Improving low-cost NO₂ sensor reliability under real-world air quality conditions by calibrating distributions instead of data points

Low-cost electrochemical NO₂ sensors enable dense air-quality monitoring, but their quantitative reliability depends strongly on calibration under field conditions. Laboratory calibration often fails to transfer to the field because temperature, humidity, cross-sensitivities, drift, ageing, and memory effects alter sensor response. Timestamp-paired co-location calibration addresses some of these effects but remains vulnerable to micro-scale heterogeneity: the sensor and reference analyser may not sample exactly the same air mass, resulting in scatter and weak point-wise agreement. This study introduces distribution-alignment calibration (DAC), which shifts the calibration target from individual timestamp-paired observations to the statistical structure of the complete co-located time series. Frequency distributions are constructed for the sensor signal and reference NO₂ concentrations, after which linear regression between matched descriptors—the mode, mean ± standard deviation, and selected percentiles—defines a transfer function that maps the sensor-signal distribution onto the reference-concentration distribution. DAC was evaluated using four NO₂ monitoring campaigns conducted at the same reference station over nearly three years. Compared with laboratory and conventional timestamp-paired co-location calibration, DAC improved the representation of baseline concentrations, short-term variability, and peak magnitudes. Descriptor-stability analysis indicated that approximately one month of continuous monitoring was sufficient for most descriptors to reach a relative variability below 10%.