Reliable radon measurements, a challenge for property owners
Joan Rey on the EPFL campus with commercial tools used to measure radon.© EPFL/Alain Herzog - CC-BY-SA 4.0
Joan Rey tested the reliability of commercial tools and artificial intelligence to measure and predict radon levels in buildings for his doctoral thesis in civil engineering. He shares his conclusions in a column that appeared in three Swiss dailies.
Radon is a naturally occurring radioactive gas and a common indoor-air pollutant. It poses a significant public-health risk in Switzerland, as it’s found in just about all our buildings and is responsible for nearly 300 lung-cancer deaths each year. The Swiss authorities began making a concerted effort in the 1990s to reduce residents’ exposure to radon gas. Because the gas is both invisible and odorless, its concentration can only be measured with specific instruments.
The Swiss Federal Office of Public Health has compiled a list of approved radon measurement services, which generally take readings over the course of a year, although this can be reduced to three months during seasons when the heating is on. These services use instruments that have been approved by the Swiss Federal Institute of Metrology. However, with the expansion of connected devices, a number of unofficial measurement systems are now appearing on the market – but public-health officials are concerned about how reliable their readings are.
Comparing data to get a clear picture
For my PhD thesis, I tested low-end measurement devices both in a laboratory and inside buildings and found that they generally give satisfactory results, albeit with large differences in performance. But these devices struggled to track radon concentrations over time. In addition, they proved to be less accurate at low radon concentrations, that is, near the value of 300 Bq/m3, which is the threshold for deciding whether renovation work is needed. It’s therefore important for everyone – real-estate professionals and the general public – to understand the limitations of these systems and be discerning in how they interpret their output.
Initial tests of our AI system resulted in accurate predictions 85% of the time.
Predicting possible hazards
Recently, various studies in Europe have looked at the use of artificial intelligence (AI) to predict radon concentrations inside buildings. As part of my thesis, I also tested an AI-driven system – the first such AI application in Switzerland. I found that, in 85% of the cases, the system correctly predicted whether annual radon concentrations would be above or below the official threshold. I’m now taking this research further at the Fribourg School of Engineering and Architecture (HEIA-FR, TRANSFORM institute, croqAIR), where I’m improving the system’s accuracy so that it can eventually be used alongside or as an alternative to direct radon measurements.
Being able to take fast, reliable radon measurements is also a societal issue, as this information can help potential property buyers decide whether to proceed with a purchase and existing owners whether to carry out energy-efficiency renovations. Further research is needed, as are updated regulations and initiatives to build awareness about radon measurements among both the general public and real-estate professionals.
Joan Rey, PhD in civil and environmental engineering, Smart Living Lab researcher, EPFL Fribourg & HEIA-FR
- This article appears in May 2025 in three local dailies – La Côte (Vaud Canton), Le Nouvelliste (Valais Canton) and Arcinfo (Neuchâtel Canton) – under a joint initiative between EPFL and ESH Médias to showcase the R&D being carried out at EPFL on advanced construction techniques.