Low-Cost Particulate Matter Sensors Performance in Northeast China Polluted Conditions

This study evaluates the field performance of three mainstream low-cost particulate matter sensors (LCS) from the networks IQAir, PurpleAir, and AirGradient in Changchun, Northeast China, a region with clean background air, frequent spring dust storms, and autumn biomass-burning smoke. Two core research questions are addressed: whether heterogeneous commercial LCS can deliver reliable PM 2.5 observations across the full pollution gradient, and which sensor platform achieves optimal stability across diverse aerosol regimes. All LCS datasets are cross-validated against China National Environmental Monitoring Centre (CNEMC) regulatory reference monitors, with overall Pearson correlation coefficients of r = 0.92–0.95 across all pollution conditions. Clear platform-specific performance differentiation is evident: AirGradient achieves the most consistent agreement with reference measurements, with minimal systematic bias; PurpleAir displays a stable linear correlation yet persistent positive overestimation after factory calibration; IQAir units show prominent inter-device variability, with some monitors exhibiting time-varying concentration bias. Sensor consistency differs markedly between two typical extreme pollution episodes: all three LCS types yield highly correlated readings during wildfire and agricultural smoke events, whereas coarse mineral dust transport significantly weakens sensor-reference consistency, with PurpleAir exhibiting the most severe performance degradation. XGBoost-SHAP attribution analysis demonstrates that meteorological conditions, ambient PM2.5 loading, and aerosol particle size jointly explain approximately 30% of the total LCS measurement errors, arising primarily from inherent light-scattering optical limitations and the absence of built-in aerosol dehumidification. This work confirms that calibrated multi-brand LCS networks serve as effective complementary tools to official air quality monitoring in this geographical region. Still, site-specific adaptive correction models are mandatory to mitigate platform-dependent nonlinear biases, especially in dust-prone regions of Northeast China.