The proliferation of low-cost air quality sensors over the past decade has democratized air quality monitoring in ways that regulatory agencies never could have achieved through traditional monitoring networks. PurpleAir alone now has over 30,000 sensors deployed globally, providing spatial coverage of PM2.5 concentrations that the EPA's several-thousand-sensor FRM network cannot approach. For researchers, city planners, and health advocates, this data is genuinely transformative.
But the same low cost that makes mass deployment possible also limits measurement accuracy. Understanding the specific ways in which low-cost sensors underperform — and the conditions under which they perform quite well — is essential for anyone building applications on top of this data.
How Low-Cost PM2.5 Sensors Work
Most consumer-grade PM2.5 sensors, including the Plantower PMS series used in PurpleAir devices, use optical particle counting (OPC) as their measurement principle. A laser illuminates a small air sample volume, and a photodetector counts the number and size distribution of particles passing through based on light scattering patterns. The sensor converts these counts to a mass concentration estimate using a factory calibration that assumes a specific particle density and size distribution.
This assumed size distribution is where things get complicated. Wildfire smoke particles have a very different size distribution than the urban aerosol mixture the factory calibration was designed for. During wildfire events, Plantower-based sensors systematically overread PM2.5 — sometimes by factors of 1.5 to 5x compared to federal reference method (FRM) instruments. AirNow and PurpleAir now apply correction factors to their data during fire smoke events, but the correction factors are approximations and introduce their own uncertainty.
Relative Humidity Effects
High relative humidity significantly affects low-cost optical sensor readings. Particles absorb water vapor and grow in physical size, scattering more light and causing the sensor to report elevated PM2.5 even when dry mass concentrations haven't changed. This effect becomes significant above about 65% relative humidity and severe above 85%.
Reference-grade instruments either heat the sample stream to reduce particle water content before measurement or apply real-time humidity corrections based on co-located measurements. Most low-cost sensors do neither. In humid climates or during wet weather, low-cost sensor readings can substantially overestimate dry PM2.5 concentrations.
Some higher-end consumer sensors, including certain IQAir and Awair models, include humidity and temperature sensors and apply on-board corrections. These perform significantly better in humid conditions, though they're substantially more expensive than the base PurpleAir units.
Sensor Drift and Aging
Low-cost sensors drift over time as the laser output degrades, the photodetector ages, and optical surfaces accumulate contamination. Published research suggests significant sensor drift on timescales of 12–18 months without recalibration. In networks where sensors are deployed and then left unattended for years, this drift creates systematic biases that can be difficult to detect without co-located reference instruments.
PurpleAir's approach of deploying sensors in pairs (each unit contains two independent sensors) allows real-time quality flagging — when the two sensors diverge by more than a threshold amount, the unit is flagged as potentially unreliable. This is a reasonable heuristic but doesn't catch drift that affects both sensors equally.
When Low-Cost Data Is Reliable
Despite these limitations, low-cost sensor networks provide genuinely valuable data in specific use cases. For detecting spatial patterns and gradients — identifying which neighborhoods have consistently higher PM2.5 than others — low-cost networks are excellent. The spatial density that makes them valuable for this purpose is precisely what FRM networks can't provide.
For detecting trend changes and anomalies relative to recent baseline conditions, low-cost sensors work well. The sensor biases tend to be consistent over the short term, so if PM2.5 doubles relative to yesterday's reading at the same sensor, that signal is almost certainly real even if the absolute number is uncertain.
For regulatory compliance assessment and health exposure research requiring accurate absolute concentrations, low-cost sensors are not suitable without co-location correction against reference instruments. The EPA's Air Sensor Guidance document provides detailed recommendations for researchers using low-cost sensors in epidemiological contexts.
How Trace AQ Uses Sensor Data
Trace AQ incorporates data from multiple monitoring tiers: EPA FRM reference instruments as the calibration anchor, AirNow AQS network as a ground truth validation dataset, and PurpleAir and other low-cost networks as spatial gap-filling inputs. We apply published humidity and smoke correction factors to low-cost sensor data before assimilation, and we weight sensor observations by their estimated uncertainty when combining them with our physics-based model output.
This multi-tier approach means that our forecasts benefit from the spatial density of consumer sensor networks while being anchored to the accuracy of reference instruments. The consumer sensors tell us where spatial gradients exist; the reference instruments ensure our absolute concentrations are accurate. Neither tier alone would give us what we need.
For teams building their own air quality monitoring applications, we recommend the same layered approach: use low-cost sensors for spatial coverage and pattern detection, but always validate against at least one nearby reference instrument when accurate absolute concentrations matter.
