numbers are scrolling nicely down the serial monitor. It feels like a small triumph — and it is. But before you trust those figures to control a fan, trigger a dehumidifier, or log a week of data for your robot's little brain, it is worth knowing that most beginner readings are wrong by a meaningful margin. Not because the sensor is broken, but because of where it sits, what is around it, and how it is being read.

Digital sensors such as the DHT22 or the BME280 are genuinely good for hobby work, typically accurate to within half a degree and two or three per cent relative humidity. Closing that gap is mostly about physics and patience rather than expensive hardware.

Placement: The Single Biggest Factor

Air temperature is only meaningful where air can actually reach the sensor. In a typical UK home, that leaves a lot of tempting spots off the list.

• Keep away from walls and windows. A sensor sitting 5 mm from a draughty single-glazed window in February will read several degrees colder than the room. Ten to fifteen centimetres of clearance is a reasonable minimum.
• Avoid corners. Corners trap stagnant air, which settles into a different temperature to the rest of the room.
• Dodge heat sources. Radiators, lamps, laptops, tumble dryers, and south-facing conservatories all skew results. So does direct sunlight — even a brief patch of sun can push a reading up by 5°C.
• Height matters. Around 1.2 to 1.5 metres above the floor is a sensible compromise for "what a person feels". Near the floor it is colder; near the ceiling it is warmer.
• Beware the robot itself. Motors, motor drivers, voltage regulators, and a Raspberry Pi-class board all shed heat. Mount the sensor on a short stalk or a small bracket away from the warm bits.

Airflow, Enclosures, and Condensation

If you are housing the sensor in a 3D-printed case, think of it as a tiny building. Solid walls trap heat and humidity in a pocket that slowly drifts away from the real world. Cut slots or drill a few 4–5 mm holes so air can move through, and position them so rain or splashes cannot run straight in.

For outdoor use — a greenhouse, shed, or garage — a louvred radiation shield makes a real difference. It blocks direct sun and rain while letting air circulate. A cheap alternative is a stack of small plant-pot saucers or a white plastic cup inverted over the sensor with a gap underneath.

Humidity is the harder measurement. Condensation forming on the sensing element will pin readings at or near 100% and can take hours to clear. If your robot moves between a warm kitchen and a cold garage, expect a settling period. In damp British winters, a sensor that has been saturated may need a day or two in a dry room to recover.

Calibrate Before You Trust

You do not need a laboratory. You need one or two reference points and the willingness to apply an offset in your code.

• Ice bath for temperature. Fill a jug with crushed ice and a little water, stir, and wait three minutes. A well-stirred ice bath sits at 0°C. Note the difference and apply it as a constant offset.
• Salt test for humidity. Put a small dish of saturated table salt slurry (salt plus a trickle of water) in a sealed jar with your sensor. After a few hours the air inside stabilises near 75% relative humidity at room temperature. Compare and offset accordingly.
• Compare against a second sensor. If you have two, note which reads high and which reads low. Averaging them is a decent workaround even without a true reference.

Offsets are rarely constant across the whole range, so treat them as a sensible correction rather than a fix. Log the offset value and the date in a comment, because you will forget.

Reading Reliably in Code

Raw single readings are noisy. A little processing goes a long way.

• Take several samples and use the median, not the mean. A median ignores the odd wild spike; a mean lets it drag the result.
• Respect sample rates. Some cheap sensors need a second or two between reads, and hammering them produces stale or garbage values.
• Add a sanity filter. If a reading jumps 10°C in ten seconds indoors, discard it and try again.
• Smooth slowly for control, log raw for analysis. A moving average is fine for deciding when to switch a fan on, but keep the unprocessed numbers if you are collecting data.

Moving Between Rooms and Going Outdoors

If your robot roams, remember that it carries its own microclimate. Give it a minute or two in a new location before recording anything, and avoid taking a reading immediately after a motor has been running hard. When heading outside, anticipate a settling period of several minutes, and be prepared for condensation if you are coming from a warm, humid house into cold air.

The golden rule is simple: a reading you understand the limitations of is far more useful than a precise-looking number you have never questioned. Place it thoughtfully, let the air reach it, nudge it with a calibration offset, and your little robot will be reporting the truth about your home — draughty hallway and all.