Estimation of radon flux spatial distribution in Rize, Turkey by the artificial neural networks method

Abstract

In this study, average radon flux distribution in the Rize province (Turkey) was estimated by the artificial neural networks (ANN) method. For this purpose, terrestrial gamma dose rate (TGDR), which is defined as an important proxy in determining radon flux distribution, was used. Input parameters that were used for ANN were the natural radionuclide (U-238, Th-232 and K-40) activity values in soil samples taken from 64 stations in Rize Province, data from ambient gamma dose rates (AGDR) directly affecting the distribution of radon flux and data of geographical coordinates. Randomly chosen 42 stations were used for ANN training and data from 22 stations were used for testing the ANN model. Performance test results gave a Pearson's r value of 0.60 (p < 0.001) and RMSE of 0.296. The area that was used for the model was divided into grids of 100 m by 100 m and a spatial distribution map was composed by using ANN predicted radon flux rates at grid nodes, whereby natural radionuclide values and Ordinary Kriging predicted values of external gamma dose rates were used for composing the map.