Proxy methods for detection of inhalation exposure in simulated office environments

Seoyeon Yun, Sailin Zhong, Hamed S. Alavi, Alexandre Alahi, Dusan LicinaJournal of Exposure Science & Environmental Epidemiology 2022 · DOI

TL;DR

This study demonstrates that stationary air quality sensors often fail to capture what office workers actually breathe due to the "personal cloud" effect. By combining stationary monitors with occupancy sensors (PIR) and physiological wearables, researchers developed models that significantly improve inhalation exposure estimates, particularly by segregating data based on whether a person is sitting or standing.

What Problem Does This Paper Address?

Most indoor air quality (IAQ) standards rely on the assumption that air in a room is "perfectly mixed." Consequently, we typically place a single sensor on a wall or in the middle of a room and assume it represents everyone’s exposure.

However, humans are active sources of pollutants. We exhale $CO_2$ and resuspend particulate matter (PM) from furniture and clothing, creating a localized "personal cloud" of higher pollutant concentrations around our bodies. This paper addresses the gap between stationary room measurements and actual inhalation exposure—the air 20 cm below your nose. The authors explore whether "proxy" methods (cheaper, less intrusive sensors) can accurately estimate this personal exposure without requiring every employee to wear a laboratory-grade respirator.

Methods

The researchers conducted 11 controlled experiments in a climate chamber configured to mimic four office layouts (shared offices, meeting rooms, and a cafeteria).

They compared "ground-truth" inhalation data (collected via sensors placed in the breathing zone of a reference participant) against three types of proxy data:

  1. Stationary IAQ Monitors: Placed at the front edge of the desk, on the desk surface, and at the ceiling-mounted exhaust vents.
  2. Contextual Sensors: Passive Infrared (PIR) sensors to detect human presence and network cameras to provide ground-truth occupancy and posture data.
  3. Physiological Wearables: An Empatica E4 wristband measuring skin temperature, heart rate, and 3-axis acceleration.

The team used Multiple Linear Regression (MLR) to build models that could predict breathing-zone $CO_2$, $PM_{2.5}$, and $PM_{10}$ levels using various combinations of these proxy inputs.

Datasets

The study utilized a primary experimental dataset generated by the authors:

  • Chamber Data: 11 experimental runs involving six human participants (3 male, 3 female) performing scripted office tasks.
  • Pollutants: Continuous monitoring of $CO_2$, $PM_{2.5}$, and $PM_{10}$.
  • Activity Logs: Scripted "sitting" (typing, reading) and "standing" (walking, using a coffee machine) activities to create dynamic occupancy profiles.

Key Insights and Ideas

  • The Posture Pivot: One of the most significant findings was that segregating data by activity (sitting vs. standing) improved model accuracy by 9–60%. Airflow patterns change drastically when a person stands up, breaking the localized thermal plume that normally carries exhaled breath away.
  • The Exhaust Advantage: For particulate matter ($PM_{2.5}$ and $PM_{10}$), sensors located at the ceiling exhaust vent were surprisingly accurate proxies for inhalation exposure (Adjusted $R^2$ up to 0.91). This suggests that the human thermal plume effectively carries particles toward the ceiling return.
  • $CO_2$ is Localized: Unlike PM, $CO_2$ was best estimated by sensors at the front edge of the desk (near the abdomen). This is likely because exhaled $CO_2$ jets often propagate downwards during sitting activities before being dispersed.
  • Sensor Fusion Works: Combining IAQ data with physiological and occupancy data doubled the accuracy of $CO_2$ exposure detection during sitting activities.

Key Results

  • Breathing Zone vs. Room: Pollutant concentrations in the breathing zone were substantially higher than stationary measurements. $PM_{2.5}$ and $PM_{10}$ levels were roughly 6.7 to 6.8 times higher in the breathing zone than at stationary monitors.
  • Model Performance:
    • PM Exposure: Stationary monitors at the exhaust vent provided a "strong" fit (Adjusted $R^2 = 0.91$ for $PM_{2.5}$).
  • $CO_2$ Exposure: A single stationary monitor was a poor predictor ($R^2 \approx 0.25$), but adding wearable data (skin temp, heart rate) and PIR data boosted the $CO_2$ estimation accuracy by 101% for sitting activities.
  • Significance: The study provides a roadmap for "smart" IAQ systems that use occupancy and posture data to adjust ventilation dynamically based on estimated personal exposure rather than bulk room averages.

Limitations and Future Work

  • Sample Size: The study relied on a small group (6 participants) and focused on a single "reference" participant for the breathing zone measurements. Individual differences in lung capacity, height, and metabolic rate could affect the results.
  • Controlled Environment: While the chamber mimics an office, it lacks the complex, unpredictable airflow of a real-world building with open windows or high-traffic corridors.
  • Intrusiveness: While wearables improved $CO_2$ detection, the authors acknowledge that requiring employees to wear wristbands for IAQ control might face privacy and comfort hurdles.
  • Future Directions: The authors suggest integrating camera-based human activity recognition (like the PifPaf algorithm) to automate posture detection, further refining the "sitting vs. standing" models without needing manual logs.

Summary updated

Abstract

This study contributes to broadening the knowledge of proxy methods for personal exposure estimation under dynamic occupancy profiles. The study recommendations on optimal monitor combination and placement could help stakeholders better understand spatial air pollutant gradients indoors which can ultimately improve control of IAQ.
BibTeX
@article{yun2022proxy,
  author = {Seoyeon Yun and Sailin Zhong and Hamed S. Alavi and Alexandre Alahi and Dusan Licina},
  title = {{Proxy methods for detection of inhalation exposure in simulated office environments}},
  journal = {Journal of Exposure Science \& Environmental Epidemiology},
  year = {2022},
  doi = {10.1038/s41370-022-00495-w},
}
Figure