Skip to main content

Calibration details by pollutant

Different pollutants are measured with different sensor technologies, so each one is calibrated differently. This article walks through how Clarity calibrates PM2.5, NO₂, Ozone, and Black Carbon.

PM2.5​

Global PM2.5 Calibration. The latest version (v2) draws on data from over 625 Clarity Monitors and 98 regulatory-grade reference sites across 84 cities. It achieves a median R² of 0.79 and an RMSE of 2.7 µg/m³, correcting for variability in aerosol composition. This calibration is preset on every Clarity device, so you get reliable data even when collocation isn't feasible. (More detail in this blog post.)

Collocation-Based PM2.5 Calibration. To tailor PM2.5 to your local conditions, Clarity deploys your Nodes near reference-grade monitors for at least four weeks, collecting data on pollutant concentrations and environmental conditions. Using an 80/20 split for model development and testing, Clarity builds a project-specific model that accounts for regional particulate characteristics. Cross-validation and quality checks add robustness, and you receive sensor-specific accuracy metrics.

NO₂​

Global NO₂ Calibration. This corrects the baseline shifts and interferences that temperature and humidity cause in electrochemical cell (ECS) sensors. It uses a machine learning model (an LGBM decision tree regressor) trained on collocation data from many global sites — 450 collocated Node-S devices, 2,000 collocation months, and 45 cities across different climates — to stabilize readings and align them with reference monitors, even without site-specific calibration.

Collocation-Based NO₂ Calibration. For the highest NO₂ accuracy, Clarity starts with a four-week collocation period alongside reference monitors.

Note: For NO₂, it's important to choose four weeks whose temperatures and humidities represent the range you'll see year-round — usually a fall or spring month.

After QA/QC checks, Clarity filters out outliers and focuses on typical conditions, then uses an 80/20 train/test split with repeated cross-validation to build a model specific to each sensor. Beyond correcting baseline shifts (as the Global calibration does), this reduces device-to-device variation and gives you quantifiable performance metrics (R² and RMSE) for each sensor.

Ozone​

Clarity's Ozone Modules use FEM-grade optical detection technology rather than low-cost sensors, so they're highly accurate out of the box. Because of this detection principle, no extra calibration is needed — factory calibration provides precise measurements. To keep that accuracy, Clarity offers a factory recalibration every two years as part of Sensing-as-a-Service.

Black Carbon​

The Black Carbon Modules also use sophisticated optical detection, so they measure precisely without extensive calibration. Factory calibration every two years — included in Sensing-as-a-Service — is enough to maintain high accuracy over time.

What's next​

Was this article helpful?​

Yes, thanks! / Not really

Still need a hand? Email us at support@clarity.io or create a support ticket, and our team will get back to you.

Last updated on