Sensor drift over time: what to expect after year one

Every sensor is a physical system, and every physical system drifts over time. The interesting part is that each measurement channel drifts in a characteristic way, and "drift" means different things for each one. The Sensirion SEN66 inside Terrestream is a single integrated module that carries CO₂, PM, VOC, NO_x, humidity, and temperature channels, and each of those channels has its own relationship with the passage of time. The Bosch BMP390L pressure sensor is essentially drift-free at the scale we report it, and the TI OPT3001 lux sensor changes only as its window accumulates dust. The firmware and dashboard handle each channel differently. Knowing the shape of each makes "is the sensor right" much easier to answer.

The SEN66's photoacoustic CO₂ channel uses Sensirion's Automatic Self-Calibration (ASC), the same approach the older SCD4x family used. The sensor assumes that, at least once per week, the room it lives in will see outdoor air (~420 ppm) for some sustained period, and it uses that minimum as a re-anchor for the calibration curve. Homes that are reliably ventilated weekly (most homes) keep the CO₂ channel within ±30 ppm of truth indefinitely. Homes that never open windows and have very tight envelopes can drift the ASC baseline upward over months, which the dashboard catches as an "always above 600 ppm even at 4 AM with no one home" anomaly.

The SEN66's PM scattering channel drifts a few percent per year from optical-window contamination: dust and aerosol residue accumulate on the laser path optics and slowly bias readings upward. The drift is gradual (single-digit percent per year in normal indoor air) and largely correctable by the firmware's zero-baseline tracking, but a sensor that has lived for five years in a heavy-cooking or smoking household will read higher than a fresh one. The SEN66's VOC and NO_x index channels are a fundamentally different beast: they are relative baseline measurements, not absolute ones. The VOC and NO_x indices are calculated against a slowly tracking baseline of the room's own normal. "Drift" does not apply in the usual sense; the baseline shifts with the room's long-term chemistry, which is the design intent. The BMP390L pressure sensor is essentially stable for the lifetime of the device; barometric pressure is one of the cleanest measurements in the whole stack.

What to do when readings look wrong: first, reality-check the room. Is something different than it was last week? A new piece of furniture, a stored cleaning product, a different cooking pattern? The sensor is usually right and the room is different. Second, take the sensor outside on a calm day and let it sit for 20 minutes; a healthy CO₂ channel should read 400-450 ppm, and the PM channel should read close to your local outdoor PM_2.5 from a nearby NIST-traceable reference. Third, the manufacturer-defined factory-reset procedures are documented; the firmware exposes the SEN66's forced CO₂ re-calibration and a VOC/NO_x baseline reset. Expected service life across the whole sensor stack is 5-10 years before noticeable degradation in real-world residential use. See sensor cross-sensitivities and where do the numbers come from for the broader instrumentation context.