Performance Evaluation and Calibration of Low-Cost PurpleAir PM_2.5 Sensors in South Asian Conditions: Dhaka, Bangladesh

In this study, we assessed the performance of PurpleAir PM2.5 sensors and developed calibration models in Dhaka, Bangladesh─one of the global hotspots most severely affected by extreme air pollution. We collocated an array of PurpleAir (PA-II-SD) sensors alongside a beta attenuation monitor (BAM: MetOne BAM-1020) across the dry and wet seasons. Specifically, we collocated 10 sensors during the wet season and 20 sensors during the dry season, collecting one month of colocation data per season, covering a wide range of pollution levels and meteorological conditions. Quality-assured hourly concentrations from different PurpleAir units have shown good consistency, with pairwise R2 values generally exceeding 0.95. We developed empirical correction models by testing 29 multiple linear regression (MLR) forms and Random Forest models. Results showed that for hourly average PM2.5 concentrations measured by PurpleAir, a simple linear correction model achieved an accuracy (nRMSE) within 17–18% of hourly BAM measurements. More complex MLR models incorporating several meteorological variables and interaction terms improved accuracy (nRMSE) slightly, to ∼15%. Random Forest models slightly outperformed all MLR models, at 12–14% (nRMSE) accuracy relative to BAM. Our findings highlight that existing correction models─particularly those developed for U.S. cities and used in the PurpleAir map─are inadequate for Bangladeshi conditions. Uncorrected PurpleAir cf_atm PM2.5 data yielded accuracy within 25% of BAM measurements. Further research is needed to assess sensor performance in rural and suburban environments and to evaluate long-term performance under diverse climatological and source conditions in Bangladesh and South Asia.