Validation of the NGU susceptibility model for debris avalanches and small to medium sized debris flows in Sørfjorden and Hallingdal Validering av NGUs aktsomhetskart for jordskred og små til mellomstore flomskred i Sørfjorden og Hallingdal
This thesis deals with the validation of the NGU susceptibility map for debris slides/avalanches and small to medium sized debris flows. The susceptibility model developed by NGU is a nation-wide model, predicting source areas, transport paths and runout lengths for events across all of mainland Norway. As a baseline for construction of housing, industry and infrastructure, it is important to have an estimation of how the model performs when applied to the real world. Two study areas were chosen for this purpose, Sørfjorden and Hallingdal. These areas are important gateways between Eastern and Western Norway, and represent different climatic and topographic conditions that enables insight into how the model performs under different conditions. A landslide inventory based on events found in the www.skrednett.no database, as well as events found by manual scanning of orthophotos, was created for both study areas. The type of event, approximate age and spatial distribution was determined for each event before being mapped by free-form polygons in ArcMap. A total of 275 and 31 events were identified and mapped in Sørfjorden and Hallingdal, respectively, most of which are debris flows. Based on the landslide inventory and NGU susceptibility model, a frequency analysis calculating the average number of modelled and mapped source areas per km2 was performed for both study areas, resulting in a ratio of 1 true event for every 4 modelled events in Sørfjorden, and 1 true event for every 40 modelled events in Hallingdal. The validation was done using confusion matrixes, calculating rates indicating model performance based on observations of overlapping modelled and mapped source areas. Only observations of source areas were used in the validation due to wanting to investigate source areas and runout separately. Because of a large imbalance between the positive and negative observations, it is difficult to give an exact number indicating model performance, but an estimate of 83% correct predictions is calculated for Sørfjorden. The performance of the NGU susceptibility model in Hallingdal is more uncertain, but it is not expected to differ greatly from the rate calculated for Sørfjorden. Manual measurements of the distance between the modelled and mapped source areas and runouts were performed for both study areas. It was found that a clear majority of the mapped runouts are confined by the model, with only 3 events in total extending beyond the modelled runout. A higher number of mapped source areas are located outside the boundaries of the model.