Abstract Radon in groundwater has long been recognized as a sensitive indicator of crustal stress. Significant changes in groundwater radon concentration before earthquakes have been documented in many studies. However, the radon concentration in groundwater may be affected by many interference factors. The anomalies before seismic activities may not be large enough to be clearly identified by the conventional statistical method. Therefore, new methods are needed to identify the possible pre-seismic anomalies. In this study, we investigated 38 years’ worth of radon time series data (1977–2015) in a hot spring to identify the possible precursor anomalies. We first identify the factors that may affect the radon fluctuations by wavelet coherence analysis, spring discharge, water temperature, rainfall and barometric pressure. All are found to be closely related to the radon fluctuation. The time series (1980–2008) were used for further decision tree analysis as a high correlation in the duration. Following this, we constructed the decision tree models based on these factors to model the “background” radon fluctuation and identify the anomalies by comparing the difference between the observed radon changes and the “background” fluctuations. The modeled “background” fluctuation is closely related to the observed data during the non-seismic activity period, with the correlation coefficient of 0.8. Following this, we compared the modeled “background” fluctuation of the radon time series with the observed one during the seismic activities period. The decision tree could identify 15 possible radon anomalies among the 24 chosen earthquakes. The identified anomalies are also supported by the anomaly changes in water temperature and spring discharge. Therefore, we believe that the decision tree method could be an efficient way to identify the possible precursor anomalies in future studies. Additionally, we explore the mechanism of radon anomalies. The plausible mechanism for the anomalous increase is that radon is continuously supplied from newly formed internal surfaces of the crack to the aquifer system. For the anomalous decrease, it might be related to radon partitioning into the gas phase and the change of mixing ratio of shallow and depth water.

中文翻译：

---

决策树法识别地下水氡前体异常

摘要 地下水中氡一直被认为是地壳应力的敏感指标。许多研究记录了地震前地下水氡浓度的显着变化。然而，地下水中氡浓度可能受到多种干扰因素的影响。地震活动前的异常可能不足以被常规统计方法清楚地识别。因此，需要新的方法来识别可能的震前异常。在这项研究中，我们调查了一个温泉中 38 年的氡时间序列数据（1977-2015），以确定可能的前兆异常。我们首先通过小波相干分析、泉水流量、水温、降雨量和气压确定可能影响氡波动的因素。发现所有这些都与氡波动密切相关。时间序列（1980-2008）被用于进一步的决策树分析，作为持续时间的高相关性。在此之后，我们基于这些因素构建了决策树模型，对“背景”氡波动进行建模，并通过比较观测到的氡变化与“背景”波动之间的差异来识别异常。模拟的“背景”波动与非地震活动期的观测数据密切相关，相关系数为0.8。在此之后，我们将氡时间序列的模拟“背景”波动与地震活动期间观测到的波动进行了比较。决策树可以在 24 个选定的地震中识别 15 个可能的氡异常。水温和泉水流量的异常变化也支持了识别出的异常。因此，我们相信决策树方法可能是在未来的研究中识别可能的前兆异常的有效方法。此外，我们探索氡气异常的机制。异常增加的合理机制是氡从裂缝的新形成的内表面连续供应到含水层系统。对于异常减少，可能与氡分入气相和浅水和深水混合比的变化有关。我们探索氡气异常的机制。异常增加的合理机制是氡从裂缝的新形成的内表面连续供应到含水层系统。对于异常减少，可能与氡分入气相和浅水和深水混合比的变化有关。我们探索氡气异常的机制。异常增加的合理机制是氡从裂缝的新形成的内表面连续供应到含水层系统。对于异常减少，可能与氡分入气相和浅水和深水混合比的变化有关。