Risk assessment for areas prone to ﬂ ooding and subsidence: a case study from Bergen, Western Norway

Bergen city centre is prone to both subsidence and ﬂ ooding. With a predicted increase in precipitation due to climate change, a higher proportion of rainfall becomes surface runoff, which results in increased peak ﬂ ood discharges. In addition, it has been predicted that sea-level rise and increasing storm surges will result in coastal ﬂ ooding. In this study, the dual hazards of ﬂ ooding and subsidence are analysed to exemplify possible risk assessment maps for areas most prone to the combination of both. Risk assessment maps are a support tool to identify areas where mitigation of subsidence and adaptation for surface water management will be most ef ﬁ cient and measures can be implemented. The results show that dual hazard assessment, like that described in this paper, can be a useful tool for decision-makers when prioritizing areas to implement measures such as Sustainable Urban Drainage Systems.


INTRODUCTION
It is expected that 60% of the world's population will be living in urban areas by 2030, and most of this area has yet to be built (UN ). The pace of urban growth may be overwhelming and exert tremendous pressure on the catchment hydro(geo)logy in general and urban drainage in particular (Marsalek et al. ). The built urban infrastructure, with asphalt and concrete-covered ground surfaces, alters hydrologic abstractions and water flow found in natural catchments (Bolund & Hunhammar ). It has been predicted that climate change will increase precipitation (Hanssen-Bauer et al. ), and a higher proportion of rainfall will become surface runoff, which, in turn, will result in increased peak flood discharges and degraded water quality (Haughton & Hunter ). In addition, the sea level is predicted to rise by up to 1 m by 2090 (Hanssen-Bauer et al. ). Changes in the urban environment due to growth in addition to climate change put the urban water cycle out of balance, thereby affecting other surface and subsurface processes, such as flooding and subsidence.
Urban areas are, to a large extent, built environments, and from that view constitute a unique environmental chal-

).
In order to provide communities with urban infrastructures that are durable and well-functioning, climate change impact and adaptability assessments are vital (Pregnolato et al. ). Flood modelling is a useful tool for planning floodways, identifying areas for mitigation measures and for bringing awareness of water issues into decision-making processes in urban areas (Fletcher et al. ; Albano et al. ; Boogaard et al. a, b; Lyu et al. ). Hence, risk assessment mapping can be further used for identifying areas for the implementation of Sustainable Urban Drainage Systems (SuDS), such as swales, to infiltrate water into the ground and to sustainably manage surface water in urban areas (Venvik & Boogaard submitted). More knowledge is needed to understand the urban water balance and the processes connected to water to prevent and counteract subsidence that can cause damage and unforeseen expenses.
Increased knowledge and understanding of the urban water cycle in the transitional zone between the built and natural environment is necessary. In the case of Bergen city centre, past research has shown that the subsidence to a large degree is driven by the depletion of water in the underlying organic-rich cultural layers (Harvold et al. ; Matthiesen et al. ). For a complete understanding of the urban water cycle, hydrological and hydrogeological studies should be included (Wakobe et al. ). Hence, we combine datasets for flood risk and subsidence to develop a risk assessment map for areas prone to damage. The case study is set in Bergen city centre (Figure 1), on the west coast of Norway.
Bergen is a coastal city where the annual precipitation is high, 2,250 mm/year (NMI ). The city is therefore prone to water-related damage caused by pluvial flooding, storm surges and stormwater flooding. the user to visualize, inquire, analyse and interpret the vast amount of (geological) data for a better understanding and problem-solving. Therefore, the risk assessment analysis presented in this paper aims to identify areas prone to the dual hazards of both flooding and subsidence. Dual hazard assessment maps, based on existing flooding and subsidence data, were executed using overlay and 'hot spot' analysis in the GIS. Results can be used as a tool to select areas that need mitigation and damage prevention measures, both for buildings and urban infrastructure. Risk assessment, shown in this case study, may be applied in urban (or rural) areas where data, such as subsidence and flooding, are available.

STUDY AREA AND DATA
Bergen is the second-largest city in Norway, located on the west coast, with an area of 464 km 2 and a population of 278,556 (SSB ). The city has an annual average temperature of 8.6 C and an annual precipitation of 2,250 mm (NMI ). The climate is predicted to become wetter with more intense and frequent downpours, which will increase the pressure on surface water runoff and stormwater management The study area has been constrained to the city centre, including the Medieval city and its surrounding area. In the city centre, the anthropogenic cultural heritage layers are thick with a rich organic content locally more than 10 m thick ( Figure 2). The old shoreline from the 12th century (Hansen ) is shown in Figure 2. Since Bergen has close to no isostatic land uplift (Mangerud ), the progressing shoreline of today is due to filling of the anthropogenic material such as waste into the bay area, Vågen. These layers are more prone to destruction due to lack of infiltration of surface water; therefore, the Bryggen project was initiated in 2010 to save the UNESCO World Heritage site of the Venvik & Boogaard submitted).

Drainage system in Bergen city
To handle the surface water and stormwater, Bergen city has a drainage system with the purpose of transporting water effectively out of the city. In the greater parts of the city, especially in the inner centre, the stormwater is brought together with the wastewater from the industry and household ( Figure 2; Bergen Kommune ). When intense rainfalls occur, the capacity of the drainage system is strained, which may cause the emission of wastewater. Since the relief in the city centre is steep ( Figure 1) and the surface has low permeability, flooding arises when large and intense rainfalls occur in short time spans. Due to climate change, events with downpour will be more intense and frequent. This, in addition to predicted sea-level rise, will give more frequent and intense flooding where there are topographic depressions (Hanssen-Bauer et al. ), as seen in Figure 3.
For this study, we included a dataset of the pipelines for wastewater and sewage. It should be noted that the sewage system may be a combined stormwater and sewage, or a separate system: these are not differentiated in the dataset (Bergen Kommune ).

Flood modelling
Pluvial, urban flooding has received increased attention over the last decade (Mignot et al. ), due to the costly damage

Subsidence data
The subsidence data used in this study were produced by the  For this study, a threshold for the PSI data was set to À1 mm, only negative vertical movement, subsidence, from À1 mm and larger was included. All data with values 0 mm or more, positive (þ) vertical movement was discarded.

METHODOLOGY -RISK ASSESSMENT APPROACH
The Geographical Information System tools such as ArcGIS and ArcGIS Pro (ESRI ) were used for the analysis in this study, with the aim of detecting areas with a risk of both subsidence and flooding. To prepare the datasets for analysis, the results from the flood model were georeferenced and vectorized and clipped against the shore. The original flood model consisted of many small and scattered polygons.
Since the focus was on areas with severe flood problems, flood polygons spaced closer than 3 m were aggregated, while the areas smaller than 10 m 2 were removed. Then, the results from the pluvial flooding were merged with the 200-year storm surge data. Only PSI points with more than 1 mm/year subsidence were used ( Figure 5). The uncertainties connected with these datasets will be discussed later.
The first and simplest overlay is a plain visual overlay of the input data, showing flood data (blue areas in Figure 6(a)) with subsidence data (red points in Figure 6

Description of the simple grid overlay method (1)
In the first method, grids of different sizes are created followed by a selection of grid cells that cover areas with a risk of both flooding (Figure 6(a)) and subsidence (Figure 6(b)). See

RESULTS AND DISCUSSION
The areas identified to be at dual risk in this study could further be targeted for mitigation measures that allow surface water to infiltrate the subsurface. Firstly, such measures would help maintain the anoxic conditions

Datasets and selected methods for analysis
A visual analysis of the input data reveals an image of a city widely affected by subsidence and flooding after heavy rainfall or storm surges, as shown in Figure 7. To make visual analysis easier, the PSI data are shown with points of increasingly darker red for higher degrees of subsidence.
The flooded areas are shown in blue. Areas most prone to flooding and subsidence become prominent in this visualization (Figure 7).

Subsidence data
It should be noted that PSI datapoints may represent points on the ground or points on the city infrastructure, such as buildings. Any hard object on the surface may reflect a signal. As such, there is always the possibility that individual points are measuring the deformation of the city infrastructure or in the building itself, and not ground subsidence.
Additionally, the PSI technique does not return any measurement in vegetated areas, such as yards or parks.
Nonetheless, more than 300,000 datapoints were used in this study providing orders of magnitude more information than could have been obtained using traditional surveying techniques. Although there are many historic buildings in the area, most have been rehabilitated in the last decade and we do not expect that building deformation is a significant part of what is measured. Therefore, we have great confidence that PSI data are suitable for the risk assessment. In this study, all PSI points with more than 1 mm subsidence are included. The exact value of vertical velocity is not used in either of the analyses, only the presence in the simple grid overlay (method 1, Figure 8) and the cluster of points in the 'hot spot analysis' (method 2, Figure 9). For method 2, a visual control of the result shows that areas of high value are also selected as hot spot areas.

Flood data
Results from the urban flood modelling, used in this study, emphasize the areas prone to flooding (Boogaard et al.  The hot spot analysis, method 2, does the narrowest selection of areas, using the aggregated flood data and a count of subsidence hot spots within each 20 × 20 m grid cell ( Figure 9). The results show that within our study area, there are several areas of significance. For a decisionmaking process, it would be easier to prioritize areas for mitigation using the 'hot spot analysis' for risk assessment mapping, as shown in Figure 9.
Risk assessment map combined with the existing drainage system As an example of usability, the risk assessment maps from the simple overlay analysis (method 1, 20 × 20 m grid cells) and the 'hot spot analysis' (method 2) have been combined with the existing drainage system. A 'near-analysis' with 3 m radii of areas in dual hazard and pipelines intersect shows areas where the drainage system is under great pressure when heavy and rapid rainfall or a storm surge occurs ( Figure 10). This is standard procedure within water management (Marsalek & Chocat ; Marsalek et al. ). However, this study shows the connected drainage pipes and manholes that in addition to high water pressure and excess surface water are prone to ground subsidence that may cause damage and disconnect the pipes ( Figure 10). In these areas, it is expected that the drainage system has a greater need for maintenance and thereby costs.  (Figures 8 and 11(b)). When using small grid sizes and without consideration of nearby objects, there is a risk of overlooking relevant areas. There is no prioritizing, and one can argue whether this map result is of any benefit to Bergen's decision-makers other than seeing that there are large areas of dual hazard. It may also contribute to a loss of information due to the cartographic overlay of the input dataset (Figure 11(a)).
Nonetheless, the result does suggest that there is a need for general guidelines for city management and building owners. At this level of detail, and if the target user group was property owners, the method can focus on buildings that are prone to flood and subsidence. A 'near-analysis' would possibly be a better alternative as exemplified with pipelines in Figure 10. The results from the 'hot spot analysis' (method 2) are more selective and areas are clearly prioritized (Figure 11(c)). For scientific research on the relationship between flooding and subsidence, or for the municipality to select areas for greater follow-up, this method gives significant results for the clearest selection of areas (Figures 9 and 11). The subsurface of any city is complex, and in Bergen, it can be roughly divided into three layers: natural ground consisting of bedrock and sediments on top, cultural layers consisting of domestic waste, with up to 100% organic matter (Matthiesen et al. ; Rytter & Schonhowd b) and anthropogenic materials, such as agglomerate, asphalt and material for drainage. The subsidence occurring is not constrained by geological structures and cannot be explained by geological processes alone. However, water, both surface water and groundwater, plays an important part in the process. Pregnolato et al. (), in their risk assessment of roads in Newcastle, UK, show that roads are prone to flooding during heavy rainfall. Similarly, the risk assessment presented in this study can help the municipality prioritize areas for mitigation or that need on-going surveillance. A current discussion in Norway is how to implement climate adaptation into best management practice for municipalities (Hanssen ). Hanssen () shows how well flood risk maps function to translate natural science information into local planning and decision-making. This shows that maps are credible and essential tools, but that they need to be brought to the table by planners and interpreted in a local context. Hanssen () conclude that local climate adaption is dependent on well-functioning interaction between multiple levels as well as disciplines and emphasis on strengthening the role of the government agencies as 'knowledge translators' ('kunnskapsoversettere'; Hanssen ).

Risk assessment as a tool for end-users
The risk assessment map methodology presented in this study aims to translate knowledge into maps to assist the end-user to select areas for implementation of, for example, SuDS by identifying areas prone to the dual hazard of flooding and subsidence. Resilience of the built environments has not been well studied (Thornbush et al. ), and results from this study may help the Bergen Municipality to plan mitigation and further adaptation to prevent areas of flooding, by increasing infiltration of surface water and decreasing flooding, as well as the processes causing subsidence as exemplified in Venvik & Boogaard (submitted).
Managing stormwater is not just important for protecting water resources and aquatic ecology but also to restore urban water cycle processes that are critical to the health of urban watersheds. These include infiltration and groundwater recharge, evapotranspiration and chemical/biological

CONCLUSIONS
There is a link between areas that suffer from subsidence and areas with an excess or shortage of water. The aim of this study was to locate areas in Bergen city centre that are prone to the dual hazard of subsidence and flooding.
This was achieved by processing existing data and maps that identify areas prone to PSI data for risk of subsidence, a flood model map and a storm surge map for areas prone to flooding.
We have demonstrated that a 'hot spot analysis', for the subsidence data within areas prone to flooding, provides an The subsurface in cities is complex due to a mixture of natural and built environments. The processes causing subsidence are not easily understood but are commonly related to water. Increasing infiltration of surface water may prevent the processes causing subsidence. Managing stormwater in this way is not only important for protecting water resources and the aquatic environmentit can help restore and maintain urban water cycle processes critical to making cities resilient to the effects of climate change.

Further work
The increased availability of data, both large datasets and timeseries, makes analyses, such as the risk assessment presented here, much more achievable. The Copernicus program is revolutionary in that it promises this type of data for decades to come, free and open. Risk assessment similar to that conducted in this study is relevant for all cities that are prone to coastal and/or pluvial flooding or possible the combination of flooding and subsidence. The www.InSAR.no service is an open access portal, displaying data used in this study, and is an example of possibilities with the upcoming EU Ground Motion Service.
The latest available PSI data and a new and updated flood model based on the latest and most detailed DEM and topographic data should be used before selecting areas in a potential follow-up of this study. This risk assessment should be also followed up by hydrological and hydrogeological field investigations to evaluate the results and to find the best management practices for the given location and problem.
This study will be expanded to categorize PSI data indicating subsidence by trends in timeseries and combining them with other datasets. This would increase the knowledge of the subsurface processes and the effects of interventions, and thereby ultimately identify effective actions to decrease effects related to the urban water cycle.
Further, end-users should be involved in the development of risk assessment maps, for example in the evaluation of the usability of prototypes, like the ones presented here. Choosing an adequate method for risk assessment with the end-user tasks in focus is important and will give more applicable results. Trying out multiple methods for analysis and visual analysis for quality control of map results was emphasized in this study and is strongly recommended in further studies.