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

An advance has been made towards a method for forecasting earthquakes several months before they occur. The method relies on changes of groundwater chemistry as earthquake precursors. In a study published in 2014, we showed that changes of groundwater chemistry occurred prior to and were associated with two earthquakes of magnitude 5 and higher, which affected northern Iceland in 2012 and 2013. Here we test the hypothesis that similar or larger earthquakes could have been forecast in the following decade (i.e. 2014–2023) based on our published findings. We found that we could have forecast one of the three greater than magnitude 5 earthquakes that occurred. Noting that changes of groundwater chemistry were oscillatory, we infer expansion and contraction of the groundwater source region caused by coupled crustal dilation and fracture mineralisation associated with the stress build-up before earthquakes. We conclude by proposing how our approach could be implemented elsewhere. Changes of groundwater chemistry have been reported before earthquakes for more than 50 years. These changes include precursory variations in δ2H and Na concentration. Here, we test the hypothesis that M ≥ 5 earthquakes affecting the same geographical region could have been forecast based on the findings published in our 2014 study. We test this hypothesis by performing a forecast analysis for the time period from 2014 to 2023, i.e., the decade after our 2014 study. In this analysis, we calculate (1) true positive rates (TPR) (i.e., sensitivity, meaning the probability of forecasting an earthquake) and (2) false positive rates (FPR) (meaning the probability of raising a false alarm) on the basis of statistically significant changes of the chemistry of the groundwater within the preceding 4–6 months. We also calculate positive predictive values (PPV) for an earthquake actually having occurred within 4–6 months of such a forecast. We found that, of the M ≥ 5 earthquakes which affected northern Iceland between 2014 and 2023, the strongest one (M 6.0) could have been forecast on the basis of statistically significant changes of the chemistry of the groundwater within the preceding 4–6 months. The weaker ones (M 5.0 and M 5.2) could not. The corresponding sensitivity was 20–32%. In addition, we calculated a false positive rate of 1–3% and a positive predictive value of 62–85%. Because this analysis was performed a posteriori, we do not claim to have forecast an earthquake. Instead, our results are consistent with a probability for forecasting a future M > 5 earthquake in our study area 4–6 months before it occurred of 20–32%. Noting this time frame, we argue that the method we propose could offer a useful complement to other probabilistic approaches with longer (e.g. 30 years) forecast windows for mainshocks and shorter (e.g. 7 days) for aftershocks. We further note that, although the method we propose is site-specific, similar approaches could also be developed elsewhere. Their successful implementation would rely on regular measurements of groundwater chemistry over a prolonged period of time.

Alasdair Skelton, Erik Sturkell, Carl-Magnus Mörth, Gabrielle Stockmann, Sigurjón Jónsson, Andri Stefansson, Lillemor Liljedahl-Claesson, Niklas Wästeby, Margareta Andrén, Elin Tollefsen, Jóhann Gunnarsson Robin, Nicole Keller, Halldór Geirsson, Hreinn Hjartarson, Ingrid Kockum