The Complexity of Indoor Air Quality Forecasting and the Simplicity of Interacting with It

The R&D team from Logitech needs research evidence to guide the next steps to develop a novel device to improve the monitoring of and interaction with indoor environmental quality, focusing on the office environments. This research aims to:  

Generative Research Goals

• Understand user awareness and engagement levels regarding air quality in meeting rooms.  
• Explore how social dynamics and comfort conditions influence users’ perceptions of air quality interventions.  
• Identify user needs for maintaining a balance between air quality control and meeting productivity.  

Design Research Goals

• Evaluate the effectiveness of predictive models in forecasting CO2 levels and triggering preventive actions.  
• Assess user interactions with proposed solutions, ensuring they are unobtrusive and socially appropriate.  
• Test the usability and discoverability of features like real-time air quality feedback and preventive action prompts.  

This research aims to address the significant impact of poor air quality on productivity and well-being, particularly in high-occupancy spaces like meeting rooms. By combining predictive algorithms with user-centered design, the solution seeks to empower users to take preventive actions, fostering long-term behavioral change and improving overall environmental quality.

To address the challenge of minimizing interruptions while maintaining air quality in meeting rooms, we conducted a field study in two office buildings in central Europe. Over five months, we deployed 24 environmental sensing devices across 28 desks and 18 meeting rooms, collecting real-time data on CO2, temperature, humidity, and other factors. The devices, co-developed with an electronics manufacturer, also allowed users to provide subjective assessments of air quality via a smartphone app. We focused on nine naturally ventilated meeting rooms, analyzing CO2 evolution patterns, user perceptions, and meeting dynamics. Additionally, we integrated outdoor weather data and meeting room occupancy information to contextualize the findings. This comprehensive approach provided insights into how air quality evolves during meetings and how users perceive and interact with their environment, informing the design of non-intrusive, user-centered solutions. 

To facilitate the data collection, I designed the following interface to allow users to quickly input their feedback through a QR code pasted on a sensing prototype and designed posters to position on the desks to remind users of the data collection activities. 

After completing the data collection, I started with Linear discriminant analysis to understand the connections between all variables. Unfortunately, there are too many of them! 

To narrow down the scope of environmental sensing, a senior UX researcher suggests starting by looking at indoor air quality, where its impact on the comfort of users is more linear. We focus on CO2 because it's a parameter that everyone understands without in-depth environmental engineering knowledge and a great indicator for indoor ventilation. 

I, therefore, focus on the CO2 data in the dataset. However, for privacy reasons, we could not log all the activities of users (e.g., their presence). I started to concentrate on meetings as normally calendar invitations are stored, and during 1-hour meetings, people normally do not go in and out -- the activity and number of users are rather stable. Hence, I started to focus on this scenario and decipher the evolution patterns of CO2. It's clustering and prediction process is documented on Miro and discussed with the senior UX researcher. 

To bridge the gap between quantitative CO2 evolution patterns and a user-centered design proposal, I conducted two online surveys to address critical questions:  

• When to Intervene: Should users be prompted to open windows before a meeting (preventing interruptions but risking thermal discomfort) or during the meeting (ensuring accuracy but potentially disrupting flow)?  
• How to Intervene: What interaction modalities and design techniques are most effective and least intrusive for notifying users during meetings?  

To address these two questions, I design two surveys. the first survey evaluated user preferences across six vulnerable scenarios, building on identified CO2 evolution patterns. The second survey extended prior research on indoor air quality (IAQ) forecasting, assessing how urgency and notification modalities (e.g., visual, auditory) influence user responses in meeting contexts. These insights informed the design of adaptive, context-aware solutions that balance air quality management with minimal disruption to user productivity and comfort.

I utilized a Generalized Linear Mixed Model (GLIMMIX in SAS) to analyze the survey data, as it is well-suited for handling categorical data that does not follow a normal distribution. This approach allowed me to account for both fixed effects (e.g., user preferences) and random effects (e.g., individual differences, contextual factors), ensuring robust and accurate insights into user behavior and preferences. By leveraging this advanced statistical method, I was able to sketch out meaningful user scenarios and inform the design of adaptive, user-centered solutions for managing air quality in meeting environments.

Building on insights from the studies, I developed a core solution and design extensions to address air quality challenges in meeting rooms while minimizing disruptions. This solution is informed by user research and designed to adapt to varying infrastructural capabilities.  

Core Solution  
Pre-Meeting Notifications:

• Based on Study 1 and Survey 1, I created a guideline to determine when pre-meeting notifications should be activated.  
• The system alerts early arrivals about rising CO2 levels and displays outdoor temperature, empowering users to take preventive actions like opening windows  

In-Meeting Notifications:

• Survey 2 revealed that urgent alerts (e.g., "High CO2 in 5 minutes") are more effective than vague warnings.  
• Notifications are displayed on a public device, as users found this modality less intrusive and more actionable.  

Design Extensions
To enhance flexibility, I explored additional design extensions based on available infrastructure:  
Ambient Displays:

• A calm blue light on windows can subtly indicate when to open windows before meetings, reducing disruptions while maintaining awareness.  Wearable Devices: 
• For users interested in monitoring, non-urgent alerts (e.g., "High CO2 in 20 minutes") can be delivered via smartwatches, minimizing social stress and interruptions.  
• Notifications are deactivated when the user is speaking, ensuring minimal disruption during critical moments.