Artificial neural network modeling of soil gas radon concentration on different lithologies for Southwest Nigeria

Abstract

Radon is a radioactive gas and a leading cause of lung cancer among non-smokers. An indicator of the level of risk associated with radon inhalation is the Geogenic Radon Potential (GRP) - a function of the soil gas radon concentration and the soil air permeability. Data is scarce on the variation of soil gas radon and GRP across different geological formations in southwest Nigeria. This study was designed to measure and develop a generalized predictive model for soil gas radon concentration and hence GRP on different bedrocks in southwest Nigeria. Soil gas radon was measured with a RAD 7 radon detector in 150 sites randomly selected over 20 major bedrocks. Soil air permeability was derived from soil saturated hydraulic conductivity measured using a constant head permeameter on undisturbed soil samples from the same sites. The GRP was obtained using Neznal relation. A 2 × 8 x 1 Artificial Neural Network (ANN) configuration culminated in the generation of a tested and validated predictive model for soil gas radon in southwest Nigeria. For the 150 sites, the range of values for soil gas radon concentration was 0.28–47.59 kBqm−3 and a mean of 10.39 ± 12.59 kBqm−3; soils on granitic bedrocks had the highest mean soil gas radon concentration (16.89 kBqm−3) and highest mean GRP (13.71). Sediments had the highest mean soil air permeability of 39.84 × 10−12 m2. Using Neznal classification, 75.30% of the sites were of low radon hazard rating. The developed predictive ANN model had a Goodness-of-Prediction (G) of 73.53%, Average Validation Error (AVE) of 0.073, Mean Bias Error (MBE) of 0.42, and Root Mean Square Error (RMSE) of 4.62kBqm−3, respectively. The model, validated using standard procedure had G, MBE, AVE, and RMSE of 86.49%, 0.61, 0.17, and 1.65kBqm−3, respectively which is an indication of good model performance.