Utveckling av skredtrösklar från grundvattennivåer i Seyðisfjörður, Island

Elevated groundwater levels are a primary trigger for water-rich landslides that commonly occur in the study area of Neðri-Botnar, Seyðisfjörður, Iceland. This study develops and evaluates locally calibrated landslide risk thresholds that are based on ground water elevation records. In the study area there are four reference and six comparison boreholes that are actively logging water levels. The reference boreholes have records from September 2020 to January 2025, while the comparison boreholes start recording later. Missing records were estimated using a simple linear regression model in which a linear regression is fitted to filtered data between a reference borehole and a newer borehole. The borehole pair with the best R2 fit was chosen to be the model and the linear regression equation was used to estimate past values using the record of the reference borehole in question. To check the validity of this method identity, residual vs fitted, and Q-Q graphs were plotted and analysed. Using the December 2020 landslide event, an event where multiple landslides fell after intense rainfall, two water-level thresholds were constructed for each borehole. The thresholds were crafted by finding the recorded or estimated minimum water level during a landslide day and adding a six-hour buffer to it. An unsafe threshold was designated to include the mean six-hour rise added to it, and an uncertain threshold included the maximum six-hour rise. These thresholds were applied to the whole time series, where missing values were estimated using the model, and then analysed. The performance of the individual borehole thresholds proved to be relatively good. And when all reliable boreholes were combined, through a unanimous voting rule, the unsafe threshold achieved perfect recall, reduced false positives to below 1%, specificity to above 99% and precision to ≈ 43%. Compared to a conventional precipitation intensity duration alarm, which produced both false positives and unheeded warnings, the groundwater level thresholds demonstrate very few false alarms, capture snow melt effects and have a stronger connection to slope stability and subsurface hydrology. These findings indicate that simple yet robust water level thresholds can improve local early warning systems and serve as a useful hazard indicator alongside precipitation derived ones.