Landslide occurrence as a response to large-scale synoptic weather types, precipitation and soil saturation in southern Norway

Debris flows and debris slides/avalanches (weather-induced landslides) represent a significant threat to infrastructures and human habitation in Norway. They are triggered by a combination of high intensity rainfall, snowmelt, high groundwater levels and high soil moisture, with rainfall being the most significant factor. They are most common in mountainous areas, in incised valleys of otherwise low relief, on steep slopes adjacent to fjords and in submarine environments. To reduce the risk and impact of landslide activity, Norway has recently established a landslide early warning system at the Norwegian Water Recourses and Energy Directorate (NVE). This thesis investigates if the large-scale synoptic weather types in combination with rain, snowmelt and soil saturation can be related to the occurrence of weather-induced landslides in southern Norway. This could allow forecasters who operate the early warning system to be prepared further in advance. As a first step, historical landslide data obtained from the Norwegian landslide database were extracted and quality controlled to derive at a final dataset of events for further analysis. This data set was subsequently examined, on a regional scale, by providing landslide statistics for climatic regions for southern-Norway. An attempt to classify the spatial patterns of weather-induced landslide data based on the climatic regions did not yield satisfying results. Improved classification results were obtained when dividing southern Norway into landslide domains (R3a). The classification provides a time series of landslide classes that was compared to time series of precipitation classes based on self-organizing maps (SOM). It showed a clear relationship between the two. Finally, the SynopVis Grosswetterlagen (SVG) classification of daily weather types was compared to the precipitation classification. Although a clear relationship also was observed here, comparing the weather-induced landslide classification to the SVG classification showed a less obvious relationship; indicating that other variables also influence the occurrence of weather-induced landslides. In order to predict the occurrence of landslides within a region, a logistic regression method was used. The dependent variable was the occurrence of landslides within a region (either landslides or no landslides). The independent variables were the SVG classes, mean daily rainfall and snowmelt data obtained from SeNorge2 grids, as well as a hydmet-index already in use in the early warning system. It was first executed on the variables separate, yielding varying results in terms of significance, odds ratio (OR) and predicted probability. An automated model selection tool was then used to select the best combination of independent variables. The results showed that in seven of the twelve regions from R3a the SVGs have the highest predictive power in terms of slide occurrence. In these regions, with the exception of one, the models are significantly better than a null model, and the models are good in predicting weather-induced landslide occurrence. The highest predictive probability of weather-induced landslide occurrence is caused by the weather type Zonal Ridge across Central Europe (BM), which yields a 90 % probability of weather-induced landslides on the west coast. This weather type has also been identified as being related to floods in the same area.