Towards a method for forecasting earthquakes in Iceland using changes in groundwater chemistry

Predicting earthquakes remains a significant challenge in seismology. While various methods exist for assessing long-term seismic hazard, short-term forecasting remains elusive. This research explores a novel approach leveraging changes in groundwater chemistry as potential earthquake precursors. The study focuses on northern Iceland, a region with known geothermal activity and frequent seismic events, making it an ideal location to investigate this relationship. Previous research, including the authors' 2014 publication, has documented correlations between changes in groundwater chemistry and subsequent earthquakes. This prior work demonstrated significant alterations in the isotopic composition (δ²H and δ¹⁸O) and concentration of elements (Na, Si, Ca, K) in groundwater samples from the Hafralækur well (HA01) before and during two M≥5 earthquakes in 2012 and 2013. The current study aims to evaluate the predictive power of these geochemical changes by testing the hypothesis that similar or larger earthquakes occurring between 2014 and 2023 could have been forecast using this method. The success of this approach could provide a valuable supplementary tool for existing seismic hazard assessment methods, especially in areas with extensive groundwater monitoring infrastructure.

The use of groundwater chemistry as an earthquake precursor has been suggested for over 50 years. Early studies noted precursory variations in deuterium (²H) and sodium (Na) concentrations. More recent research, including the authors' 2014 study, has provided further evidence linking geochemical anomalies to seismic activity. These studies highlight the potential of integrating multiple geochemical parameters and statistical analyses to improve the reliability of earthquake forecasts. However, the specific mechanisms driving these geochemical changes and the limitations of using such methods for prediction need further investigation. The literature review also includes discussions on stress triggering, fault zone properties and the limitations of previous attempts at earthquake forecasting based on geochemical data. The research highlights the importance of long-term monitoring and careful analysis of groundwater data for developing effective forecasting models.

This study utilizes a decade-long (2014-2023) dataset of groundwater chemistry measurements from the Hafralækur well (HA01) in northern Iceland. The well penetrates basaltic rocks and basalt-derived sediments, providing a continuous record of groundwater conditions. Weekly samples were collected for analysis of stable isotopes (δ²H and δ¹⁸O) and dissolved element concentrations (Na, Ca, Si, K). To enhance the interpretation of isotopic data, the authors calculated parameters 'd' (deuterium excess, a proxy for water-rock interaction) and 'd'' (a proxy for mixing between different meteoric water sources). Earthquake data were obtained from the USGS and Global CMT Project. The authors employed a variety of analytical techniques including ICP-OES for element analysis and cavity ring-down spectroscopy for stable isotope measurements. Statistical analyses focused on identifying statistically significant changes in groundwater chemistry preceding M≥5 earthquakes. True positive rates (TPR), false positive rates (FPR), and positive predictive values (PPV) were calculated to assess the method's forecasting performance. A binomial test was used to assess the likelihood of earthquakes occurring within 4–6 months of groundwater chemistry changes compared to random occurrence. Furthermore, spectral analysis was performed to identify periodicities in the oscillatory groundwater responses. The analysis considers the strain radius around the epicenter of each earthquake based on established empirical relationships and compares this with the co-seismic dilational or compressional strain at the Hafralækur well.

The analysis of groundwater chemistry data from 2014 to 2023 revealed that the largest M≥5 earthquake (M 6.0) in the study period could have been forecast using the proposed method. This method involves identifying statistically significant changes in the parameter d' (a proxy for mixing of meteoric waters), which exceeded 2σ from its cumulative mean and lasted for 1, 2, or 3 consecutive measurements. The sensitivity (TPR) of the method ranged from 20% to 32%, meaning there is a 20-32% chance of forecasting an earthquake using this method. The false positive rate (FPR) was low (1-3%), indicating a low probability of false alarms. The positive predictive value (PPV) was high (62-85%), suggesting a high probability that a forecast would correctly predict an earthquake. Oscillatory patterns in the groundwater chemistry data, with a periodicity of about 75 days, were observed before several earthquakes. This oscillatory behavior is hypothesized to be related to cyclic expansion and contraction of the groundwater source region in response to crustal dilation coupled with mineralisation along microfractures. While element concentration anomalies were observed, only those associated with the 2012 and 2013 earthquakes were statistically significant. Co-seismic dilational strains at Hafralækur were small, suggesting that the observed groundwater responses are likely caused by micro-fracturing.

The findings of this study support the hypothesis that changes in groundwater chemistry can serve as precursors to earthquakes, at least in certain geological settings. The relatively high PPV and low FPR of the forecasting method suggest that it could provide a useful complement to other probabilistic earthquake forecasting methods, especially given its relatively shorter (4-6 month) prediction window compared to others that might offer decades-long forecasts or are focused on the days preceding aftershocks. The oscillatory nature of the geochemical changes and their association with micro-fracturing provide insights into the underlying physical processes causing these anomalies. The relatively low sensitivity (20-32%) emphasizes the need for further research to improve the reliability of the method and to explore its applicability to other regions. The study does not rule out the potential role of other geochemical parameters or different statistical approaches in enhancing forecasting accuracy.

This study demonstrates the potential of using changes in groundwater chemistry as a short-term earthquake forecasting tool. The method, while site-specific, shows promising results in northern Iceland, successfully forecasting one of three M≥5 earthquakes. The oscillatory nature of the geochemical signal provides insights into the underlying physical mechanisms. Further research should focus on improving the sensitivity of the method, testing it prospectively, and investigating its applicability to other geological settings. The successful implementation of similar methods elsewhere depends critically on long-term high-quality groundwater chemistry datasets.

The study's primary limitation is the retrospective nature of the analysis. While the findings are consistent with the possibility of earthquake forecasting, prospective testing is crucial to validate the method's predictive capability. The site specificity of the method is another limitation; the unique geological and hydrogeological conditions in northern Iceland might not be replicated elsewhere. The relatively low sensitivity highlights the need for further research to refine the forecasting approach and to potentially identify additional parameters for improved prediction. Future work should also consider other potential influencing factors on groundwater chemistry that could generate false positives or negatives.