How Many PM2.5 Sensors are Needed to Estimate Daily and Annual Deployed-Network Mean Concentrations in City-Scale Sensor Networks?

Abstract

Abstract Low-cost PM2.5 sensors make it possible to deploy dense networks, but the density required depends on the monitoring objective and local variability. We examined how well smaller subnetworks reproduce deployed-network daily and annual or study-period mean concentrations using data from Dhaka, Bangladesh (N = 35; annual mean 55.15 μg/m3), Lucknow, India (N = 71; annual mean 61.80 μg/m3), and the nine-month Open Air Chicago network (N = 277; period mean 10.24 μg/m3). For each subnetwork size n ≥ 2, mean estimates from 10,000 simple-random-without-replacement draws were compared with the corresponding deployed-network mean; the n = 1 errors were enumerated exactly. At n = 10, the period median absolute percentage error (MdAPE) was 3.9%, 5.6%, and 1.9% in Dhaka, Lucknow, and Chicago, respectively, and the across-day median daily MdAPE ranged from 2.6% to 5.8%. For a single sensor, the period MdAPE ranged from 5.2% to 17.8%, whereas the corresponding 95th-percentile absolute percentage error (APE) ranged from 18.5% to 59.9%, demonstrating that typical and tail performance must be considered together. Within these three finite networks, n = 10 reproduced the target means with typical errors below 6%, although upper-tail performance remained objective- and city-specific. Estimating neighborhood-scale gradients, hotspots, exceedance classification, and population exposure, however, requires network designs tailored to those objectives. Overall, approximately 10 well-maintained, spatially distributed sensors recover annual and typical daily means with an MdAPE below 10%.