Hierarchical modeling of indoor radon concentration: how much do geology and building factors matter?

Exploring statistical and machine learning techniques to identify factors influencing indoor radon concentration

Radon is a radioactive gas with a carcinogenic effect. The malign effect on human health is, however, mostly influenced by the level of exposure. Dangerous exposure occurs predominantly indoors where the level of indoor radon concentration (IRC) is, in its turn, influenced by several factors. The current study aims to investigate the combined effects of geology, pedology, and house characteristics on the IRC based on 3132 passive radon measurements conducted in Romania. Several techniques for evaluating the impact of predictors on the dependent variable were used, from univariate statistics to artificial neural network and random forest regressor (RFR). The RFR model outperformed the other investigated models in terms of R2 (0.14) and RMSE (0.83) for the radon concentration, as a dependent continuous variable. Using IRC discretized into two classes, based on the median (115 Bq/m3), an AUC-ROC value of 0.61 was obtained for logistic regression and 0.62 for the random forest classifier. The presence of cellar beneath the investigated room, the construction period, the height above the sea level or the floor type are the main predictors determined by the models used.

Geologic, seasonal, and atmospheric predictors of indoor home radon values

Exposure to tobacco smoke and radon cause lung cancer. Radioactive decay of naturally occurring uranium in bedrock produces radon. Seasonality, bedrock type, age of home, and topography have been associated with indoor radon, but the research is mixed. The study objective was to examine the relationships of geologic (soil radon and bedrock) and seasonal (warm and cold times of the year) factors with indoor home radon values in citizen scientists' homes over time, controlling for atmospheric conditions, topography, age of home, and home exposure to tobacco smoke. We collected and analyzed indoor radon values, soil radon gas concentrations, and dwelling- and county-level geologic and atmospheric conditions on 66 properties in four rural counties during two seasons: (1) summer 2021 (n = 53); and (2) winter/spring 2022 (n = 52). Citizen scientists measured indoor radon using Airthings radon sensors, and outdoor temperature and rainfall. Geologists obtained soil radon measurements using RAD7 instruments at two locations (near the dwelling and farther away) at each dwelling, testing for associations of indoor radon values with soil values, bedrock type, topography, and atmospheric conditions. Bedrock type, near soil radon levels, home age, and barometric pressure were associated with indoor radon. Dwellings built on carbonate bedrock had indoor radon values that were 2.8 pCi/L (103.6 Bq m-3) higher, on average, compared to homes built on siliclastic rock. Homes with higher near soil radon and those built <40 ago were more likely to have indoor radon ⩾4.0 pCi/L (148 Bq m-3). With higher atmospheric barometric pressure during testing, observed indoor radon values were lower. Seasonality and topography were not associated with indoor radon level. Understanding relationships among bedrock type, soil radon, and indoor radon exposure allows the development of practical predictive models that may support pre-construction forecasting of indoor radon potential based on geologic factors.

Machine learning in environmental radon science

Temporal dynamic as well as spatial variability of environmental radon are controlled by factors such as meteorology, lithology, soil properties, hydrogeology, tectonics, and seismicity. In addition, indoor radon concentration is subject to anthropogenic factors, such as physical characteristics of a building and usage pattern. New tools for spatial and time series analysis and prediction belong to what is commonly called machine learning (ML). The ML algorithms presented here build models based on sample and predictor data to extract information and to make predictions. We give a short overview on ML methods and discuss their respective merits, their application, and ways of validating results. We show examples of 1) geogenic radon mapping in Germany involving a number of predictors, and of 2) time series analysis of a long-term experiment being carried out in Chiba, Japan, involving indoor radon concentrations and meteorological predictors. Finally, we identified the main weakness of the techniques, and we suggest actions to overcome their limitations.

Radon levels and indoor air quality after application of thermal retrofit measures-a case study

This study was conducted to evaluate the influence of thermal retrofit on radon levels in workrooms, and to determine whether the radon concentration in the building changes after the application of retrofit measures. In the first survey, digital Airthings Corentium Home radon detector was used for 1-month radon measurements during the heating season 2018/19. The daily averaged radon concentrations varied from 37 to 573 Bq/m³ for 10 selected workrooms, while hourly averaged radon measurements showed extreme variations from 6 to 1603 Bq/m³ due to radon fluctuations. In second survey, passive radon technique based on charcoal canister test kit was conducted in all basement workrooms in spring 2021. The averaged radon concentrations grouped according to flooring type in workrooms were 327 Bq/m³ for parquet, 227 Bq/m³ for ceramic tiles, 146 Bq/m³ for vinyl flooring and 71 Bq/m³ for laminate. Besides thermal insulation and airtight windows, noticeable differences in indoor radon concentration within the renovated building are primarily caused by different types of flooring. It includes various types of insulation from the ground/concrete slab: laminate, parquet (wood blocks), vinyl flooring, and ceramic tiles. Detailed analysis point out that laminate is more efficient way for radon protection than other types of flooring. An efficient ventilation system should be installed to avoid increasing occupational radon exposure and to provide healthy and comfortable indoor environment.

INDOOR RADON MONITORING IN SELECTED BUILDINGS IN VYDRNÍK (VIKARTOVCE FAULT, SLOVAKIA)

To test the relation of indoor radon concentration to fault zone, a survey was conducted in selected buildings in Vydrník, situated close to the neotectonically active Vikartovce fault trace. Monitoring was performed using RamaRn track detectors during a period of 1 year. Annual average of indoor radon ranged from 60 to 940 Bq/m3, with an average of 160 Bq/m3. Seasonal variation of indoor radon concentration was confirmed, with minimum in summer (June-August). The results confirmed that radon concentration in rooms above the cellar was lower than in rooms with the direct contact with subsoil. High indoor radon up to 940 Bq/m3 was detected only in one of monitored houses situated close to the generally E-W trending Vikartovce fault and can be attributed to this fault zone used as pathway for radon migration. Influence of nearby situated NW-SE trending crosscutting fault on radon concentration is probable too.

Designing a Multicriteria WebGIS-Based Pre-Diagnosis Tool for Indoor Radon Potential Assessment

Radon (222Rn) is a well-known source of indoor air contamination since in its gaseous form it is a reported source of ionizing radiation that belongs to the group of rare gases. Radon occurs naturally in soils and rocks and results from the radioactive decay of its longer-lived progenitors, i.e., radium, uranium, and thorium. Radon releases itself from the soil and rocks, which mainly occurs in outdoor environments, not causing any kind of impact due to its fast dilution into the atmosphere. However, when this release occurs in confined and poorly ventilated indoor environments, this release can result in the accumulation of high concentrations of radon gas, being recognized by the World Health Organization (WHO) as the second cause of lung cancer, after smoking. Assessing the indoor radon concentration demands specific know-how involving the implementation of several time-consuming tasks that may include the following stages: (1) radon potential assessment; (2) short-term/long-term radon measurement; (3) laboratory data analysis and processing; and (4) technical reporting. Thus, during stage 1, the use of indirect methods to assess the radon occurrence potential, such as taking advantage of existent natural radiation maps (which have been made available by the uranium mineral prospecting campaigns performed since the early 1950s), is crucial to put forward an ICT (Information and Communication Technology) platform that opens up a straightforward approach for assessing indoor radon potential at an early stage, operating as a pre-diagnosis evaluation tool that is of great value for supporting decision making towards the transition to stage 2, which typically has increased costs due to the need for certified professionals to handle certified instruments for short-term/long-term radon measurement. As a pre-diagnosis tool, the methodology proposed in this article allows the assessment of the radon potential of a specific building through a WebGIS-based platform that adopts ICT and Internet technologies to display and analyze spatially related data, employing a multicriteria approach, including (a) gamma radiation maps, (b) built environment characteristics, and (c) occupancy profile, and thus helping to determine when the radon assessment process should proceed to stage 2, or, alternatively, by eliminating the need to perform additional actions.

Predictors of indoor radon levels in the Midwest United States

Radon (Rn) is a natural and toxic radioactive gas that accumulates indoors, mainly in low-ventilated underground floors and basements. Several factors make prediction of indoor Rn exposure in enclosed spaces challenging. In this study, we investigated the influence of soil, geology, topography, atmospheric variables, radiation, urbanization, community economic well-being, and monthly and yearly variations on indoor Rn concentrations. We analyzed 7,515 monthly indoor Rn measurements in 623 zip codes from two U.S. States (Michigan and Minnesota) during 2005-2018 using a random forest model. Using Shapley Additive exPlanations (SHAP) values we investigated the contribution of each factor using variable importance and partial dependence plots. Factors that predict indoor Rn differed between states, with topographical, geological and soil composition being most influential. Cross-validated Pearson correlation between predictions and measurements was 0.68 (RMSE = 47.8 Bq/m³) in Minnesota, and 0.67 (RMSE = 52.5 Bq/m³) in Michigan. Our results underline the importance of soil structure for radon exposure, presumably due to strapped Rn in soil. The differences across states also suggest that Rn studies performing model development should consider geographical variables, along with other factors. As indoor Rn levels are multifactorial, an understanding of the factors that influence its emanation and build up indoors will help better assess spatial and temporal variations, which will be useful to improve prevention and mitigation control strategies.Implications: Radon exposure has become a year-round problem as people spend most of their time indoors. In North America, radon exposure is increasing over time and awareness related to its health effects remains low in the general population. Several factors make prediction of indoor radon exposure in enclosed spaces challenging. In this study, we used random forest to investigate the influence of factors on indoor radon in the Midwest United States. We found that topography, geology, and soil composition were the most influential factors on indoor radon levels. These results will help better assess spatial and temporal variations, which will further help better prevention and mitigation control strategies.

Mapping indoor radon hazard in Germany: The geogenic component

Indoor radon is considered as an indoor air pollutant due to its carcinogenic effect. Since the main source of indoor radon is the ground beneath the house, we utilize the geogenic radon potential (GRP) and a geogenic radon hazard index (GRHI) for predicting the geogenic component of the indoor Rn hazard in Germany. For this purpose, we link indoor radon data (n = 44,629) to maps of GRP and GRHI and fit logistic regression models to calculate the probabilities that indoor Rn exceeds thresholds of 100 Bq/m³ and 300 Bq/m³. The estimated probability was averaged for every municipality by considering only the estimates within the built-up area. Finally, the mean exceedance probability per municipality was coupled with the respective residential building stock for estimating the number of buildings with indoor Rn above 100 Bq/m³ and 300 Bq/m³ for each municipality. We found that (1) GRHI is a better predictor than GRP for indoor radon hazard in Germany, (2) the estimated number of buildings above 100 Bq/m³ and 300 Bq/m³ in Germany is ~2 million (11.6% of all residential buildings) and ~ 350,000 (1.9%), respectively, (3) areas where 300 Bq/m³ exceedance is greater than 10% comprise only 0.8% of the German building stock but 6.3% of buildings with indoor Rn exceeding 300 Bq/m³, and (4) most urban areas and, hence, most buildings (77%) are located in low hazard regions. The implications for Rn protection are twofold: (1) the Rn priority area concept is cost-efficient in a sense that it allows to find the most buildings that exceed a threshold concentration with a given amount of resources, and (2) for an optimal reduction of lung cancer risk areas outside of Rn priority areas must be addressed since most hazardous indoor Rn concentrations occur in low to medium hazard areas.

The new Austrian indoor radon survey (ÖNRAP 2, 2013-2019): Design, implementation, results

The delineation of radon prone areas is one of the central requirements of the European Council Directive 2013/59/EURATOM. It is quite a complex task which usually requires the collection of radon data through an appropriate survey as a first step. This paper presents the design and methodology of the recent Austrian radon survey (ÖNRAP 2, 2013-2019) and its implementation. It details the results of the nationwide survey as well as correlations and dependencies with geology and building characteristics. The paper also discusses the representativeness of the survey as well as advantages and disadvantages of the selected approach. For the purpose of establishing a new delineation of radon prone areas in Austria we distributed approximately 75,000 passive long-term radon detectors. They were offered to selected members of the voluntary fire brigades and this resulted in about 50,000 radon measurements. Thus, a return rate of about 67% was achieved. The distribution of the radon results closely follows a log-normal distribution with a median of 99 Bq/m³, a geometric mean of 109 Bq/m³, and a geometric standard deviation factor of 2.29. 11% of the households show a mean radon concentration above the national reference level of 300 Bq/m³. Important data on building characteristics and the location of the measured rooms were collected by means of a specific questionnaire and a measurement protocol that were handed out together with the radon detectors. We were able to identify significant correlations between the indoor radon concentration and geology, the year of construction, and the coupling of the room to the ground (basement yes/no, floor level). Being a geographically-based and not a population-weighted survey, the comparison of building characteristics with the Austrian census data confirms that rural areas are over-represented in this survey. As a summary, the selected approach of conducting passive long-term radon measurements in selected dwellings of members of the voluntary fire brigades proved to be an efficient method to collect reliable data as a basis for the delineation of radon prone areas. The next step was to eliminate factors that influence the measured radon concentration through appropriate modelling. Based on the results predicted by the model radon areas are then be classified. This will be presented in a subsequent publication.

Developing a method for predicting radon concentrations above a reference level in new montenegrin buildings

Dependence of indoor radon concentrations (IRCs) in the ground floors of 1200 buildings across Montenegro on 11 factors was analyzed. A group of 734 buildings, for which none of the analyzed factors was missing, was further analyzed using the logistic regression method, in order to develop a prediction model for IRC occurrence above the national reference level for new buildings (200 Bq/m³). Applying the forward stepwise method, and based on likelihood ratios, five explanatory variables-municipality, type of building, presence of basement, window frames, and period of construction-were selected for including into the final logistic regression model for predicting probability of IRC > 200 Bq/m³. The final model explained 77.1% of the observed IRCs, while the obtained Area under the Curve of 0.8018 classified the model as having a very high predictive ability. Achieving similar values for both the final prediction model and the validation model, for sensitivity, specificity, and accuracy, confirmed the applicability of the developed model.

EVALUATION OF REPRESENTATIVENESS OF SAMPLES USED FOR INDOOR RADON SURVEYS

The estimation of the indoor radon exposure of the population of a country is generally carried out by the means of surveys designed in order to have sample representativeness as a target (population-based survey). However, the estimates of radon concentration distributions could be affected by biases if sampling was not random or in case of differences between sample and target population characteristics. In this work, we performed a preliminary check of the representativeness of the sample used for the second Italian national survey aimed to evaluate radon concentration distribution in each Province. We found that sampled dwellings are mostly located in the main administrative centres, where average radon concentration is generally lower, as compared with the other towns of the Province. The potential source of bias identified in this work suggests to carefully control the occurrence of a sampling imbalance between 'main' cities and other cities of Province and to take it into account in data analysis.

The FIRST large-scale mapping of radon concentration in soil gas and water in Romania

In the framework of the last Council Directive 2013/59 (Euratom, 2014) laying down basic safety standards for protection against the dangers arising from exposure to ionizing radiation, the problem of radon was assumed in Romania at national level by responsible authorities through the design and development of a National Radon Action Plan and an adequate legislation (HG nr. 526/2018). In order to identify radon risk areas, however, it is necessary to perform systematic radon measurements in different environmental media (soil gas, water, indoor air) and to map the results. This paper presents an atlas of up-to-date radon in soil and water levels for central and western part of Romania. The radon in soil map includes data from 2564 measurements carried out on-site, using Luk3C radon detector. The Luk-VR system was used to measure radon activity concentration from 2452 samples of drinking water. The average radon activity concentration was 29.3 kBq m^-3 for soil gas, respectively 9.8 Bq l^-1 for water dissolved air. Mapping of radon can be a useful tool to implement radon policies at both the national and local levels, defining priority areas for further study when land-use decisions must be made.

Confluent impact of housing and geology on indoor radon concentrations in Atlanta, Georgia, United States

Radon is a naturally released radioactive carcinogenic gas. To estimate radon exposure, studies have examined various risk factors, but limited information exists pertaining to the confluent impact of housing characteristics and geology. This study evaluated the efficacy of housing and geological characteristics to predict radon risk in DeKalb County, Georgia, USA. Four major types of data were used: (1) three databases of indoor radon concentrations (n = 6757); (2) geologic maps of rock types and fault zones; (3) a database of 402 in situ measurements of gamma emissions, and (4) two databases of housing characteristics. The Getis-Ord method was used to delineate hot spots of radon concentrations. Empirical Bayesian Kriging was used to predict gamma radiation at each radon test site. Chi-square tests, bivariate correlation coefficients, and logistic regression were used to examine the impact of geological and housing factors on radon. The results showed that indoor radon levels were more likely to exceed the action level-4 pCi/L (148 Bq/m³) designated by the U.S. Environmental Protection Agency-in fault zones, were significantly positively correlated to gamma readings, but significantly negatively related to the presence of a crawlspace foundation and its combination with a slab. The findings suggest that fault mapping and in situ gamma ray measurements, coupled with analysis of foundation types and delineation of hot spots, may be used to prioritize areas for radon screening.

Impact of ventilation systems and energy savings in a building on the mechanisms governing the indoor radon activity concentration

For a given radon potential in the ground and a given building, the parameters affecting the indoor radon activity concentration (IRnAC) are indoor depressurization of a building and its air change rate. These parameters depend mainly on the building characteristics, such as airtightness, and on the nature and performances of the ventilation system. This study involves a numerical sensitivity assessment of the indoor environmental conditions on the IRnAC in buildings. A numerical ventilation model has been adapted to take into account the effects of variations in the indoor environmental conditions (depressurization and air change rate) on the radon entry rate and on the IRnAC. In the context of the development of a policy to reduce energy consumption in a building, the results obtained showed that IRnAC could be strongly affected by variations in the air permeability of the building associated with the ventilation regime.

A Comparative Study of Indoor Radon Levels between Two Similar Dwellings Using CONTAM Software

Considered the second biggest risk factor for lung cancer in humans after smoking, indoor radon gas (222Rn) is produced by the radioactive decay of 226Ra, which is mainly present in building materials and soil. The current study proposes a methodology for estimating radon levels inside a dwelling using the multizone airflow and contaminant transport modeling software CONTAM, with the main purpose to improve indoor air quality by optimizing mechanical ventilation. The generation rate of indoor radon sources was investigated in relation to the total surface of each room and was subsequently used for identifying the rooms that require special attention. The methodology described should assist in modeling indoor radon contamination, which can be used to optimize the ventilation in a dwelling.

The use of gamma-survey measurements to better understand radon potential in urban areas

Accounting for as much as 14% of all lung cancers worldwide, cumulative radon progeny exposure is the leading cause of lung cancer among never-smokers both internationally and in the United States. To understand the risk of radon progeny exposure, studies have mapped radon potential using aircraft-based measurements of gamma emissions. However, these efforts are hampered in urban areas where the built environment obstructs aerial data collection. To address part of this limitation, this study aimed to evaluate the effectiveness of using in situ gamma readings (taken with a scintillation probe attached to a ratemeter) to assess radon potential in an urban environment: DeKalb County, part of the Atlanta metropolitan area, Georgia, USA. After taking gamma measurements at 402 survey sites, empirical Bayesian kriging was used to create a continuous surface of predicted gamma readings for the county. We paired these predicted gamma readings with indoor radon concentration data from 1351 residential locations. Statistical tests showed the interpolated gamma values were significantly but weakly positively related with indoor radon concentrations, though this relationship is decreasingly informative at finer geographic scales. Geology, gamma readings, and indoor radon were interrelated, with granitic gneiss generally having the highest gamma readings and highest radon concentrations and ultramafic rock having the lowest of each. Our findings indicate the highest geogenic radon potential may exists in the relatively undeveloped southeastern part of the county. It is possible that in situ gamma, in concert with other variables, could offer an alternative to aerial radioactivity measurements when determining radon potential, though future work will be needed to address this project's limitations.

THE NATIONAL RADON ARCHIVE AS A USEFUL TOOL FOR DEVELOPING AND UPDATING THE NATIONAL RADON ACTION PLAN

International recommendations and regulations require developing of National Radon Action Plans (NRAPs) to effectively manage the protection of workers and population from radon exposure. In Italy, a NRAP was published in 2002 and several activities have been carried out in this framework. Information and data regarding these and previous activities have been collected in a National Radon Archive (NRA). Activities carried out by institutionally involved institutes and agencies include several national and regional surveys, involving more than 50 000 indoor environments (dwellings, schools and workplaces), and remedial actions performed in ~350 buildings, largely in schools. Data collected in the NRA allowed also to estimate that lung cancer deaths attributable to radon exposure in Italy are ~3400 per year. On-going developments of the Italian NRA finalized to effectively use it as tool for developing, monitoring and updating the NRAP are also described.

The use of mapped geology as a predictor of radon potential in Norway

Radon exposure is considered to cause several hundred fatalities from lung-cancer each year in Norway. A national map identifying areas which are likely to be exposed to elevated radon concentrations would be a useful tool for decision-making authorities, and would be particularly important in areas where only few indoor radon measurements exist. An earlier Norwegian study (Smethurst et al. 2008) produced radon hazard maps by examining the relationship between airborne gamma-ray spectrometry, bedrock and drift geology, and indoor radon. The study was limited to the Oslo region where substantial indoor radon and airborne equivalent uranium datasets were available, and did not attempt to test the statistical significance of relationships, or to quantify the confidence of its predictions. While it can be anticipated that airborne measurements may have useful predictive power for indoor radon, airborne measurement coverage in Norway is at present sparse; to provide national coverage of radon hazard estimates, a good understanding of the relationship between geology and indoor radon is therefore important. In this work we use a new enlarged (n = 34,563) form of the indoor radon dataset with national coverage, and we use it to examine the relationship between geology and indoor radon concentrations. We use this relationship to characterise geological classes by their radon potential, and we produce a national radon hazard map which includes confidence limits on the likelihood of areas having elevated radon concentrations, and which covers the whole of mainland Norway, even areas where little or no indoor radon data are available. We find that bedrock and drift geology classes can account for around 40% of the total observed variation in radon potential. We test geology-based predictions of RP (radon potential) against locally-derived estimates of RP, and produce classification matrices with kappa values in the range 0.37-0.56. Our classifier has high predictive value but suffers from low sensitivities for radon affected areas. We investigate an alternative classification method based on a Naïve Bayes classifier which results in similar overall performance. The work forms part of an ongoing study which will eventually incorporate airborne equivalent uranium data, as and when new airborne data become available.

Relationships between indoor radon concentrations, thermal retrofit and dwelling characteristics

A monitoring campaign was conducted on a sample of more than 3400 dwellings in Brittany, France from 2011 to 2014. The measurements were collected using one passive dosimeter per dwelling over two months during the heating season, according to the NF ISO 11665-8 (2013) standard. Moreover, building characteristics such as the period of construction, construction material, type of foundation, and thermal retrofit were determined using a questionnaire. The final data set consisted of 3233 houses with the measurement results and the questionnaire answers. Multivariate linear regression models were applied to explore the relationships between the indoor radon concentrations and building characteristics, particularly the thermal retrofit. The geometric mean of the indoor radon concentration was 155 Bq m^-3 (with a geometric standard deviation of 3). The houses that had undergone a thermal retrofit had a higher average radon concentration than those that had not, which may have been due to a decrease in air permeability of the building envelope following rehabilitation work that did not systematically include proper management of the ventilation. Other building characteristics, primarily the building material and the foundation type, were associated with the indoor radon concentration. The indoor radon concentrations were higher in older houses built with granite or other stone, with a slab-on-grade foundation and without any ventilation system.

Geostatistical simulations for radon indoor with a nested model including the housing factor

The radon prone areas definition is matter of many researches in radioecology, since radon is considered a leading cause of lung tumours, therefore the authorities ask for support to develop an appropriate sanitary prevention strategy. In this paper, we use geostatistical tools to elaborate a definition accounting for some of the available information about the dwellings. Co-kriging is the proper interpolator used in geostatistics to refine the predictions by using external covariates. In advance, co-kriging is not guaranteed to improve significantly the results obtained by applying the common lognormal kriging. Here, instead, such multivariate approach leads to reduce the cross-validation residual variance to an extent which is deemed as satisfying. Furthermore, with the application of Monte Carlo simulations, the paradigm provides a more conservative radon prone areas definition than the one previously made by lognormal kriging.