Dynamic calibration of low-cost PM2.5 sensors using trust-based consensus mechanisms

Abstract

Low-cost particulate matter (PM) sensors enable high-resolution urban air quality monitoring but face challenges from offsets, scaling mismatches, and drift. We propose an adaptive trust-based calibration framework that first corrects systematic errors and then dynamically adjusts model complexity based on sensor reliability. Extensive simulations and real-world deployment in Zurich, Switzerland validate the approach. Each sensor’s trust score integrates four indicators: accuracy, stability, responsiveness, and consensus alignment. High-trust sensors receive minimal correction, preserving baseline accuracy, while low-trust sensors leverage expanded wavelet-based features and deeper models. Results show mean absolute error (MAE) reductions of up to 68% for poorly performing sensors and 35–38% for reliable ones, outperforming conventional calibration methods. By using trust-weighted consensus, the framework reduces dependence on large training datasets and frequent re-calibrations, ensuring scalability. These findings demonstrate that dynamic, trust-driven calibration can substantially enhance low-cost sensor network accuracy across both controlled scenarios and complex real-world environments.