How Accurate Are Low-Cost Air Quality Sensors? A 10-Month Study of Three-Tier Comparison with Three Calibration Methods and Cross-Site Transferability

Low-cost sensors could increase the spatial granularity of air-quality monitoring at 700–1000 times less cost than reference stations (2026 prices), but few field studies evaluate them for gases. This study evaluates a three-tier architecture (low-cost metal oxide and optical, mid-cost electrochemical, high-cost government reference) over 10 months at two London sites, 1.6 km apart. The low-cost sensors and mid-cost NO2 sensor were calibrated with manufacturer curves, linear regression on temperature and relative humidity, and machine learning (ML). After field calibration, the low-cost PM2.5 sensor explains over half the reference variance at both sites, and linear calibration equals ML. The low-cost NO2 sensor agrees less with its reference than the PM2.5 sensor. The low-cost sensors explain over half the variance of a factory-calibrated mid-cost electrochemical CO sensor at both sites only with ML. The mid-cost NO2 and low-cost O3 sensors fail under all three methods. Calibrations trained on April–October 2025 stop describing November 2025–February 2026 data for NO2 at both sites and CO at one, whereas PM2.5 stays stable. Transferred between sites, PM2.5 calibrations reach R2≥0.56 and gas calibrations R2≤0.28. Low-cost PM2.5 sensors calibrated at a reference site could serve neighborhood-scale monitoring where reference stations are absent, whereas NO2 needs co-location.