Comparing wavelet-based artificial neural network, multiple linear regression, and ARIMA models for detecting genuine radon anomalies associated with seismic events

This study presents a comprehensive analysis of continuous subsurface soil radon with the environmental parameters and their correlation with seismic events in the Indo-Myanmar subduction zone. Discrete wavelet transformation was applied to the standardized values of soil radon and environmental parameters for denoising, and the resulting data were used as inputs for multiple linear regressions (MLR), multilayer perceptron (MLP) and autoregressive integrated moving average (ARIMA) models. The predicted radon concentrations by the models were then compared using a hypothesis test to select the best models for correlation with seismic events. The correlation coefficient between the measured radon and predicted radon was found to be highest in case of ARIMA model. Furthermore, t-test and the corresponding p value were used to check the significance of the model. Finally, the residual radon was calculated for ARIMA model and the anomalous variation in residual radon was use to correlate with seismic events that occurred around the study area. A significant variation in residual radon was observed few days before a shallow epicenter seismic event of M5.4. The precursor time between the earthquake and observed radon anomaly was calculated to be 24 days. The wavelet based ARIMA model applied was useful to reduce the effect of environmental parameters on soil radon and find a true radon anomaly attributed to seismic event.