A faulty sensor can look like a changing city
An outdoor temperature sensor reports an abrupt rise. Is the street getting hotter, or has the instrument lost its radiation shield? The gold standard of urban sensor failure detection distinguishes environmental signals from equipment faults before decisions are based on the readings.
Define failure modes before deployment
Sensors may stop transmitting, repeat stale values, drift gradually, produce implausible spikes or report incorrect timestamps. Each type of failure needs a suitable detection and recovery method.
Monitor freshness as well as connectivity
A device can appear online while its last valid measurement is old. Record the actual observation time, expected reporting interval and missing-data periods.
Use physical plausibility carefully
Thresholds and sudden-change checks can identify suspicious readings. However, genuine extremes and local microclimate differences must not be automatically discarded.
Compare with relevant context
Nearby reference observations, site conditions, maintenance records and calibration history can help explain anomalies. Two sensors at different exposure conditions need not agree.
Trace the entire data pathway
Problems can originate in the sensor, power supply, communications network, gateway, clock synchronisation or processing software. Diagnose the affected stage before replacing equipment.
Flag uncertainty transparently
Preserve appropriate raw data and mark suspect measurements. Silently filling gaps or overwriting observations can compromise long-term research.
Give faults an owner
An alert should lead to investigation, repair where necessary and verification of trustworthy measurements. Closing a ticket before validating the result is not full recovery.
Break-test: the sensor freezes during a heatwave
If the same value repeats for hours, can the system detect that readings are stale rather than interpreting them as stable weather? What alternative evidence remains available?
Measure reliability meaningfully
Useful measures include detection delay, data completeness, false alarms, successful diagnosis, verified repair and quality of recovered observations. Device uptime alone is insufficient.
Sources and connected knowledge
See Urban Sensor Network Resilience, Sensor Data Quality Assurance, Microclimate Sensor Calibration and NEA.