Identification of radon anomalies in soil gas using decision trees and neural networks

The transport of radon (222Rn) from the ground towards the surface is influenced by a number of geophysical and geological parameters, among them seismicity. Prior to an earthquake, the formation of stress causes changes in the strain field. The displacement of rock mass within the earth’s crust before an earthquake leads to changes in gas transport from deep layers in the earth to the surface [5]. As a result, larger quantities of radon are released from the pores and fractures of the rocks towards the surface. This may be considered as an anomaly in the concentration of radon. Because of seismicity, changes in underground fluid flow may account for anomalous changes in concentration of radon and its progeny [8]. A small change in velocity of gas [6] into or out of the ground causes a significant change in radon concentration at shallow soil depth as changes in gas flow disturb the strong radon concentration gradient existing between the soil and the atmosphere. A small change in gas flow velocity causes a significant change in radon concentration. Thus, monitoring of radon in soil gas is a means of detecting changes related to an earthquake. For small earthquakes, it is often impossible to identify an anomaly caused by a seismic event and not by meteorological or hydrological events. Therefore, the implementation of more advanced statistical methods in data evaluation appears to be essential [1, 3, 7]. In this contribution, the 32-month time series of radon concentration together with the environmental parameters (air and soil temperature, barometric presIdentification of radon anomalies in soil gas using decision trees and neural networks Boris Zmazek, Saso Džeroski, Drago Torkar, Janja Vaupotic, Ivan Kobal