AAAR 37th Annual Conference October 14 - October 18, 2019 Oregon Convention Center Portland, Oregon, USA
Abstract View
Performance Evaluation of Light Scattering PM2.5 Sensors for Deployment in an Urban Sensing Network in Bangalore, India
JONATHAN GINGRICH, Mark Campmier, Advaitha Byereddy, Shayan Charolia, Heather Howton, Brian Mai, Meenakshi Kushwaha, Elbin Savio, Adithi Upadhya, Sreekanth Vakacherla, Julian Marshall, Joshua Apte, University of Texas at Austin
Abstract Number: 887 Working Group: Air Quality Sensors: Low-cost != Low Complexity
Abstract Low-cost aerosol devices offer the potential to fill critical data gaps on air pollution for cities. Here, we present data on the performance evaluation of 60 PurpleAir (PA) PA-IISD devices in Bangalore, India.
We established an ambient air evaluation site on the ~15m roof of a suburban office building in June-July 2019 with two beta-attenuation monitors (BAM-1022, US EPA FEM designation). Each PA device consists of two Plantower light-scattering sensor modules (PMS5003), resulting in 120 total sensors tested. Colocation of all devices lasted from June 20 to July 8, 2019, resulting in 441 hourly measurements. Hourly BAM PM2.5 concentrations averaged 10-40 µg m-3.
113 sensor modules were included in the analysis, as seven modules failed during colocation. Sensor-to-sensor comparisons between the PurpleAirs resulted in a high correlation. The hourly median pairwise R2 for all combinations of the sensors was 0.99, although 15 modules had median R2 values consistently below 0.90. The NRMSE ranged between 0.01 and 0.46 with a median of 0.02. The median pairwise slope was 0.97.
Because BAM measurements are imprecise at their native hourly averaging time, the correlation between the PA sensors and the BAM greatly increased as averaging time increased. For hourly data, there was weak correlation between PA and BAM measurements (median R2 = 0.19, median NRSME = 0.34, median slope= 0.46). For daily averaged data, the relationship improved dramatically (median R2= 0.81, median NRMSE = 0.11, median slope = 0.71). Care must be taken in comparing the PA data to BAM measurements for hourly measurements. At the daily time scale, the PA sensors track BAM readings well and are capable of resolving moderate (~10-20%) concentration differences.