Inside or Outside
In October 2023, San Francisco had another bad air day. Smoke from Northern California fires pushed the AQI into the 200s, which are dangerous levels, especially for exercise. Public warnings urged people to stay indoors and avoid cardio activity.
I had a workout session scheduled with a physical trainer on a breezy hill and brought along a portable air quality sensor. Just outside my house, the AQI was over 200. But on the hilltop, it dropped to the 30s, which was fine for jumping jacks and running sprints.
When I got back into my house, I measured it inside, and it was at 300, even worse than outside.
The coarse measurements weren’t helpful. It was the fine ones that provided the information I needed to make better lifestyle choices.
I went to a local store, bought a HEPA air purifier, and watched the AQI numbers slowly crawl down to more acceptable levels.
What I realized is that having many air quality sensors throughout a home or small location provides useful information. I had already designed similar sensors for my Sensors and Soundscapes installations, so I built more and deployed about eight of them inside and outside my house over the following weeks.
Scalable Options
Purple Air provides a great crowd-sourced service, but at ~$240 per sensor, it’s not a scalable option.
People spend a lot of time inside their houses, and what is useful is to see how your bedroom compares to your kitchen. Particulates from your stir-fry dinner drift into other rooms. Maybe there’s circulation in some places, maybe there isn’t.
I use the same family of Plantower sensors as Purple Air, and can make a my own sensor package with an ESP32 chop for around $40 per sensor. The data gets displayed on a small screen and also pushed to an MQTT server for long-term analysis and immediate web display.
Mars College Prototype
At Mars College—a DIY pop-up art community in the desert—I prototyped an air quality (AQ) mesh network. In Bombay Beach, dust storms can cause poor air quality, and asthma rates are higher than average. Yet little scientific research has been done in the area.
I deployed my custom AQ sensors across the desert, in trailers, and in the homes of Bombay Beach to collect data. I wanted to explore whether this data can be useful for changing behavior, whether indoor air quality is worse than outdoor, and if any patterns can be identified.
What I learned
My sensors were low-cost, deployable and reliable. None of the Plantower sensors themselves
broke. I soon learned that during the dust storms, I saw higher counts of the large particulates (PM10s vs PM1.0).
I made a simple p5.js sketch that could show me live data, within the minute. I could see spikes in some of the kitchens when meals were cooked. If folks were smoking outside near the lounge, you could tell. Soldering irons were also a culprit.
In the trailers, during a dust storm, the dust would often get inside and swirl around the trailers afterwards, which meant running HEPA air filters just after these storms was helpful. The visual display prompted my own behavior change.
Did the behaviors of others also change? I was told that folks started ventilating their kitchens better. Some of the “Martians” went outside to vape. People reported looking at the simple temporary website I made to check to see the conditions outside.
The more complex data-analytics, well that never really happened. It was beyond my skillset and required more time than I had available.
Future Possibilities
The pilot project was a success, and I received a small grant to develop more sensors for Mars College 2025.
I’m currently refining the tech to make the sensors more configurable across different Wi-Fi networks, and also working on open-sourcing the hardware and software.
I’m also reaching out to organizations to test the prototypes. Their low cost makes them ideal for temporary sites, like a land trust monitoring particulates from local campfires.
The data can remain private, with a simple API for organizations to access and control how it’s used.
I’m curious to see where this goes next.
