Connections between precipitation, stream flow and landslide occurrence in Hordaland
county
Sara Kristine Muri Bugge
Thesis submitted for the degree of Master in Geosciences
(Geohazards) 60 credits
Department of Geosciences
Faculty of mathematics and natural sciences
UNIVERSITY OF OSLO
Connections between precipitation, stream flow and landslide occurrence in
Hordaland county
Sara Kristine Muri Bugge
© 2017 Sara Kristine Muri Bugge
Connections between precipitation, stream flow and landslide occurrence in Hordaland county
http://www.duo.uio.no/
Printed: Reprosentralen, University of Oslo
Abstract
Western Norway is each year exposed to landslides, causing considerable damage in property, especially related to infrastructure. These landslides are triggered by a combination of soil moisture and water supply (e.g. precipitation). Although empirical thresholds, developed for the landslide forecasting service in Norway, already exist, a study of the characteristics of the precipitation and stream flow events connected to the landslides have not previously been conducted for Hordaland.
This thesis investigates the features of both the rainfall events that triggered landslides, in terms of intensity and duration, and stream flow events occurring at the same time as landslides. In order to do this, landslide events from the national mass movement database in Norway were analysed and the rainfall connected to the events were investigated. The first step of the analysis involved a thorough review of the mass movement database. The landslides reported in the database were analysed based on their spatial and temporal dis- tribution, and eventually connected to rainfall data from nearby rain gauges, or stream flow data from water discharge stations.
A method developed by Melillo, Brunetti, Peruccacci, Gariano, and Guzzetti (2015) with the purpose of reconstructing rainfall events has been tested for Hordaland. This method involves an algorithm, that was developed for the automatic identification and reconstruction of rainfall events, and it has previously not been tested in Norway.
Principles from the method mentioned above has been adopted in order to be able to char- acterise stream flow events occurring at the same time as landslides, caused by the same rainfall event. A 5 km buffer zone around the rain gauge stations were used in the precipita- tion analysis in order to find landslides where the weather conditions would be represented by the rain gauge station, whereas the full catchment of water discharge stations were used in the corresponding stream flow analysis.
The results show that the method from Melillo et al. (2015) is capable of reconstructing the rainfall events, and that the same principles, with some modification, also can be applied when reconstructing stream flow events. However, many landslides appear to have been re- ported only when the weather conditions triggering the landslides are out of the ordinary, and more investigation and study is required to ensure that the algorithm is able to reconstruct these events properly. For the stream flow events, criteria were suggested for the reconstruc-
tion of such. However, the set criteria met some problems when applied generally, and it was suggested that the characterisation of said events are better used for special cases, such as a landslide event coupled with stream flow data.
Acknowledgements
First of all, I want to thank my supervisors. A big thank you to Graziella Devoli for intro- ducing me to the topic of landslide thresholds, for all the help and suggestions during this period, and for always being available for answering questions, a discussion or supervision.
I also want to thank her for the opportunities at NVE, both in connection with this thesis, and otherwise. I would like to thank my co-supervisor, Kolbjørn Engeland, for good super- vision and support during the work with this thesis, and for helping with the extraction of stream flow data. You have provided lots of good advice along the way, and been helpful at introducing hydrology, for me, a relatively unknown subject.
Thank you Massimo Melillo for all the help you have provided with the algorithm and prepa- ration of the data when I got stuck. Even though it really was not your responsibility, you have taken the time to reply all my e-mails, answering any question I have, explaining and assisting me during my work with the algorithm. It is much appreciated.
Further, I would like to thank all my fellow students. Thank you Linda, for taking the time to read through my thesis, even though you are so busy with your own studies. Thanks to all the students at room 219 at ZEB, for lunch breaks, Friday quizes and both academic and extracurricular discussions.
Thanks to my family for their support and encouragement during my studies, and for never questioning my academic choices. Last, but not least, thank you Magne, most of all just for being you, for your moral support and for putting up with me during ups and downs through my studies.
Sara Kristine Muri Bugge
Contents
1 Introduction 1
1.1 Background . . . 1
1.2 Motivation . . . 4
1.3 Objectives . . . 5
1.4 Thesis outline . . . 5
2 Natural hazards 6 2.1 Extreme weather . . . 6
2.2 Flood . . . 6
2.3 Mass movement . . . 8
2.3.1 Mass movement classification in Norway . . . 9
2.3.2 Snow avalanches and slush flows . . . 10
2.3.3 Landslides in rock . . . 11
2.4 Slope deformation and rock avalanches . . . 13
2.5 Landslides in soil . . . 14
2.6 Triggering factors for landslides in soil . . . 17
3 Literature review 19 3.1 Landslide thresholds in the world . . . 19
3.2 Landslide thresholds in Norway . . . 20
3.3 Rainfall events and stream flow events . . . 22
4 Study area 26 4.1 Climate . . . 27
4.2 Geology . . . 30
5 Data 31 5.1 Landslide data . . . 31
5.2 Precipitation data . . . 31
5.3 Stream flow data . . . 32
6 Methods 34 6.1 Quality control of landslide database . . . 34
6.2 The algorithm to identify rainfall events . . . 34
6.2.1 How the algorithm works . . . 36
6.2.2 Changes made to the algorithm . . . 39
6.3 Identification of streamflow events . . . 39
6.4 Limitations of data . . . 43
6.4.1 Landslide data . . . 43
6.4.2 Precipitation data . . . 43
6.4.3 Stream flow data . . . 44
7 Data processing 45 7.1 Landslide data . . . 45
7.2 Precipitation data . . . 45
7.3 Stream flow data . . . 47
8 Results 49 8.1 Quality control of landslide data . . . 49
8.2 Rainfall events . . . 52
8.3 Stream flow events . . . 56
9 Discussion 61 9.1 Landslide data . . . 61
9.2 Performance of the algorithm . . . 62
9.3 Rainfall events . . . 63
9.4 Stream flow data . . . 67
10 Summary and conclusion 71 10.1 Landslide database . . . 71
10.2 Precipitation events . . . 71
10.3 Stream flow events . . . 72
10.4 Conclusion - use of hydrometeorological data in EWS . . . 73
11 Future work 74
References 76
Appendix 81
A Rainfall event metrics 81
List of Tables
1 A summary of the methods, in reference to the objectives and tools used. . . 34
2 The variables and the values used in the analysis . . . 39
3 List of the precipitation stations used in the analysis . . . 46
4 List of the water discharge stations used in the analysis . . . 47
5 Regulation of watercourses . . . 48
6 Table of extreme weather events connected to dates with high landslide occur- rence . . . 50
7 A summary of the results. Landslides are sorted in four quarters through the year. . . 56
8 Overview of stations used in the analysis, with corresponding number of land- slides within their catchments. . . 57
9 The minimum, maximum and average stream flow values for each station, between 2000 and 2015. . . 59
10 The minimum, maximum and average stream flow values when landslides oc- curred for each catchment. . . 60
List of Figures
1 An example of landslide triggering thresholds, from Cepeda and Devoli (2008).
Option C show the boundary between triggering and non-triggering rainfall events. . . 2 2 The frequency of mass movement events and the yearly costs of such events
along the main infrastructure network in Norway. It is evident that Hordaland has a high frequency in events, and also the highest costs in the country each year. Figure modified from Frauenfelder et al. (2013) . . . 4 3 The flood in the river Opo in Odda, October 2014. Five houses were destroyed
by the water flow, and 45 people evacuated. Photo by Marit Hommedal/NTB Scanpix (Nationen, 2014). . . 7 4 An overview of damages caused by extreme storms and floods in Hordaland
reported by Finans Norge (2017). It is evident that storms (left axis) cause significantly more damage than floods (right axis). . . 8 5 The updated Varnes classification system, reprinted from Hungr, Leroueil, and
Picarelli (2014) . . . 9 6 Classification of mass movements in Norway (reprinted from GEO365 (2013)). 10 7 The figure illustrates the difference in water content between a snow avalanche
and a slush flow. . . 11 8 Rock falls can come in various sizes, as the images in a) and b) illustrates. . 12 9 Overview of spatial distribution of all rock falls and rock slides (types 110,
111, 112, 113 in the national mass movement database) in Hordaland between 2000 and 2015. The landslide database contains 4025 rock falls and rock slides. 13 10 A screenshot from the online map service from NGU, showing the areas in
Hordaland where slope deformation has been mapped. . . 14 11 Overview of spatial distribution of all types of landslides in soil in Hordaland
between 2000 and 2015 . . . 15 12 The debris slide that struck Hatlestad Terrasse in September 2005. Photo by
Hallgeir V˚agenes/VG/Scanpix (Aftenposten, 2011). . . 16 13 Debris flows in Bogelia, Vaksdal, in May 2015. Photo by Rune Sævig (Bergens
Tidende, 2015). . . 16 14 The thresholds used in the EWS in Norway. The colors represent the levels,
green (1) to red (4), with increasing degree of landslide danger. . . 23 15 The different rainfall patterns discovered in the study of the Barcelonnette,
France, area. From Cepeda, Colonnelli, Meyer, and Kronholm (2012) . . . . 24
16 Hordaland, situated on the west coast of Norway . . . 26 17 The normal precipitation and temperature curve for Bergen - Florida. Figure
is produced from www.eklima.no . . . 28 18 Precipitation and temperature normals for the south of Norway 1971-2000.
Data is collected from www.senorge.no. . . 28 19 Normal annual amount of snow and snow depth for the south of Norway 1971-
2000. Data collected from www.senorge.no. . . 29 20 A visual overview of the 24 catchments and the corresponding stations within
Hordaland. The station ID (e.g. 63.12) corresponds to the list of stations in table 4 (presented in a later chapter). . . 33 21 A framework for the algorithm, from Melillo et al. (2015) . . . 35 22 These rainfall sub-events, together with the reconstructed full rainfall event
are examples of the output from the algorithm. . . 38 23 The stream flow curve consists of two components: the baseflow and the direct
flow. The figure is a simplification, just to illustrate the basic components of the stream flow curve. . . 40 24 An illustration of some of the principles of hydrograph separation. Three main
techniques are shown; the constant slope method, the concave method and the constant-discharge method. Figure from McCuen et al. (1989). . . 41 25 An example of the graphs used to visualise the stream flow and landslide data.
The dates are given with a 48 hour resolution. . . 42 26 The raw data as downloaded from eKlima for the Eikanger Myr station. The
downloaded file includes other data such as snow depth (SA) and snow cover (SD) and temperature (TAM/TAN/TAX). However, only the precipitation data (RR) has been used from this file. . . 46 27 The same data as shown in the previous figure, but transformed into the format
compatible for the algorithm. Note how the daily precipitation value has been distributed over several hours. . . 47 28 The distribution of landslides in Hordaland, n=674. . . 49 29 The distribution per year of soil landslides used as a base for the analyses of
this thesis, n=674 . . . 51 30 The distribution of landslides per month, per year, n=674 . . . 51 31 The typology of the 674 landslides. The number behind the typology indicates
how the landslides are registered in the landslide database. . . 52 32 The distribution of landslides within buffer zones of the 25 rain gauges used
in the analysis. Number of landslides: 195. . . 53
33 The distribution of landslides used in the rainfall event analysis, per year, n=195 54 34 Overview of the seasons when the landslides occurred. N=130 landslides. . 54 35 These rainfall sub-events, together with the reconstructed full rainfall event
are examples of the output from the algorithm. . . 55 36 The 6 catchments and 64 landslides used in the stream flow analysis. The
names refer to the station each catchment is connected to. . . 57 37 An example of two landslides that occurred in the catchment of Sandvenvatn
station. One landslide occurred a few days before the largest peak, while the other landslide occurred just after the peak. The figure illustrates good correspondence between the landslide data and the stream flow data. . . 58 38 The landslide occurred well before the rise in stream flow. This example is
from Bulken, and the landslide might not be rainfall induced. . . 58 39 An example of the results from the 46450 Røldal station. Four landslides were
reconstructed from this station, all with similar results. . . 64 40 An example of the division of stream flow events (between red bars) with
sub-events (shaded area between green bars). . . 69
1 Introduction
1.1 Background
A natural hazard is “any natural process or phenomenon that may cause loss of life injury or health impacts, property damage, loss of livelihoods and services, social and economic disruption or environmental damage” (UNISDR, 2009). Such hazards can have geologic (e.g.
earthquake, landslide, volcanic eruption), atmospheric (e.g. storms, precipitation1, snow), hydrologic (e.g. floods, drought) or epidemic (e.g. diseases, wildfires) origin (Smith, 2013).
Common for all is that they are caused by changes in the natural environment.
Landslides and snow avalanches cause major disasters on a global scale every year (Nadim, Kjekstad, Peduzzi, Herold, & Jaedicke, 2006). Because landslides can be triggered by a variety of factors, such as heavy rain or earthquakes, they can occur in many places for a number of reasons. There are many examples of major landslide disasters, on all continents (excluding Antarctica), such as the Vargas tragedy, Venezuela, in 1999 where more than 30 000 people were killed (Wordatlas, 2017).
Some natural hazards, such as landsldies, can be impossible to prevent, however, there are ways to reduce the damage potential. Early warning systems (EWS) can offer powerful and cost-effective alternatives to traditional structural measures, and makes it possible to provide early information about a possible event before it occurs (Geoprævent, 2017). An EWS is defined by the United Nations as “the set of capacities needed to generate and disseminate timely and meaningful warning information to enable individuals, communities and organi- zations threatened by a hazard to prepare and to act appropriately and in sufficient time to reduce the possibility of harm and loss” (UNISDR, 2009). Many different types of EWS for landslides exists, such as alarm, warning and forecasting systems (St¨ahli et al., 2015), each with their specific areas of application. Alarm systems are often connected to an ongoing hazard, such as at he Illgraben debris flow alarm system. Warning systems are directed towards detecting changes in the environment, such as opening of cracks, before the event occurs (St¨ahli et al., 2015). The surveilance and investigation connected to ˚Aknes in Møre and Romsdal is an example of a warning system.
Many forecasting systems base their warnings on thresholds. A threshold is the minimum or maximum level of a quantity that is needed for a process to take place (White, Motter-
1Precipitation in this thesis involves only rainfall. Rainfall and precipitation are used throughout this thesis both referring to the same.
shead, & Harrison, 1996). When used in connection with rainfall-induced landslides, such thresholds may be linked to soil moisture, rainfall or other hydrological conditions that are likely to trigger landslides (Guzzetti, Peruccacci, Rossi, & Stark, 2007). An example of such thresholds is presented in figure 1. The thresholds are either defined on physical or empirical bases. The physical based thresholds are often confined to a single slope, or smaller area, as it is difficult to acquire detailed information of hydrological, morphological, geological and soil characteristics for a larger area (Guzzetti et al., 2007).
The empirically based thresholds are, on the other hand, based on records of historical landslide events and their corresponding rainfall metrics, such as the hourly precipitation measured at a nearby rain gauge. The minimum threshold means that no landslides will occur for conditions below a set threshold, while a maximum threshold indicates that landslides always will occur when this threshold is exceeded.
Figure 1: An example of landslide triggering thresholds, from Cepeda and Devoli (2008). Option C show the boundary between triggering and non-triggering rainfall events.
Norway has an operative landslide forecasting system, which was launched in 2013 with the goal to avoid losses of lives and properties as a result of landslides. The Norwegian Water Resources and Energy Directorate (NVE) is responsible for the day-to-day operation of the
predicting the level of danger of mass movements based on monitoring the causative condi- tions, such as precipitation and soil moisture. Like for other forecasting systems, the data interpretation is conducted on a regular basis, where experts analyse sensor data and consult models to forecast the regional danger levels (St¨ahli et al., 2015).
Landslides are a common feature in Norway. Vestlandet (Western Norway) is each year exposed to different natural hazards and landslides are amongst the most frequent. The topography is dominated by steep mountain sides, fjords and valleys. A network of highways and railways along the mountainsides, and villages in the valley bottoms and along the fjords make the infrastructure and inhabitants of Hordaland possible targets of landslide accidents.
In 2016 alone, landslides and snow avalanches caused damage for more than 4 million NOK (Finans Norge, 2017) in Hordaland. In addition to direct damage cost, various indirect costs are generated due to landslides. Br˚athen, Husdal, and Rekdal (2008) calculated that the cost of closing a road due to landslides costs between 2000 and 80 000 NOK per hour for the 7 counties most exposed to landslides in Norway, in which Hordaland is included. According to the InfraRisk-project, closed roads, clearing up after an event and the repairs connected to the event cost more than 100 million NOK each year. Figure 2 show a visual presentation of the costs of mass movement events in Norway, showing that both the number of events each year and the costs related to mass movement events in Hordaland is substantial. EWS are well established tools that can aid in the process of reducing the impacts of a natural hazard, both regarding the prevention of loss of life, and the reduction of economic and material losses.
It is well established that many landslides are triggered by precipitation (Caine, 1980; Hungr et al., 2014; Jakob & Weatherly, 2003; Guzzetti et al., 2007; Melillo et al., 2015), therefore many forecasting systems apply precipitation thresholds or other thresholds, as mentioned, when “forecasting” possible landslide events. However, Guzzetti et al. (2007) argues that standards for rainfall conditions are lacking. The literature reviewed in the article show that different researchers have different aproaches to their investigations. This causes a variety in the results, in terms of the periods of antecedent rainfall used in the studies. The reconstruc- tion of rainfall events are often done subjectively by expert observers with varying emphasis on the duration of antecedent rainfall when defining thresholds. An automated method for rainfall conditions reconstruction will not only save time, as the manual identification of rain- fall events is a time consuming processes, but also reduce the uncertainty in defining rainfall events that triggered landslides and remove the element of subjective assessments (Melillo et al., 2015).
Figure 2: The frequency of mass movement events and the yearly costs of such events along the main infrastructure network in Norway. It is evident that Hordaland has a high frequency in events, and also the highest costs in the country each year. Figure modified from Frauenfelder et al. (2013)
1.2 Motivation
Although many studies focus on thresholds, not many studies on the caracteristics of the rainfall events triggering the landslides has been conducted. As an attempt to mitigate this, Melillo et al. (2015) developed an algorithm with the goal of objectively reconstruct rainfall events that have resulted in landslides. This algorithm is in this thesis tested on landslide triggering rainfall conditions in Norway for the first time.
Similarly, little emphasis has been put on stream flow data in landslide investigations. In places where the precipitation stations are not present a stream flow station might provide an indication of the weather conditions in the area. They might also be useful for conditions where other factors, that precipitation gauges are unable to measure such as snow melt, are dominating. Stream flow data is already used for flood warning, but could be used for landslide warning as well, as an indication of the soil moisture content. Floods and landslides often occur at the same time, or during the similar weather conditions.
1.3 Objectives
This thesis will investigate how EWS can get increasingly better by using data from multiple sources. It will therefore be important to investigate the applicability of rainfall events infor- mation and stream flow data in the EWS for weather-induced landslides. The main objective of the study is to identify the typical meteorological conditions (in terms of precipitation in- tensity and duration) and hydrological conditions (in terms of stream flow) during landslide events in Hordaland, that can contribute to a more specific landslide warning. This aim will be achieved through a series of secondary objectives:
• characterise landslide events by analysing their spatial and temporal distribution.
• characterise rainfall events in terms of rainfall intensity and duration by the use of a specific algorithm (explained in section 6.2).
• identify the typical levels of stream flow events (if any) during landslide occurrence by applying the basic principles from the algorithm to stream flow data.
1.4 Thesis outline
This thesis is split into several chapters. Chapter 2 presents the main natural hazards found in Hordaland. Many types of natural hazards are introduced, such as extreme weather, flood and mass movement issues of the area, but a special emphasis is placed on landslides in soil. The theoretical background of landslide thresholds, and rainfall and stream flow events connected to landslides, is presented in chapter 3, and Hordaland as study area is presented in chapter 4. The chapter gives the basic information on the topography, climate and geology of the area. The following chapters involve the process of conducting three analyses in order to answer the objectives stated above, and the chapters are arranged to fit each of the secondary objectives related to the landslide events, rainfall events or stream flow events.
The data and methods used for the analyses are presented in chapters 5 and 6, respectively.
The methods chapter gives an outline of the algorithm used, in addition to explaining how the same principles for identification of rainfall events can be applied to stream flow data.
The methods involves a degree of data processing in order to prepare the data for use with the algorithm, and this process is explained in chapter 7. Results are presented in chapter 8, and further discussed in chapter 9. The discussion is summarized and a conclusion is given in chapter 10.4. Some suggestions for future work are given in chapter 11.
2 Natural hazards
Hordaland is each year exposed to natural hazards, that cause large problems, especially for the infrastructure. NVE’s plan to identify which municipalities and counties that should be prioritized for mass movement hazard mapping (Øydvin et al., 2011), show that the Vest- landet counties, including Hordaland, are where mapping should be prioritised, because many people and properties in the area are exposed to potential mass movement accidents.
2.1 Extreme weather
The term “extreme weather” is, in Norway, derived from the Norwegian Meteorological Insti- tute’s (MET) extreme weather warnings (Meteorologisk Institutt, 2017a). These warnings are sent out when the weather situation can threaten life and values if precautions are not made (Meteorologisk Institutt, 2017a). Phenomena that can cause dangerous weather situations are strong winds, large amounts of precipitation, storm surges and waves, or a combination of these. Floods, landslides and snow avalanches can be a result of extreme weather situations.
Storms cause the most damage of the weather situations mentioned above. More than 35000 reported damages, see figure 4, caused by storm occurred in Hordaland between 1980 and 2017 (Finans Norge, 2017). As figure 4 show, two years stand out; 1994 had 6919 reported damages, while 2015 had 10074. Meteorologisk Institutt (2017b) has published a list of all the extreme weather situations in Norway since December 1994. The extreme weather Nina in 2015 can account for most of the reported damages in 2015, causing damage of around 450 million NOK (Finans Norge, 2017). The most recent extreme weather that affected Horda- land was named Vidar, and occurred in January 2017. This extreme weather was related to high tidal water, low air pressure and onshore wind, thus affecting the coastal areas of Hordaland the most.
2.2 Flood
Floods cause damage in Hordaland each year. Floods are a hydrological hazard, in Norway most often caused by excessive snow melt and precipitation (Engeland et al., 2016). Floods occur when water rises and inundates land (Smith, 2013). They are a result of, but not lim- ited to, overflowing rivers, frozen soils preventing water to infiltrate, or water from the sea invading low-lying coastal areas (Smith, 2013). Floods are measured by the peak river flow, discharge (m3s−1). The stage of a flood indicates the maximum height the water reaches, and this is often what is important in terms of hazard potential.
Some floods occur in combination with extreme weather events, such as the floods in Oselva, Opo and Vosso in 2005 (extreme weather Kristin). Roald (2013) mentiones several rivers in Hordaland with flood events, both from historical and recent times. The oldest news articles reviewing floods in Norway dates back to 1789 and the flood called Storofsen (Roald, 2013), but other sources, such as flood marks on rocks or buildings have been found, dating back even further. A mark carved into the wall of the church at Voss dates back to a flood in 1604 (Holmqvist, 2015).
In Hordaland, Roald (2013) highlights especially Hardanger and Sunnhordland as areas with large floods. The most recent, significant floods were the 200-year flood in the Vosso river and large flood in the Opo river in October 2014. The flood in the Vosso river was especially large, the largest flood recorded since the stream flow measurements started at Bulken in 1892 (Holmqvist, 2015). According to Finans Norge (2017), 3019 damages were reported as a result of flood from 1980-2017, costing more than 368 million NOK 2. This is illustrated in figure 4, where especially 2005, but also 2014 stand out.
Figure 3: The flood in the river Opo in Odda, October 2014. Five houses were destroyed by the water flow, and 45 people evacuated. Photo by Marit Hommedal/NTB Scanpix (Nationen, 2014).
2per 27.02.2017
Figure 4: An overview of damages caused by extreme storms and floods in Hordaland reported by Finans Norge (2017). It is evident that storms (left axis) cause significantly more damage than floods (right axis).
2.3 Mass movement
Mass movement is the collective term for displacement of material down a slope under the force of gravity (Smith, 2013). Mass movement is a collective term and includes two main categories; landslides and snow avalanches. Both types can then further be classified into subtypes. Landslides are classified according to the materials involved, or the mode of trans- portation, such as slide or fall (Hungr et al., 2014). According to Varnes (1978), there are three different types of material that each can move down slope in seven different ways, with a total of 29 different subtypes. Flows are the most extensive type of movements, with a total of 12 subtypes. Hungr et al. (2014) proposed an updated version of the Varnes classifi- cation system, in which there are 32 different types of landslides spread over six main types of movement and two main types of materials.
Figure 5: The updated Varnes classification system, reprinted from Hungr et al. (2014)
2.3.1 Mass movement classification in Norway
Terminology differs between Norwegian and English. This thesis uses English terminology, but an overview of the Norwegian terminology and the corresponding English terms (Jensen et al., 2015) is also presented. Figure 6 shows a compilation of how mass movement are classified in Norway. One significant difference between English and Norwegian terminology, is that the Norwegian word “skred”, a general term for all mass movements, includes mass movements in snow. The corresponding word “landslide” in English excludes mass move- ments in snow. This chapter contains definitions, descriptions and examples of the most common landslide types found in Hordaland. The word “landslide” is used as a general term for mass movement in soil, unless otherwise noted.
Figure 6: Classification of mass movements in Norway (reprinted from GEO365 (2013)).
2.3.2 Snow avalanches and slush flows
Snow avalanches are defined as the flow of snow down a slope (NVE, 2016b). Snow avalanches can further be divided into subtypes such as slab avalanches or loose snow avalanches. They can also be classified based on the water content (dry or wet snow avalanches). Snow has the ability to undergo large changes, depending on the relative content of water in the snow pack, thus changing the density (Smith, 2013). The snow pack fails with similar failure me- chanics as landslides, when the shear stress exceeds the shear strength of the material. Most snow avalanches fails on slopes between 30°and 45°(Smith, 2013), often triggered by heavy precipitation, melting or other loading, such as skiers (Smith, 2013).
Slush flows are defined as rapid, floodlike mass movement of saturated snow (NVE, 2013).
The slush flow contains much more liquid water than snow avalanches, and behaves more like flow slides than traditional snow avalanches (Hestnes, 1998). Slush flows can release in slopes of very low inclination, which makes them a hazard not only in areas with steep slopes. Slush flows can be triggered at slopes of only 5°to 25°. Triggering factors are high precipitation rates and rapid snow melt, high water content in the snow or snow avalanches that curb water (NVE, 2013).
At least 185 people have been killed in snow avalanches or slush flows in Hordaland in the last 500 years (Furseth, 2006). The oldest known written record about a snow avalanche
(a) Snow avalanche in Odda in 1993. Photo by Grind.no (2015)
(b) A slush flow in Jostedalen. Photo by NRK (2013)
Figure 7: The figure illustrates the difference in water content between a snow avalanche and a slush flow.
accident is from year 1180. One of the worst snow avalanche disaster in Hordaland in more recent time, according to Furseth (2006), occurred in 1947, when 13 students were struck by a snow avalanche near Voss, killing five of them.
2.3.3 Landslides in rock
Landslides in rock includes all types of landslides involving bedrock. They can further be subdivided into rock falls or rock slides. These types are the most frequent landslides in rocks in Hordaland.
Rock falls are rock fragments falling, bouncing or gliding down a mountain slope. Rock falls occur in fragmented rock, and are often associated with climate and geology. Weathered rock slopes develop fracture systems, and these systems will eventually lead to failure. Rock slides are characterised by the detachment of a rock mass that disintegrates during the de- scent down a slope. The volume of a rock slide is larger than that of rock falls, and they move rapidly downslope. The material from both rock falls and rock slides deposits in a scree where the landscape flattens (Devoli et al., 2011).
The topography of the area, geology and level of weathering and fracture pattern in the rocks are all factors which has an impact on the release of either rock falls or rock slides.
Increased pore water pressure from rain or snow melt can cause instabilities (Øydvin et al., 2011). Freeze and thaw cycles are also known to be a cause of rock falls (Dorren, 2003).
When temperature drops, water in the fractures freezes, and the ice expands. Due to the expanding ice, the rock is displaced, and eventually it detaches and falls down.
(a) A rockfall in Kvinnherad, 04.07.16. Photo from Regobs.
(b) A rockfall in Mundheim. Photo by NVE.
Figure 8: Rock falls can come in various sizes, as the images in a) and b) illustrates.
In Hordaland, rock falls are common throughout the county, and the size varies from just small rocks falling from rock faces and road cuts, to larger boulders detaching from rock slopes, as illustrated in figure 8. From 2000 to 2015, more than 4000 rock falls or rock slides were reported in Hordaland (NVE, 2016a), displayed in figure 9. This is a significant number of events, amounting to an average of more than 250 events per year. These data are the raw data from the National landslide database (skrednett.no). No correction has been done to these event reportings, and the database can thus contain double recordings or misregis- trations, leading to a higher number of events than what is actually the reality. However, it is emphasised that the number, with or without corrections, is high.
Figure 9: Overview of spatial distribution of all rock falls and rock slides (types 110, 111, 112, 113 in the national mass movement database) in Hordaland between 2000 and 2015. The landslide database contains 4025 rock falls and rock slides.
2.4 Slope deformation and rock avalanches
Slope deformation is a general term concerning the large scale deformations of mountains (Devoli et al., 2011). The rate of the deformation is slow, often only mm/year, and the movement can be recognised by scarps, benches, cracks, trenches or bulges, however not fully developed failures. Rock avalanches occur when the slope deformation leads to failure of the slope. Such landslides involve massive volumes, and can move e.g. by falling or flowing, or in a combination of these. Rock avalanches have large runout zones, and can cover e.g. entire valley bottoms. The mobility and range of the rock avalanches are what distinguishes them from rock slides (Devoli et al., 2011). The Norwegian Geological Survey (NGU) has since 2005 systematically mapped areas where slope deformation that can lead to rock avalanchs in the future occurs. Figure 10 show an overview of the areas in Hordaland where slope deformation has been mapped.
Figure 10: A screenshot from the online map service from NGU, showing the areas in Hordaland where slope deformation has been mapped.
2.5 Landslides in soil
Landslides in soil cover a range of different subtypes, but only some of them are common in Hordaland. Debris slides, debris flows and debris avalanches are the three subtypes that most often occur. “Debris” is the common denominator for all three types, indicating that the material in the landslides can have different size, shape and composition. Debris means that the material is a mixture of different components, such as sand, gravel, boulders and finer sediments (Hungr et al., 2014), and even organic material and vegetation from the top soil.
Debris slides are planar slides, usually shallow or small sized. Debris avalanches are larger, and can often start as a debris slide(Hungr et al., 2014). Debris avalanches are fast-moving shallow flows where the soil water content can vary, usually unsaturated or not fully satu- rated. These landslides are not confined to channels, and can happen outside of rivers and streams (Highland, Bobrowsky, et al., 2008).
Debris flows, on the other hand, are confined to channels, and are thus a hazard for a specific path or deposition area. Debris flows often occur during floods, and acts as a surging flow of saturated debris (Hungr et al., 2014). The channels does often not have permanently running water (Øydvin et al., 2011), and the material in the flows can come from moraine material, weathered material, top soil (organic material), or from previous debris flows. Sediments from the flows are deposited along the edges of the landslide path as levees, and in a fan at the base of the slope. Debris flows can extend over large distances, often covering several kilometres, due to high water content.
Figure 11: Overview of spatial distribution of all types of landslides in soil in Hordaland between 2000 and 2015
Figure 11 show an overview of the landslides that occurred in Hordaland between 2000 and 2015. According to Furseth (2006), at least 11 people have died from landslides in Hordaland through historic times. Three people were killed in 2005, during the extreme weatherKristin.
A debris slide struck the residential area of Hatlestad Terrasse in Bergen (figure 12).
Figure 12: The debris slide that struck Hatlestad Terrasse in September 2005. Photo by Hallgeir V˚agenes/VG/Scanpix (Aftenposten, 2011).
Similarly to the rock falls, most of the landslides reported in the landslide database are occur along the road and railway network in Hordaland, and the landslides cause considerable damage and problems. In May 2015, several debris flows in Bogelia caused the main road and railway between Bergen and Voss to close over several days. The scale of the debris flows is illustrated in figure 13.
Figure 13: Debris flows in Bogelia, Vaksdal, in May 2015. Photo by Rune Sævig (Bergens Tidende, 2015).
2.6 Triggering factors for landslides in soil
Landslides are all results of different climatic conditions. According to Wieczorek and Glade (2005), two groups of climatic factors can trigger landslides. The primary climatic factors are a direct trigger of the slides, such as intense rainfall or rapid snow melt. Other primary trigger mechanisms, such as volcanic eruptions or earthquakes, will not be reviewed in this thesis, as they are not considered to be of any importance for triggering landslides in Horda- land.
Intense rainfall, or rainstorms, are widely known for triggering landslides (Guzzetti, Peruc- cacci, Rossi, & Stark, 2008). The water from the precipitation events percolates into the soil, and increases the pore water pressure. This leads to loss of effective strength, and the soil collapses in a landslide. How much water infiltrates the soil depends of soil properties, such as permeability, porosity, and thickness of the soil. The pore water pressure builds up when the soil is no longer able to transport or drain the water, and this can lead to landslides.
Intense rainstorms, where a large amount of water is added to the soil at once, cause a rapid build up of the pore water pressure.
As snow melts, the water flows steadily into the soil. The soil might, however, not be able to drain properly if the snow melts rapidly, and the water infiltration increases (Wieczorek
& Glade, 2005). This, similar to the intense rainfall, leads to an increase in the pore water pressure, and can eventually lead to a slope failure.
The secondary climatic factors include antecedent rainfall. Antecedent rainfall is the term concerning the precipitation that happened before the primary climatic incident that trig- gered the landslide. The antecedent rainfall supplies water to the soil, facilitating conditions that will allow landslides if primary factors sets in. The soil moisture already present can have an essential role in whether a slope failure occurs or not. A study by Wieczorek (1987), cited in Wieczorek and Glade (2005), showed that no storm of any intensity resulted in slope failures unless the antecedent rainfall previous in the season had exceeded 280 mm, indicating that the antecedent rainfall plays an important role. There is an overall agreement on the importance of the antecedent rainfall, but there are, however, different opinions on the time period significant for the build up of the antecedent soil moisture (Wieczorek & Glade, 2005).
Different studies display varying measures of antecedent periods, ranging from a few days to several months (Guzzetti et al., 2007).
In a master thesis, Jørandli (2016) found that most landslides in the western parts of Norway
occur between October through March. For some years (2000, 2005, 2011 and 2013), the landslide occurrence could be attributed to extreme weather. One conclusion from the thesis was that seasonal variations, where most precipitation in the period between October and March, comes with the westerly winds, and falls over the west coast of Norway, and that these conditions trigger landslides (Jørandli, 2016).
3 Literature review
A literature review was performed in order to get an overview of the information available on the field of the landslide thresholds. The literature is extensive, and only a selection of the available literature is presented in this chapter.
3.1 Landslide thresholds in the world
Many studies around the world has been conducted on landslide thresholds. The work of Caine (1980) is one of the most famous. The study developed global threshold values based on a number of landslide events, using data from 73 published observations of debris flow events and the corresponding rainfall intesity and duration. The study resulted in a general threshold for shallow landslides I = 14.82D−0,39 3. According to Caine, the results give a simple definition of the rainfall necessary to cause instability. Caine does, however, not take into account the antecedent rainfall prior to landslides. Later studies has stressed the impor- tance of antecedent rainfall.
Some studies attempts to develop global landslide thresholds (Caine, 1980; Innes, 1983), while other take on the studies of landslide thresholds of specific regions or countries (Jakob
& Weatherly, 2003; Glade, 1998; Jakob, Owen, & Simpson, 2012). Guzzetti et al. (2007) re- viewed landslide thresholds for the Central European Adriatic Danubian South-Eastern Space (CADSES) area, and developed new, lower limit rainfall thresholds that can cause (shallow) landslides when exceeded. The study established that the minimum intensity likely to trigger slope failures decreases with increasing duration. Based on this, Guzzetti et al. (2007) de- termined minimum intensity-duration and normalized intensity-duration thresholds for the area. The results indicated that lower average rainfall intensity is required to cause landslides in areas with mountain climate.
Jakob (2003; 2012), conducted studies of debris flow thresholds in connection with a regional debris flow warning system in North Vancouver, Canada. The 2003 study analysed hydro- climatic variables from 25 storms that caused debris flows, and comparably 25 storms that failed to initiate debris flows. The study identified important factors for landslide prediction, and also found that antecedent conditions, such as snow melt and antecedent rainfall, are im- portant for developing a threshold. The 2012 study reviewed a real-time debris flow warning system, where a system based on discriminant function analysis of a variety of hydrometric
3I= Intensity (mmh−1), D=duration (h)
variables was developed. The analysis identified three important variables; the 4-week an- tecedent rainfall, the 2-day antecedent rainfall and the 48-hour rainfall intensity during the storm. The system is connected to a rain gauge, which provide real-time rainfall data, and the antecedent rainfall is automatically calculated every hour.
The goal of the study conducted by Guzzetti et al. (2008) was to update the work of Caine.
They established a new, minimum rainfall intensity-duration threshold for possible initiation of shallow landslides by using precipitation data from a literature search. The study found that antecedent rainfall and soil moisture was important for landslide initiation when rainfall events exceeds 48 hours, and a worldwide landslide warning system was proposed based on the results from the study.
Crozier (1999) used the Antecedent Water Status Model in a New Zealand study aiming at developing a methodology for the prediction of landslides occurring in Wellington within a 24 hour window. The model is based on the assumption that a critical water content in the soil is required to initiate a failure. In the study, Crozier found that the model was able to provide a level of prediction, but that there were many factors that limited the results. This study differs from the other studies mentioned above, as it is more concerned with the status of soil moisture content than the intensity and duration of precipitation. He is accompanied by Waltham (2009), who states that ground water is the single most important factor in triggering landslides (Waltham, 2009, p. 72).
3.2 Landslide thresholds in Norway
In 1954, a large storm struck the northern parts of Norway, in the counties Nordland and Troms. Jørstad (1964) investigated for the first time the meteorological and hydrological conditions that led to several landslides. He argued that there is a certain percentage of the yearly normal of precipitation that should indicate how much precipitation is needed before landslides are initiated. Although his study area was in the northern parts of Norway, he presented percentages for other parts of Norway as well. For the western parts of Norway this percentage is 5-8 % of the yearly normal precipitation. In addition, a general expression for a critical amount of precipitation for landslide initiation for the whole of Norway was pre- sented: RDcrit = 40 + 0,058 ¯Ryears. Calculations show that the critical value in the western parts of Norway is 110 mm precipitation.
about debris flows. The work was based on work conducted by NGI, and a literature review.
Sandersen (1988) investigated the connection between rainfall and debris flow initiation, and found that different climatic areas have different critical values for precipitation intensity.
The antecedent rainfall also play a role, influencing the critical precipitation intensity (it lowers). For coastal areas with marine climate (typical western parts of Norway), it was found that a critical period in order for unstable conditions to occur, is at least 6 hours, and corresponds to rainfall of 40-60 mm. This causes an intensity of 6,7-10 mmh−1 (Sandersen, 1988, p.29).
In a more recent report, Sandersen, Bakkehøi, Hestnes, and Lied (1996) investigated the weather conditions prior to 30 debris flows, 27 rock falls and 4 rock slides, including rain- fall, air temperature, wind speed and air humidity. The study found that the the governing process for slide initiation was meltwater production in the areas with continental climate (eastern parts of Norway), while meltwater in addition to precipitation was important in the areas with a marine climate (western coast of Norway). In the western parts of Norway, slides are most frequent from August to December, which are months with heavy precipita- tion coinciding with heavy snow melt (Sandersen et al., 1996). The study confirmed what was previously noted by Jørstad (1964), that slopes in areas with higher annual rainfall have ad- justed to higher precipitation intensities. This was also noted later, by Guzzetti et al. (2007).
Sandersen et al. (1996) also specified an expression for the critical water supply needed for slide initiation by using the percentage of mean annual precipitation: Pcrit = 1,2xD0,6 4. Meyer, Dyrrdal, Frauenfelder, Etzelmuller, and Nadim (2012) conducted a study on debris flows triggered by extreme precipitation events and rapid snow melt in Norway. All to- gether, 502 slides across Norway and their associated hydrometeorological conditions were documented and investigated, and two different thresholds were calculated:
1. a thresholds based on absolute ID relationships
2. a threshold based on ID relationships normalized by the local precipitation day normal They found that, depending on duration, the absolute threshold intensities varies between 15-107 mm/day for debris flow initiation, while the normalized thresholds varies from 6 to 250 mm/day. The study agree with Sandersen et al. (1996), and states that intense rainfall and rapid snow melt are the most common triggers of landslide initiation in Norway. Op- posed to the other two studies mentioned, Meyer et al. (2012) do not use data directly from
4P= critical water supply as a percentage of mean annual precipitation, D=duration (h)
meteorological stations. In the study, gridded data from senorge.no was used. The whole of Norway is divided into grids of 1 x 1 km, where daily observations of temperature and precipitation is extrapolated from hydrometeorological data available from nearby hydrome- teorological stations (Engeset, 2016).
Over a period of four years, from 2009 to 2013, three NVE projects investigated threshold values for landslides based on statistical analyses of the connection between historical land- slides hydrometeorological variables simulated by the HBV model (Colleuille, Haugen, &
Beldring, 2010; Cepeda, Sandersen, Ehlers, Bell, & Luca, 2012; Cepeda, 2013b, 2013a). The results were later summarised in S. Boje, Colleuille, and Devoli (2014). It was found that the best combination of variables, that predicts landslides, were the relative water supply and the relative soil saturation. The relative water supply is a percentage of the annual mean precipitation for the period 1971-2000, while the relative soil saturation is the sum of water in the soil- and ground water parts of the HBV model normalised by the maximum simulated value for the period 1990-2008. The study by S. Boje et al. (2014) show that landslides occur when the relative soil saturation is above 60 % and the relative water supply is lower than 8
% 14.
The thresholds from the NVE are visualised in form of a 1x1 km2 grid map called “hydmet index” at xgeo.no. The warning levels from the EWS service are issued based on an index map and expert knowledge. There are four levels of warning; green level (1), yellow level (2), orange level (3) or red level (4). The landslide hazard is evaluated for the current day plus the consecutive 3 days.
3.3 Rainfall events and stream flow events
Most studies of rainfall events are done in connection with landslide thresholds. Not many studies are devoted to the investigation of the rainfall events themselves, or to discovering the characteristics of rainfall events that lead to landslides. One of the difficulties when studying the antecedent rainfall, is to define the starting time of the event.
When investigating the weather conditions, Sandersen et al. (1996) found three major weather patterns responsible for triggering debris flows, where two could be atrributed to the west- coast climate of western Norway:
• Heavy rainfall of typically 1 day duration with a concentration of rainfall within 1 to 4 hours during the storm.
Figure 14: The thresholds used in the EWS in Norway. The colors represent the levels, green (1) to red (4), with increasing degree of landslide danger.
• Rainfall in combination with snow melt of typically 3-7 days duration.
(Sandersen et al., 1996, p.99).
In the Safeland-report (Cepeda, Colonnelli, et al., 2012) used statistical and empirical models for predicting critical meteorological elements. In general, the study found that debris flows generally are initiated for rainfall less than 12 hours, while antecedent moisture conditions plays an important role for soil slides, and these conditions need to be accounted for in thresh- old models. A rise in antecedent precipitation is expected to increase the pore pressures, and thus landslide susceptibility, conversely reducing threshold levels (Cepeda, Colonnelli, et al., 2012, p. 22).
For one of the datasets in the Safeland-report (Cepeda, Colonnelli, et al., 2012), from Barcelonnette in France, an analysis of the antecedent precipitation was conducted. The object was to analyse the relationship between two types of mass movement (debris flow and shallow soil slides). Several periods, ranging from 180 to 2 days before the landslide event, were investigated, and 4 main types of rainfall (type A-D), illustrated in figure 15, were
identified. Each of the rainfall types were then tied to a type of slide, where debris flows were most typical for type A and B, and soil slides were most typical for type C. The study also found that debris flows initiated after a 7-day dry period, followed by heavy rainfall, either registered by all the rain gauges in the area, or just by one rain gauge. It was found that soil slides needed at least 40 mm rain over a 7-day period to trigger (Cepeda, Colonnelli, et al., 2012).
Figure 15: The different rainfall patterns discovered in the study of the Barcelonnette, France, area. From Cepeda, Colonnelli, et al. (2012)
An area on the west coast of Norway, Norangselva, was also part of the study. It used a model of antecedent rainfall with values ranging between 1 and 360 days, and found that the
It is noted that hourly data was not available for the analysis, and that such data could significantly improve the thresholds.
The connection between stream flow data and mass movement has previously not been ex- plored in any extent. In a report from NVE considering floods and landslides in Trøndelag in 2012, it is noted that there is a high correlation between landslide events, daily precipi- tation and water discharge (Væringstad & Devoli, 2012). Similarly it is noted a connection between the time of landslides and hydrometeorological conditions in an equivalent report from 2013, about landslides and flood in Troms in 2012 (Devoli, Engen, Hoseth, & Petterson, 2013). However, both reports separates landslides and flood (stream flow data) in the further descriptions of the current events.
In her master thesis, Kr˚abøl (2016) simulated hydrological conditions for rainfall induced landslide events by using the Distance Distribution Dynamics model (DDD). The object was to determine if the model has any capacity to predict hydrological conditions triggering landslides is represented in the model - and how this can relate to landslide occurrence.
The study found that the model has the capacity to predict hydrological conditions, and especially overland flow proved to be relevant for landslide occurrence. 90 % of the landslides were characterised by a sharp gradient in saturation and/or prolonged saturation. However, she also found that soil can be completely saturated without triggering landslides, and that the occurrence of landslides seems to linked to additional input of rainfall.
4 Study area
Hordaland is a county on the west coast of Norway. It covers 15422 km2 (Hordaland Fylkeskommune, 2016), and has 32 municipalities. Bergen is the largest city in Hordaland, where around 54 % of the population live. The coast is characterized by sounds and straits opening into the North Sea, while deep fjords carve their way into the interior of the county.
Figure 16: Hordaland, situated on the west coast of Norway
The highest parts in Hordaland are found on the mountain plateau Hardangervidda, at the glacier Hardangerjøkulen (1863 m.a.s.l.), in the eastern parts of Hordaland. Hordaland is home to two ice caps (plateu glaciers), Folgefonna and Hardangerjøkulen, each with several minor outlet glaciers. Hordaland has important infrastructure connecting the county to the rest of Norway. Three major highways run through Hordaland, and E16 is the most important road between Bergen and Oslo. Several minor highways connects the different parts of Hordaland. The railroad “Bergensbanen” is an important source of transportation, both for transportation of people and goods. Because of the topography in Hordaland, most of the large transportation routes are set along the steep sides of the fjords, many of which are prone to landslides, as previously illustrated by figures 9 and 11.
4.1 Climate
The western parts of Norway are known as the wetter regions of Norway. The climate is characterized by a temperate climate with high precipitation rates (Dannevig, 2009). The amount of precipitation varies throughout the county due to the mountainous areas inland, that shields the inner parts from the precipitation. This causes large variations in precip- itation, such as the difference between Masfjorden (3537 mm per year) and Eidfjord (840 mm per year) (Uni Research Klima, 2015). Figure 17 show the monthly normal values for Bergen Florida, displaying the seasonal variations in both temperature and precipitation, while the normal precipitation for Hordaland compared to the rest of Norway, in addition to the normal temperaturesis, is illustrated by figure 18. The temperature is around 0°C during the winter along the coast, but considerably lower in the more mountainous regions (Norsk Klimaservicesenter, 2016). Figure 19 show the distribution of snow in Hordaland, measured in water equivalent together with an overview of the snow depth in southern Norway. As the figure show, the inner parts of Hordaland, the most precipitation rich areas, usually receive the largest amounts of snow.
Figure 17: The normal precipitation and temperature curve for Bergen - Florida. Figure is produced from www.eklima.no
(a) Normal annual precipitation measured in mm.
(b) Normal annual temperature measured in Celsius.
Figure 18: Precipitation and temperature normals for the south of Norway 1971-2000. Data is collected from www.senorge.no.
The report “Klimaprofil Hordaland”, published by Norsk Klimaservicesenter (2016), predicts the expected climate for Hordaland in year 2100. According to the report, a temperature
(a)Normal annual amount of snow in mm water
equivalents (b) Normal annual snow depth.
Figure 19: Normal annual amount of snow and snow depth for the south of Norway 1971-2000. Data collected from www.senorge.no.
increase of 4°celsius, and an increase in the yearly precipitation of 15% is expected. This will, assuming the projected estimates holds, lead to increase in both flood and landslide activity in Hordaland due to the increased intensity of precipitation.
Gottschalk, Jensen, Lundquist, Solantie, and Tollan (1979) defined regions with common hydrological regimes in the Nordic countries. The basis for the regions were variations in the runoff through a year, and the characteristics were related to either high or low water. The high water category includes three different divisions:
H1: Dominant snowmelt high water. An area is classified as H1 if the three months with highest average runoff belong to spring or early summer (typically May-July).
H2: Transition to secondary rain high water. An area is classified as H2 when the second highest or third highest monthly runoff takes place in autumn (typically: October, Novem- ber, on the Scandinavian peninsula - early in the west and late in the east and vice versa in Finland).
H3: Dominant rain highwater. An area is classified as H3 when the highest monthly runoff takes place in autumn or early winter (typically November- December).
The low water category includes a corresponding division:
L1: Dominant low water flow in winter, caused by snow accumulation. An area is classified as L1 when two months with the lowest runoff both belong to winter or early spring (typically:
February-March).
L2: Transisiton zone, when the two months with the lowest runoff do not belong to the same
time of the year (typically: February and July).
L3: Dominant summer low water caused by high evapotranspiration and/or low precipitation when the two months with the lowest runoff belong to summer or early autumn (typically:
June-August) (Gottschalk et al., 1979, p.280-281).
Accoring to Gottschalk et al. (1979), the west coast of Norway exhibit a variety of hydro- logic regimes, with H3L3 at the coast, which changes to H2L3 further inland, and ends up as H1L1 for most of the inner parts of Hordaland. Due to the importance of water supply for initiations of landslides, this classification of the hydrologic regimes in Hordaland should indicate that most landslides occur in the autumn or early winter, and in the spring/early summer when the rate of snow melt is high.
4.2 Geology
Most of the bedrock in Hordaland is gneiss and granite, underlying sedimentary rocks such as phyllite and limestone (Bryhni & Thorsnæs, 2014). The geology is strongly influenced by the Caledonian orogeny, such as ”Bergensbuene” around Bergen. During the orogeny large areas were overthrusted by bedrock and the sea floor from the west due to the collision between Laurentia and Baltica (Bryhni & Thorsnæs, 2014).
However, the last ice age has had the largest impact on the landscape in Hordaland. The icecap of the last glaciation covered the land all the way to the continental margin, and the ice carved out large valleys and deep fjords. The last parts of the ice cap melted around 9000 years ago (Mangerud, 1976), leaving the landscape shaped the way it looks today. The glaciers excert enormous forces on the bedrock below, and crushes the underlying layers.
Almost all of the detrital sediments found in Hordaland originated during the last glaciation, and the material covers large areas of the county (Mangerud, 1976). These sediments can be found both in steep slopes where thin moraines are prevalent, or in flat areas where espe- cially crushed phyllite provides good agricultural land (Mangerud, 1976). Moraine material is prone to landsliding, and the last glaciation has thus been a large contributor to why Hordaland is so exposed to landslides today.
5 Data
This chapter describes the data used in the analysis, which are all collected from open sources available online such as eKlima.met.no or xgeo.no. The landslide data has been downloaded from the national mass movement database, through NVE. The following data was used:
1. Landslide data 2. Precipitation data 3. Stream flow data
The criteria for choosing the landslides and hydrometeorological stations used in the analysis will be further elaborated in the following sections.
5.1 Landslide data
The mass movement database includes all types of mass movements, also snow avalanches.
A data set only containing mass movement events in Hordaland was extracted from the main data base. By 14.09.16, the data set for Hordaland contained 6733 mass movement events events.
5.2 Precipitation data
Per 25.01.2017, there are 37 operational stations in Hordaland. Most stations are owned by the Norwegian Meteorological Institute (MET), one is owned by the Norwegian National Rail Administration (now BaneNOR). The 37 stations cover most parts of Hordaland, except the outermost areas along the coast and the remote areas of the Hardangervidda mountain plateau. In addition, there are several other stations owned by other institutions, such as Landbruksmeteorologisk tjeneste (LMT), or even privately owned stations. These additional stations have not been used in this thesis.
The stations are located at elevations between 12 m.a.s.l and 1260 m.a.s.l. 19 of the stations are located at 108 m or higher. The longest operating station started in 1895 (Bulken). 15 of the stations opened between 2000 and 2015. Stations not operational at time of data download has not been included in this thesis, even if it was operational some years of the study period (2000 to 2015). The object of this thesis is to investigate the factors which
can be applied to an operational early warning system. The stations that are currently op- erational are the most important source of information for a working EWS. Stations which has been operational at a previous time can be good sources for additional information and to strengthen the dataset, but an inclusion of these stations would make this study far to extensive.
The data downloaded contain raw data from MET’s e-portal eKlima.no. Hourly data is available for some of the stations, however, the data from eKlima is stated as one daily value regardless if hourly data for a station is available or not. Hourly data is available from xgeo.no, and has been added to the daily data where possible. The precipitation is measured in millimeter (mm).
All the stations used in this thesis have long data series of daily precipitation values, but not all of them cover the entire range of years used in this analysis. There are also gaps where data is missing in the data series. These gaps can range from a few days, to months and even years for some stations. All stations have some kind of gap in the data series.
5.3 Stream flow data
The stream flow data was downloaded from NVE. There are, per 02.04.17, 24 operational stations in Hordaland measuring stream flow from which stream flow data was extracted.
All data is owned by NVE. The earliest station opened in 1892 (Bulken), while the newest station, Bjoreio, opened in 2006.
The stations cover a large portion of he central parts of Hordaland. The parts close to the coast, and the interior surrounding Hardangerfjorden are not covered by any stations. Ele- vation of the stations ranges from 19 m.a.s.l. to 1007 m.a.s.l. Half of the stations are located at 223 m or higher elevations. The data downloaded contain hourly data series of stream flow measurements from each station. The hourly data has been corrected for potential ice jams. The unit of the stream flow data is m3s−1.
The catchments used in the analysis were downloaded as a shape file from NVE
(http://nedlasting.nve.no/gis/). The file contains the total catchments with station, meaning all the catchments associated to a type of measuring station. Figure 20 show the distribution of the catchments form the shape file connected to a water discharge station.
Figure 20: A visual overview of the 24 catchments and the corresponding stations within Hordaland. The station ID (e.g. 63.12) corresponds to the list of stations in table 4 (presented in a later chapter).
6 Methods
The following sections will give an outline of the methods used in this thesis. A summary of the methods, connected to the relevant objective, is presented in table 1. Section 6.1 addresses the extraction and control of the landslide database, while section 6.2 explains the background of the algoritm used and how it works in connection with reconstruction of rainfall events. The last section, 6.3 addresses the method used for identification of stream flow events resulting in landslides.
Table 1: A summary of the methods, in reference to the objectives and tools used.
Objective Method Main tool used
Quality control of landslide database
Extraction of data, and manual control of landslide database
ArcGis, Skredreg- istrering, xgeo.no Identification of rain-
fall events
Algorithm R
Identification of stream flow events
Manual evaluation of landslide events and stream flow
Excel
6.1 Quality control of landslide database
All landslides were extracted to ArcGis by NVE’s online downloading tool. The landslides were grouped based on typology, as only landslides in soil (excluding clay slides) were of interest for the further analysis. The landslides were then grouped based on the year they occurred, and then further sorted by the month. This was done in order to easily and visually, discover any initial patterns of occurrence.
The quality control of the database was done to ensure that the data in the final analysis are without any double registrations, that the landslides have the right typology registered, and that they all have the correct time and place registered.
6.2 The algorithm to identify rainfall events
The method used in this thesis to identify rainfall events that trigger landslides, is based on an algorithm for the reconstruction of landslide inducing rainfall events developed by Melillo et al. (2015). The algorithm was made as an attempt to remedy the lack of standards for measuring rainfall events that triggered landslides, as pointed out by Guzzetti et al. (2007).
The algorithm connects precipitation data from single rain gauges to data from a regional or local landslide database in order to identify and reconstruct the rainfall events that have caused landslides. The aim is that ”use of the algorithm shall contribute to reducing the uncertainty in the definition of landslide-triggering rainfall events, to compile large catalogues of rainfall events with landslide and to determining reliable rainfall thresholds for possible landslide occurrence” (Melillo et al., 2015, p. 311). The algorithm itself does not calculate threshold values, but can be used with other methods to obtain such values. A framework for the algorithm is presented in figure 21.
Figure 21: A framework for the algorithm, from Melillo et al. (2015)