A Network Calibration Approach Improves the Accuracy and Long-Term Stability of a Low-Cost Air Quality Mesonet in New York City

TL;DR

A new calibration method improves the accuracy and stability of low-cost air quality sensors in New York City over 16 months.

Contribution

The Network Calibration Algorithm (NCA) enhances long-term sensor performance by compensating for drift and degradation.

Findings

NCA-calibrated sensors showed consistent or better performance than prior studies.

The method compensates for sensor drift and degradation over 16 months.

Field limit of detection estimates were provided for each low-cost sensor.

Abstract

A new calibration
approach, the Network Calibration Algorithm (NCA),
was developed and applied to low-cost sensors measuring PM2.5, O3, NO2, NO, and CO at 38 New York State
Mesonet sites in the New York City Metropolitan Area. A single low-cost
sensor package (the “keystone” package) was colocated
alongside regulatory-grade (reference) instruments at the New York
State Department of Environmental Conservation Queens College monitoring
site for 16 months. For each pollutant, hourly data from the keystone
package and reference instruments were used to train a single calibration
model that was subsequently applied to all packages at field sites
across the network. The calibration models included multiple linear
regression (MLR) for CO and a hybrid approach that combined MLR with
a Random Forest model for PM2.5, O3, NO2, and NO. The performance of the NCA-calibrated low-cost sensors
was quantified using multiple evaluation data sets, with a focus on
accuracy and long-term stability over the 16-month period. The performance
statistics were consistent with or better than previous reports for
similar low-cost sensors, and the NCA was able to compensate for sensor
degradation and drift. Empirical estimates of the field limit of detection
for each of the low-cost sensors are presented.

Figures