Assessment of pluvial flood exposure and vulnerability of residential areas

Author’s Accepted Manuscript Assessment of pluvial flood exposure and vulnerability of residential areas Tonje Grahn, Lars Nyberg

www.elsevier.com/locate/ijdr

PII: DOI: Reference:

S2212-4209(16)30791-9 http://dx.doi.org/10.1016/j.ijdrr.2017.01.016 IJDRR492

To appear in: International Journal of Disaster Risk Reduction Received date: 16 December 2016 Revised date: 27 January 2017 Accepted date: 29 January 2017 Cite this article as: Tonje Grahn and Lars Nyberg, Assessment of pluvial flood exposure and vulnerability of residential areas, International Journal of Disaster Risk Reduction, http://dx.doi.org/10.1016/j.ijdrr.2017.01.016 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Assessment of pluvial flood exposure and vulnerability of residential areas Tonje Grahn1,2, Lars Nyberg1,2 1

Department of Environmental and Life sciences, Karlstad University Universitetsgatan 2, 658 88 Karlstad, Sweden 2

Centre for Climate and Safety, Karlstad University Universitetsgatan 2, 658 88 Karlstad, Sweden [email protected] [email protected]

Abstract Floods are a large problem around the world but the understanding of flood risks is hampered by a lack of data and knowledge about flood losses at different scales. The objective of this study was two-fold 1) to assess available temporally and spatially distributed data of rain events and flood damages during those events, regarding the usefulness of these data to quantify precipitation-related hazards and consequences, and 2) to assess the potential for deriving reliable damage functions based on the information compiled under objective 1. The study examined 2140 individual observations of insurance payouts for residential buildings caused by 49 different rainfall events in Sweden. Radar data were used to extract daily precipitation amounts and to capture the spatial and temporal distribution of the rainfalls. This study demonstrates that including the duration of a rainfall, as opposed to only the aggregated amount of daily precipitation, is highly important in estimating the extent of damage. Furthermore, higher rainfall intensities increased the number of damaged properties but had less influence on the mean damage cost per property. In order to draw conclusions from damages at the micro level, both availability and detail level of data must be improved.

Keywords: Insured flood losses; flood damage; pluvial flood damage; data availability; flood risk assessment

Introduction Floods are the most common natural hazard in the world [1]. It causes serious losses in terms of lives and damage to buildings and infrastructure [2]. One of the most important indicators of the consequences of such hazards is the level of economic losses which is usually 1

expressed as direct economic damage. Economic losses have dramatically increased over the past decades [3, 4, 5], and with the damage inflicted upon society by flooding it has underlined the need for risk mapping and risk assessments to support risk reduction [6]. In quantitative natural hazard assessments risk is commonly expressed as function of the probability of a hazard, the exposure caused by the hazard and the vulnerability of exposed objects (R=P x E x V). In order to perform reliable quantitative risk assessments we need to understand how various damage-inducing factors contribute to changes in the three risk components so that we can increase our understanding of the interactions between society and floods [7]. So far, most focus of the scientific literature has been on the hazard part (P) of events and not so much on the actual consequences (E x V) caused by hazards [8]. One crucial input to quantitative risk assessments is the potential consequences in residential areas. The most commonly used approach to quantify flood exposure and vulnerability to residential areas is to apply residential flood damage functions (RFDF), also referred to as stage damage functions, vulnerability functions or depth damage functions [3, 9, 10, 11, 12, 13, 14]. RFDFs express objects’ vulnerability to flooding in relative terms (e.g. percentage of total value) or absolute values (e.g. the actual replacement cost of damages), at object level or in aggregated form (e.g. for a land use area) [11, 15, 16, 17]. Further, vulnerability is most commonly expressed as a function of water level inside buildings exposed to large river floods, coastal floods or flash floods [18]. Pluvial floods have been identified as the flood type most likely to increase in severity as a result of climate change [19]. Many European countries have already suffered large consequences due to heavy rainfall. For example, in 2007, England was repeatedly struck by severe flooding. Two-thirds of the damage could later be related to pluvial flooding [20]. Heavy rainfall also caused record floods in Central Europe in 2013 and 2016 with substantial damage to property [4, 21]. Bouwer [5] reports that the Netherlands are expected to suffer an increase in losses due to heavy rainfall of up to 47 % by the year 2040, signalling future challenges. Regardless of future climate scenarios, however, intense rainfall events need to be properly analysed since these events have very large consequences at the present time as well. Sweden has historically experienced substantial consequences caused by floods [22]. For example, the city of Malmö experienced a cloudburst in August 2014 that caused more than 4500 insurance claims and was estimated to cause more than 250 million SEK in payouts counting only insured losses [23]. During 2009-2011 more than 100 intense rainfalls were identified in Sweden that caused damage and disruption in varying degrees [24]. Private flood damage to residential property reported to Swedish insurance companies have been increasing steadily over the past 25 years (Fig. 1). The majority of damages are reported to insurance companies in June, July and August, when intense summer rains are common (Fig.1) Cloudbursts are expected to become more frequent events in Europe with global warming [25], and the intensity of heavy summer rains in Sweden is expected to increase by 10-15 percent [26]. Hazardous events, such as the Malmö 2014 event, are therefore expected to

2

occur more frequently in the future which is a concern to the insurance industry [27] as well as to governmental agencies at both local and national levels [28]. Estimation and modelling of flood damage is highly dependent on data availability. Data sets of past events are useful for modelling damage as they give an idea of possible affected areas, the expected magnitude of events, their frequency and possible impacts on vulnerable elements [29]. Damage data and inundation data are, however, often not available for different parts of watersheds [18]. Empirically based information that links water levels inside buildings to damage is not available in Sweden. The only country that we are certain that this kind of information has been systematically collected is in Germany, and then for flooding of large rivers [11, 13, 30]. Pluvial floods differ from these types of floods by predictability, temporal aspects such as duration, and pathways of excess water [31]. Pluvial Flood impacts on urban areas are especially complex to analyse and because of the complexity of urban contexts standardized RFDFs exclusively for pluvial flood damage have not yet been produced [32]. Data availability and the level of detail of available data are large barriers to deriving empirically based RFDFs. Exploring the potential of data that is available is therefore important for development of alternative approaches of damage assessments to make this applicable to countries that don’t have their own RFDFs and don’t have the comprehensive datasets needed to derive such functions. The most comprehensive loss databases are held by insurance companies which are rarely available to be analysed by anyone outside the companies [33]. In this study the objective was to assess available flood hazard and flood damage data with the purpose of evaluating its applicability to quantitative pluvial flood risk assessments and to measure the causal effect that damage inducing variables have upon flood damages. The paper has the following structure. Section 2 gives a brief overview of flood damage inducing variables emphasized in peer-reviewed literature. Section 3 describes the data assessed and applied in this study. Section 4 analyses the causality of potential damage inducing variables upon damages. Results and data are discusses in section 5, and section 6 concludes the study.

3

a

b

Figure 1: a: Yearly distribution of payouts made by Länsförsäkringar insurance group, 19872013 (left hand side), b: Distribution of payouts per month (right hand side). Number of payouts is represented on the Y-axis. The actual number of payouts cannot be displayed in the figure due to the confidentiality agreement. The figure does, however, reflect the Swedish flood damage trend and the seasonal flood damage pattern.

Damage influencing variables Scientific literature on flood damage risk factors serves as guidance for the data assessment in this study. A search was made in peer reviewed scientific literature in order to identify risk factors that might affect pluvial flood damage and therefore should be quantified in order to estimate future pluvial flood damages. The following indicators of risk were identified: damage cost or number of damages [32, 34, 35], water depth inside buildings [9, 12], volume and intensity of rain [20, 36], spatial distribution of rain [20, 37], sewer system capacity and sewer system failures [20, 36, 38], impermeable surfaces [36, 39], population density [18, 32, 38, 40], building characteristics (residential use, material, number of floors) [9, 12, 35], whether the buildings have been in direct contact with water or contaminated water [41], and self-protective behaviour [42, 43, 44]. Urban city centres are known to be more vulnerable due to their urbanized land cover and the lack of capacity of their drainage systems [36, 38, 39]. Concerning precipitation data, radar data are superior to gauge data in risk assessments since it provides a better representation of spatial variations in rainfall and indications of intensities [45].

4

Data description 1.1.1 The data used in this study is described in the following sections. The different dependent and explanatory variables used in the statistical analysis are described and named in capital letters. Descriptive statistics of all variables are shown in Table 1.

3.1 Insurance data In Sweden, flood damage to residential property is covered by basic home insurance. The insurance covers damage to buildings and moveable property when the flood is caused by rain of an intensity of 1 mm per min. or 50 mm per day, by snowmelt or by increasing water levels in lakes, rivers and streams. Further, the water must have entered a building by flowing from the surface directly into a building through valves, windows or door openings, or through sanitary pipes. There are presently no special demands on policyholders related to risk reduction efforts that can affect the conditions for receiving payout other than the general request for carefulness and to follow guidelines concerning water utility systems. No active choice must be made to include flood insurance in a person’s home insurance, the price of the policy is not connected to the actual flood risk in a specific area, and no policyholder is refused flood insurance in their home insurance [47]. To obtain a mortgage on a house, using the property as collateral, the property must be properly insured by an insurance provider accepted by the creditor [46]. The insurance coverage amongst homeowners in Sweden can therefore be assumed to be close to 100 percent. In this study, market shares are defined as the proportions of homes that are insured by an insurance company relative the total number of homes in the country.

3.1.1 Insurance compensations The damage data used in this study came from the Länsförsäkringar insurance group. These data contain information on the individual insurance amounts paid to policyholders as compensation for flood damage, the date the damage occurred and whether there was damage to buildings (DETACHINSURE) or movable property (MOVEINSURE). It does not contain information that specifies type of flood or how the water entered the building. Due to poor data resolution it cannot be ensured that the payouts are solely related to pluvial flooding. The only payouts that are included in the study are compensation for damage that has occurred when rain has been documented. Known damaging events caused by increased water levels in lakes, rivers and streams are not included in the study, however, it cannot be excluded that the data set also contains damage that has been caused by e.g. overflow of minor rivers and streams due to heavy or ongoing rain. Concerning spatial resolution, the damages were identified at municipal level meaning that the exact localization of the flood-damaged 5

property was not revealed. What can be detected is that flood damage compensations follow a highly skewed distribution (Fig. 2). The reason is that the insurance compensations are characterized by many small payouts. Median values are low, with higher mean values due to a few large payouts (Table 1). The compensation amounts vary between 53 and 970,628 SEK per individual payout (2013 year price level). The amounts do not include the deductible which is set to 10 % of the damage, or a minimum of 10,000 SEK for structural damage and 3,000 for damage to movable property. These deductibles apply to damage caused by flooding1. Other rates apply to other sources of damage. For further statistical regression analysis, the insurance observations were transformed to normality by using their natural logarithm (ln). They can then be used to produce efficient and unbiased estimations of flood damage. Länsförsäkringar insurance group is Sweden’s largest insurance group and covers approximately 35 percent of the national Swedish home insurance market. There are regional differences in this rate and the rates vary between 23 and 65 percent for the municipalities included in this study. Different rates were used to estimate the aggregated amounts (AGGINSURE) paid by the Swedish home insurance industry per rainfall. The same rates were used to estimate the total number of payouts (PAYOUTS) for each event. The relation between aggregated insurance compensation per rainfall event and number of payouts per event are displayed in Fig. 3

1

The rates are not constant and must be checked before applying them to payouts made outside the timeline of this study.

6

1.1.2 Figure 2: Distribution of individual insurance compensations at object level in the range of 53- 400,000 SEK. 23 observations within the range of 401,000-971,000 lies outside the range of the X-axis

a

b

Figure 3: Scatter plot of estimated number of payouts per rain event and estimated total insurance cost to residential property per rain event. a: All observations, b: Observations limited to Number of payouts <300

7

3.2 Population density With information from Statistics Sweden, SCB, population densities (POPDENS) for every municipality and year was calculated as the ratio of the number of inhabitants to the size of the land area (km2) of the municipalities.

3.3 Radar data of precipitation Radar data were used in this study to determine the amount and intensity of precipitation at the date and municipality where flood damage had occurred according to insurance records. Radar data for Sweden, provided by the Swedish metrological and hydrological institute (SMHI), is available from the year 2000 and onwards. The output of the radar data is presented as images of mean precipitation in 3-hour intervals in a rectangular area of 34 x 30 km2. An example of radar images is displayed in Fig. 4, revealing both the amount of precipitation and the temporal perspective. A total of 49 different rainfall events occurring during the months of June, July and August between the years 2000 and 2013 in 13 different municipalities with urban centres were included in the analysis. Geographical maps of impacted municipalities were overlaid with radar images for the relevant dates to determine which parts of the municipalities were exposed to rainfall. Daily precipitation (between 00.00 and 24.00) for the 49 rainfall events varied between 12 and 200 mm. Information on water depths inside buildings are, as far as we know, not being systematically documented by insurance companies or other stakeholders at present time in Sweden. Instead this study uses the aggregated daily maximum amounts of rain (PRECIPDAILY). The information has been extracted from radar maps for the dates and municipalities where flood damage had been compensated by the insurance company. In addition to the rain amount on the day when damages occurred also precipitation for the day prior to damage (PRECIPRIOR) were tested as a damage influencing variable. This was done to see if any effect of ongoing or repeated rainfall could be detected. Apart from the amount of daily precipitation, the intensity of rainfall, combining both amount and duration was tested. Existing well-defined intensity values within the scientific literature are, as far as we know, only related to extreme rain. An example is Karagiannidis et al. [48], who studied extreme precipitation, related to cyclones (excluding summer thunderstorms) and defined 60 mm/day as extreme conditions. For light and moderate rainfall, no applicable definition was found that could be adapted to this study. Therefore, an intensity variable (INTENSE) was derived based on three criteria. The first criterion was that rainfall generating more than 60 mm/day was considered intense. The second criterion was that rainfall generating precipitation of >40 mm in a timespan of < 9 hours was considered intense. The third criterion was that rainfall generating precipitation of =>25 mm in <3 hours was considered Intense. As an example, Fig. 4 displays two different rainfall events both with 8

aggregated precipitation of 40 mm/day. The event on 25 July is deemed not intense while the event on 16 Aug is deemed intense. By visually analysing geographical maps overlaid by radar images, it was also possible to determine whether high volumes of rain exposed the more central urbanized areas of the municipalities during different rain events. Urban city centres are known to be more vulnerable, partly due to their urbanized land cover and the lack of capacity of drainage systems. It has not been possible, however, to quantify complexities such as changes of impermeable surfaces or drainage system capacity for the relevant years and municipalities included in this study (years 2000-2013). Therefore, on the basis of visual analysis, a binary variable representing urban exposure (URBEXPOS) was derived to test the suspected increased vulnerability related to urban complexity. The variable is not able to identify and measure any specific aspect of urban vulnerability but might test whether central rain exposure is a determining factor for urban vulnerability. The urban exposure binary variable was 1 if the centre of the rainfall clearly exposed central parts of the municipality (using the Swedish centre tätort as a delimiter), and 0 if the centre of the rainfall clearly fell over the outskirts of the municipality. Binary variables were also derived to analyse whether the time of day of a rain event affects the size of insurance payments. Due to the often rapid onset of summer rains there is limited time for residents and home owners to minimizing damage, especially if they are not at home. The DAY variable had the value 1 if the event occurred in the time span 09.00-18.00, otherwise the value of 0. The NIGHT variable had the value of 1 if the event occurred in the time span 00.00-06.00, otherwise the value of 0. Note that the DAY variable is not the inverse of the night variable. a

b

9

25 July 2007 Kristianstad, 40 mm/day-

16 Aug 2002 Kristianstad, 40 mm/day –

Not intense

Intense

Figure 4: Example of intense and not intense rain. Daily precipitation in 3-hour intervals (8 images for 00:00-24:00). Each rectangle represents mean precipitation during 3 hours for an area of 34 x 30 km2. Radar images have been used to classify rain events as Intense or Not intense. In event a, 40 mm were distributed throughout a period of 21 hours. In event b, most of the 40 mm of rain fell within a span of 3 hours. Both rainfalls caused damages.

Table 1: Description of the variables used in the regression analysis of pluvial flood damage impact (SEK = 2013 year price level) Variable

N

Min

Max

Median

Mean

St.dev.

Source

MOVINSURE

1126

155

777,748

13,928

28,252

45,361

Insurance

(SEK)

DETACHINSURE

Company

1023

53

970,628

45,708

(SEK)

82,333

106,160

Insurance Company

10

AGGINSURE

49

60326

74,721,480

1,917,790

6,422,425

13,200,000

(SEK)

PAYOUTS

Insurance Company

49

5

1,251

63

115

194

Insurance Company

PRECIPDAY (mm)

49

12

200

50 (80*)

56 (71*)

35 (30*)

SMHI

0*

250*

35*

44*

44*

SMHI

(2,286*)

PRECIPPRIOR

49

(mm)

(2,276*)

INTENSE (binary)

49

0

1

1

0.63

0.49

SMHI

URBEXPOS

49

0

1

0 (1*)

0.45

0.50

SMHI,

(binary)

(2,287*)

(0.78*)

(0.41*)

maps

POPDENS

2,286*

16*

1,930*

106*

402*

570*

SCB

DAY (binary)

49

0

1

0

0.3

0.46

SMHI

0

1

0

0.41

0.49

SMHI

(2,260*)

NIGHT (binary)

49 (2,040*)

MOVINSURE = insurance compensation, moveable property; DETACHINSURE = insurance compensation, detached houses; AGGINSURE = estimated aggregated insurance compensation per rain event; PRECIPDAY = precipitation amount on the day of damage; PRECIPPRIOR = daily precipitation amount on the day prior to damage; INTENSE = intense or not intense rainfall; URBEXPOS = centre of rainfall within or outside urban area; POPDENS = population density; PAYOUTS = estimated total number of payouts per rain event; DAY = rainfall occurring between 09.00-18.00;, NIGHT= rainfall occurring between 00.00-06.00. *The numbers relates to analysis of individual insurance claims when these numbers diverts from the numbers in the analysis of the aggregated insurance data.

11

Statistical analyses of residential flood damages Regression analyses were used to analyse the relationships between insurance data and potential explanatory variables. The analysis was conditional upon flood damage occurring and payments being made from the insurance company to policyholders. All monetary values were adjusted to the price level of 2013 using consumer price index. The sizes of insurance payouts at object level and aggregated at municipality level were used as dependent variables, while meteorological, geographical and demographic characteristics, along with temporal aspects, were tested as explanatory variables (also called independent variables or predictors). Binary explanatory variables were used when categorization of data was necessary. Five different hypotheses have been tested with regard to magnitude of flood damage. The different hypotheses are represented by 5 regression functions (Table 2). Models 1, 2 and 3 represent hypotheses concerning insurance claims at object level. The purpose of the models was to test if and to what extent damage influencing variables affected the size of individual insurance compensations at object level. Model 1 represents damage to moveable household property. Insurance compensation for moveable property had lower median and mean values than payments for damage to residential detached buildings (Table 1). Models 2 and 3 represent damage to residential detached buildings. The main difference between the two models is that model 2 used AGGDAILY as the precipitation variable and model 3 used the INTENSE variable. More explanation to the choice and combination of explanatory variables is found in the results section below. Models A and B both tested if changes in rain characteristics affected the total aggregated amount the insurance industry paied in total compensations to households per rainfall event. Model A and B differ in one specific aspect. While model A used AGGDAILY as the precipitation variable, which reflect only the amount of rain, model B used the INTENSE variable, which adds the temporal aspect of duration to the rainfall events. Both models also tested if and to what extent the total compensations per rainfall were affected by the spatial distribution of the rainfall, population density and number of payouts per rainfall. Regression coefficients displayed in the tables 3 and 4 have been recalculated and expressed as percentages to better communicate the effect of the damage inducing variables (explanatory variables).

Table 2: Overview of hypothesis tested in section 5

12

Model 1

ln(MOVINSURE)=β0 + β1INTENSE + β2PRECIPPRIOR + β3DAY

Section 4.1

Model 2

ln(DETACHINSURE)=β0 + β1INTENSE + β2PRECIPPRIOR + β3DAY + β4POPDENS

Section 4.1

Model 3

ln(DETACHINSURE)=β0 + β1PRECIPDAY + β2PRECIPPRIOR + β3NIGHT+ β4POPDENS +

Section 4.1

β6URBANEXPOS

Model A

ln(AGGINSURE)=β0 + β1PRECIPDAY + β2URBEXPOS + β3POPDENS + β4PAYOUTS

Section 4.2

Model B

ln(AGGINSURE)=β0 + β1INTENSE + β2URBEXPOS + β3POPDENS + β4PAYOUTS

Section 4.2

4.1 Residential damage at object level Effects on size of damage costs for moveable property (model 1) were estimated using the following variables: Intense rain events, aggregated daily precipitation the day prior to damages, and rainfall occurring during the day. These three explanatory variables were all statistically significant. Intense rain caused the mean value of insurance compensation to be 42 percent higher than payouts caused by less-intense rainfall events. For rainfall occurring during the day, when fewer residents are expected to be at home, mean insurance payouts 13

were 42 percent lower than for rain occurring between 18:00-09:00. This contradicts the hypothesis made in section 3, that residents can reduce damage when they are home (by moving inventories away from flooded parts of the house). It must be noted, however, that no certain knowledge about the actual whereabouts of the residents was included in the analysis. Further, Table 1 shows that the distribution of the DAY variable was skewed, with only 30 percent of the rainfall events in the sample occurring in the 09:00-18:00 timespan. The occurrence of rain the day prior to damage observations decreased the insurance payout by 0.03 percent for every mm of rain the previous day. The R2 value of model A was very low (0.03), explaining only 3 percent of the total variation in damage costs. Models 2 and 3 concern individual insurance payouts at object (residential building) level caused by flood damage to detached houses. Model 2 estimated effects on mean damage cost using the explanatory variables of intense rain events, aggregated precipitation from the previous day and population density. Intense rain events had a significant effect, increasing the mean size of payouts by 92 percent compared to those resulting from less-intense rains. The occurrence of rain the day prior to the damage event decreased insurance compensations by 0.2 percent for every mm of rain the previous day. Insurance claims from rainfall events occurring during the day (9:00-18:00) were 30 percent lower than those occurring during evenings or at night (18:00-09:00). Population density did not have significant effect on the mean size of individual damage payouts. The R2 value of the model was low (0.03). Model 3 added one more variable to increase the explanatory power of the damage function: precipitation exposing central urban parts of the municipality. The URBEXPOS variable was, in this sample, correlated to the INTENSE variable (corr. coeff. = 0.56). This was expected since both variables contain some common information. This correlation can affect the size of estimated effect of the individual variables. Since we are interested in individual effect as well as the explanatory power of the model, we decided to not estimate the variables within the same function. Therefore, aggregated daily precipitation (mm) was used to represent precipitation characteristics in model 3 instead of INTENSE. Additionally, the day variable (9.00-18.00) was correlated to aggregated daily precipitation, so the night variable (24.0006.00) was used instead. Rain exposing residential houses in central urban areas of the municipality caused a 69 percent increase in mean insurance compensations for individual damages compared to rainfall exposing residential houses in less central and urban parts of the municipality. Aggregated daily precipitation did not have a significant effect on the size of payouts for building damage when all observations were included in the analysis. However, the rainfall event with a precipitation of 200 mm/day can be interpreted as an outlier due to the very large amount of rain at this event compared to the other events in the sample but also since the compensations caused by this event had a low mean value compared to the overall mean insurance compensations. Model 3 was run a second time without the outlier resulting in statistically significant estimates of daily precipitation. Because there was one single event

14

(29 individual damage observations for 1 rain event), the effect of including/excluding these observations cannot be concluded. It is possible, however, that the most extreme events follow a different distribution than small, medium and large rainfall events. The inability of the PRECIPDAILY variable to consistently explain variation of flood damages coincides with the result of the other analysis in this study. The occurrence of rain the day prior to a damage event decreased mean insurance payouts by 0.2 percent for every mm of rain the previous day. Properties exposed to rain during the night received 43 percent higher insurance compensations than those exposed to rain occurring between the hours of 06.00-24.00. The R2 value in model 3 (0.05) was somewhat higher than that in model 2 (0.03) but is still very low, meaning that the model only explains a very small part of the total variation in mean damage cost. Residuals for all the performed regression analyses were normally distributed and homoscedastic and support the assertion that the OLS analysis was unbiased and effective.

Table 4: Regression result. The dependent variable is the natural logarithm of the size of insurance compensation for damage at object level. Standard errors are in brackets. Explanatory variables

Intercept

PRECIPDAILY, mm

Model 1

Model 2

Model 3

Ln MOVINSURE

Ln DETACHINSURE

Ln DETACHINSURE

9.413***

10.041***

10.136

(0.122)

(0.148)

_

_

-0.0004 (0.0016)

INTENSE (binary)

PRECIPPRIOR

DAY (binary)

0.35**

0.65***

(0.121)

(0.168)

-0.003**

-0.002*

-0.002*

(0.001)

(0.001)

(0.0011)

-0.349***

-0.273*

-

(0.077)

(0.103)

15

_

NIGHT (binary)

_

_

0.5234*** (0.103)

URBEXPOS(binary)

_

_

0.3568** (0.1216)

POPDENS

_

-0.00006

-0.0001

(0.0001)

(0.0001)

N

1116

1124

1036

R2

0.03

0.03

0.05

*0.05 **0.01 ***0.001

4.2 Residential damage aggregated at municipality level This section analyses effects of explanatory variables upon the total aggregated compensation paid by insurance companies to residents for flood damages using two different models, models A and B (Table 2). Model A uses aggregated maximum daily amounts of rain, urban exposure, population density and the number of payouts made by the insurance industry to model the aggregated cost of insurance damage per rainfall. Aggregated daily precipitation did not have a significant effect upon the total sum of insurance compensations. Neither did the categorical variable, urban exposure, which represents rainfall occurring in the central parts of a municipality, or the population density variable. The only significant variable in the model was the variable that in advance where expected to result in a positive significant effect: number of payouts per rain event. Model B analysed the aggregated cost of insurance damage per rainfall event using the variables Intensity, Urban Exposure, Population density and Number of payouts. Intense rains which added a time element to the rain amounts had a statistically significant effect on the total sum of insurance compensations per rainfall. Rainfall events that were categorized as intense in this study caused the total sum of insured property losses to be higher than for less

16

intense rain events. The mean aggregated insurance compensation was 1922 (coeff. 1.072) percent higher for intense rain. Number of payouts had a significant effect on the size of aggregated insurance payouts, meaning that the total amount compensated by the insurance industry increases when the number of individual insurance payouts increases. Figure 5 displays the relationship between aggregated insurance compensations at municipal level and the number of payouts at municipal level per rainfall in relation to intense and non-intense rainfall events. The R2 value for model B was 0.57, meaning that 57 percent of the total variation in aggregated insurance compensation in the sample could be explained by a model containing information on the expected number of payouts and the expected share of intense rainfalls3.

Table 3: Regression result. The dependent variable is the natural logarithm of the estimated aggregated insurance compensations at municipality level. Standard errors are in brackets. Explanatory variables

Intercept

PRECIPDAILY, mm

Model A

Model B

Ln AGGINSURE

Ln AGGINSURE

13.604***

13.261***

(0.349)

(0.280)

0.005

_

(0.005)

INTENSE (binary)

_

1.072*** (0.323)

URBEXPOS(binary)

0.493

0.372

(0.350)

(0.319)

2

The percent change from the estimated coefficient is (e (1.072) -1)* 100 =192 %

3

Leaving number of payouts out of the model gives and R2 value of 0.35.

17

0.0000

0.0001

(0.0004)

(0.0002)

0.005***

0.004***

(0.001)

(0.001)

N

49

49

R2

0.48

0.57

POPDENS

PAYOUTS

*0.05 **0.01 ***0.001

18

1.1.2.1 Figure 5: Aggregated flood damage at municipal level. Actual and estimated relationship between LN (aggregated insurance compensations) and number of payouts (refunds) per municipality, with 95% confidence interval. Lefts side picture represents normal rain, middle-picture represents intense rain, and right side picture represents all rain.

Discussion Data availability and data homogeneity are large issues when analysing flood damages in Sweden. In an ideal study all the factors identified in section 2 would be tested. However, due to data scarcity and inhomogeneity among data sets concerning spatial and temporal overlaps this was not possible. The absence of large, systematically collected datasets hinders the use of quantitative flood risk assessment for policy purposes. By assessing and analysing insurance data, radar data, geographical maps and data available at Statistics Sweden (SCB), this study has explored the potential for detecting pluvial flood damage inducing variables and estimating their effect upon insured residential damages in Sweden. The basis for empirically derived RFDF’s are usually derived from water levels and flood damages occurring in large catchments caused by river floods, and are therefore suspected of being inappropriate for estimating Swedish pluvial flood damages. The existing literature on flood damages in developed countries includes studies conducted in countries as Germany, England and the Netherlands, e.g. Merz et al. [15], which are considered to be ahead of the Scandinavian countries in assessing flood damage, especially concerning the detail level of damage data available for analysis. For pluvial flood damage, however, the scientific literature is generally scarce. The study reveals that flood damage caused by pluvial flooding can occur even at low levels of precipitation. According to the radar data, damages occurred at levels as low as 12 mm/day. Further, daily precipitation of 25-40 mm can cause more damage than daily precipitation of 80-100 mm, depending on the temporal and spatial distribution of the rainfall. This is in agreement with the study of Hurford et al. [46] also demonstrating that pluvial flooding appears to occur at lower rainfall intensities and shorter return periods than those used as thresholds in different extreme Rainfall alert services. In our study, the intensity variable, taking both duration of the rainfall and aggregated amount of rain into account, had a statistically significant effect on flood damages while aggregated amount of rain by itself had not. This shows the importance of applying meteorological information that reflects the temporal perspective of a rainfall. In analysing insurance compensations at object level, the amount of precipitation the day prior to reported damage was included. The intention was to potentially capture the effect of a more or less saturated ground prior to the date of damages, seeking an explanation of why very low 19

precipitation amounts can cause great damage. The expectation was to see more damage if rain occurred the previous day, but the result pointed in the opposite direction. A 20 percent increase in the aggregated daily precipitation on the previous day leads to a decrease in mean payout amounts of between 4 and 6 %. This is valid for damages to movables as well as structural building damage. The effect does not have very high practical significance but is possibly an indication of a “warning effect,” which might lead to better preparedness among residents and homeowners (e.g., by checking gutters, valves, and drains) than when rain develops quickly and unexpectedly. Local vulnerabilities can be reduced through interventions among citizens and households [49]. Grothmann and Reusswig [44] claimed that self-protective behaviour by residents of flood-prone urban areas can reduce monetary flood damage by 80 percent and reduce the need for engagement from rescue services in emergent flood risk management. Some people are, however, more susceptible to harm than others due to their different capacities to address hazards [50]. The effect of self-protective behaviour is one of the aspects of pluvial flood risk that was not possible to include in this study but that we suspect might have had large effects on the size of the individual damage to residential property at object level. The reluctance of the insurance industry to share the exact location of the source of their flood damage compensations makes it impossible to combine the insurance data with information on resident’s behaviour. Further analysis of the effect of resident’s behaviour must be analysed using a different approach, e.g. a survey approach. The reluctance to share exact damage location has competitive reasons since the information can be used by competing insurance companies to gain market share. Insurance companies are also protective of their reputation and want to be perceived as loyal to their customers. There are known uncertainties both in the radar data and the damage data used in this study. The insurance payouts can be compensation for consequences other than direct physical damage. For example, compensation could cover the cost of a property inspection to decide if damage has occurred or the cost of drying out basements. These insurance compensations are often among the lower payments and rarely reflect structural damage to a property [47]. They do, however, still reflect the costs of consequences due to flooding. There is potential for improvement in both the accuracy of observations and the temporal and spatial distribution of the observations in damage data and weather data. Uncertainty in the data is a concern, and validating the radar data against gauge data is one possible approach to improving the accuracy of the precipitation-based variables in the analysis. Since radar data is better at capturing the spatial distribution of a rainfall compared to gauge data, the potential of using radar data for analysing damaging effects of precipitation ought to be good as long as the spatial resolution on damage data is good. In this study the spatial resolution of damage data was too poor to fully utilize the spatial information in the radar images. Another limitation following the non-precise location of damage data is that all geographical factors, such as topography, sealed surfaces, local runoff patterns, etc., cannot be used to explain the variation in damages. This is probably one major reason why the R2-values for the regression models were moderate for models A and B and very low for models 1-3. To 20

improve the explanatory capacity of regression models in the area, a combination of geography, meteorology and local hydrology, together with construction information on residential buildings would be needed.

Conclusions The objective of this study was to assess available flood hazard and flood damage data for quantitative flood risk assessment and to test and measure causal effects between extent of damage and damage influencing variables. Concerning damage to property at object level we found that daily rain amounts by itself could not explain variation in damage size. Adding a time perspective to the daily amounts did however significantly contribute, and rain amounts of >25 mm in less than 3 hours, >40 mm in less than 9 hours, and >60 mm in less than 24 hours increased the size of damage costs at object level. Further, rainfall occurring during night time and rainfall exposing central areas of the municipality caused higher damage cost, but when rainfall occurred the day prior to reported damages we detected somewhat lower costs at object level. Despite high statistical significance of the above mentioned damage influencing variables they all had low practical significance on the actual size of mean damage cost. The R2 values of the models (models 1-3) are very low, explaining only 3-5 percent of the total variation in insurance compensations. This means that the largest part of flood damage costs at object level, both to buildings and moveable property, are caused by variables that have not been possible to quantify by the available data set. In their current state, the models are not suitable for making inferences on the size of insured losses to residential property at object level. Other variables with potentially high explanatory power, such as topography, surface runoff index, drainage capacity, the portion of impermeable surfaces and building-specific characteristics, should be identified, quantified and integrated into the regression functions to make better estimations and to further develop pluvial RFDSs at object level. While we could not see practical significance of rain intensity on object level, rain intensity did have both statistical and practical significance upon the total estimated aggregated damage costs to residential areas at municipality level. The overall conclusion that can be drawn from analysis in section 4.1. an 4.2 is that rain intensity did not affect the mean compensation size per payoutbut it increased the total number of exposed and damaged properties, and thereby raised the overall cost of rainfall events. The main conclusion of this study concerns the importance of data availability, quality, and detail level to enable quantitative pluvial flood risk assessment. Discontinuity and heterogeneity of different datasets (damage data versus precipitation data) leaves very few possibilities for deriving RFDFs in a Swedish (or similar) flood context. By not having a responsible authority for framing, collecting and merging data concerning precipitation, floods, damages, technical systems, demographics etc., risk assessors at all levels are left with 21

relying on sporadic and unsystematic data documented by different private and public organisations. Overall, data availability concerning pluvial flood damages and their corresponding risk factors are poor. More efforts are needed to identify and quantify the risk factors with the largest impacts on pluvial flood damages in order to derive meaningful quantitative risk assessment tools for risk reduction policy purposes.

Funding: The study is funded by the Swedish Civil Contingencies agency (MSB). The funding part had no involvement in study design, analysis or interpretation of results. Acknowledgement: I wish to express my thanks to the Länsförsäkringar Alliance for sharing information on damage claims, and to meteorologist Barbara Blumenthal for her expert opinion and assistance in deriving categorical variables.

References [1] Guha-Sapir, D., Hoyois, P. & Below, R. (2014). Annual disaster statistical review 2013: The numbers and trends. Brussels: Cred 50 pp. Available at http://cred.be/download/download.php?file=sites/default/files/ADSR_2013.pdf. [2] Wang, Y., Li, Z., Tang, Z. & Zeng, G. (2011). A GIS-Based Spatial Multi-Criteria Approach for Flood Risk Assessment in the Dongting Lake Region, Hunan, Central China. Water Resources Management, 25 (13), 3465-3484. [3] Kreibich, H. & Thieken, A.H. (2008). Assessment of damage caused by high groundwater inundation. Water Resources Research, 44 (9). [4] MunichRe (2014). TOPICS GEO. Natural catastrophes 2013 Analyses, assessments, positions. Munich. [5] Bouwer, L.M. (2013). Projections of Future Extreme Weather Losses Under Changes in Climate and Exposure. Risk Analysis, 33 (5), 915-930. [6] Council of European Communities (2007). Directive on the assessment and managementof flood risks. (Official Journal of the European Union, 2007/60/EC). [7] Merz, B., Elmer, F., Kunz, M., Mühr, B., Schröter, K. & Uhlemann-Elmer, S. (2014).The extreme flood in June 2013 in Germany. Houille Blanche, (1), 5-10. [8] Jonkman S.N., Lentz A., Vrijling J. K. (2010). A general approach for the estimation of loss of life due to natural and technological disasters. Reliability Engineering and System Safety 95 (2010) 1123–1133. doi:10.1016/j.ress.2010.06.019

22

[9] Meyer, V., Becker, N., Markantonis, V., Schwarze, R., Van Den Bergh, J.C.J.M., Bouwer, L.M., Bubeck, P., Ciavola, P., Genovese, E., Green, C., Hallegatte, S., Kreibich, H., Lequeux, Q., Logar, I., Papyrakis, E., Pfurtscheller, C., Poussin, J., Przyluski, V., Thieken, A.H. & Viavattene, C. (2013). Review article: Assessing the costs of natural hazards-state of the art and knowledge gaps. Natural Hazards and Earth System Science, 13 (5), 1351-1373. [10] Jongman, B., Kreibich, H., Apel, H., Barredo, J.I., Bates, P.D., Feyen, L., Gericke, A., Neal, J., Aerts, J.C.J.H. & Ward, P.J. (2012). Comparative flood damage model assessment: Towards a European approach. Natural Hazards and Earth System Science, 12 (12), 3733-3752. [11] Elmer, F., Seifert, I., Kreibich, H. & Thieken, A.H. (2010). A Delphi method expert survey to derive standards for flood damage data collection. Risk Analysis, 30 (1), 107-124. [12] Messner, F., Penning-Rowsell, E., Green, C., Volker, M., Tunstall, S. & van der Veen, A. (2007). Evaluating flood damages: guidandce and recommendations on priciples and methods. T09-15]06-01). [13] Thieken, A.H., Müller, M., Kreibich, H. & Merz, B. (2005). Flood damage and influencing factors: New insights from the August 2002 flood in Germany. Water Resources Research, 41 (12), 1-16. [14] Smith, D.I. (1994). Flood damage estimation - a review of urban stage-damage curves and loss functions. Water SA, 20 (3), 231-238. [15] Merz, B., Kreibich, H., Schwarze, R. & Thieken, A. (2010). Review article "assessment of economic flood damage". Natural Hazards and Earth System Science, 10 (8), 1697-1724. [16] Green, C., Viavattene, C. & Thompson, P. (2011). Guidance for assessing flood losses. CONHAZ Report. WP6 Final Report) Middelsex University [17] Meyer, V., Schwarze, R., Becker, N., Markantonis, V., Bergh, J. C. J. M. V. D., Bouwer, L. M., Bubeck, P., Ciavola, P., Genovese, E., Green, C., Hallegatte, S., Kreibich, H., Lequeux, Q., Logar, I., Papyrakis, E., Pfurtscheller, C., Poussin, J., Przyluski, V., Thieken, A. H. & Viavattene, C. (2014). Assessing the Costs of Natural Hazards – State of the Art and the Way Forward. Hydrometeorological Hazards. John Wiley & Sons, Ltd. [18] Freni, G., La Loggia, G. & Notaro, V. (2010). Uncertainty in urban flood damage assessment due to urban drainage modelling and depth-damage curve estimation. Water Science and Technology, 61 (12), 2979-2993. [19] Houston, D., Werritty, A., Bassett, D., Geddes, A., Hoolachan, A. & McMillan, M. (2011). Pluvial (rain-related) flooding in urban areas: the invisible hazard. Joseph Rowntree foundation. [20] Pitt, M. (2008). The Pitt Review- Learning lessons from the 2007 floods. 22 Whithall, London: Cabinet Office. [21] Floodlist (2016). Europe. http://floodlist.com/europe/page/3 (accessed 2016-11-09) [22] Swedish Government Official Reports (2007). Sweden facing climate change- threats and opportunities. SOU 2007:60, Stockholm: Final report from the Swedish Commission on Climate and Vulnerability. [23] DHI (2015). Skyfall- en översikt. Presentation by Erik Mortensson (in swedish). http://www.offsak.se/upl/files/114529.pdf (accessed 16-12-15)

23

[24] MSB (2013). Pluviala översvämningar. Konsekvenser vid skyfall över tätort- en kunskapsöversikt (Pluvial flooding. Consequences of downpour over urban areas). https://www.msb.se/RibData/Filer/pdf/26609.pdf (In Swedish, downloaded 2016-11-09). [25] Nikulin, G., Kjellström, E., Hansson, U., Strandberg, G., Ullerstig,A. (2011). Evaluation and future projections of temperature, precipitation and wind extremes over Europe in an ensemble of regional climate simulations. Tellus A, 63 (1). [26] Hernebring, C., Dahlström, B. & Kjellström, E. (2012). Regnintensitet i Europa med fokus på Sverige- ett klimatperspektiv (in swedish). 2012:16) The Swedish Water and Wastewater Association. [27] Insurance Sweden (2016). High time for more climate adaptation. http://www.svenskforsakring.se/Startsidan-EN/Huvudmeny/Ourwork/Undersidor/Climate/Undersidor/High-time-for-more-climate-adaptation/ (accessed 201611-09) [28] Government Offices of Sweden (2016). Utredning för att stärka klimatanpassning i Sverige (Investigation to enhance climate adaption in Sweden). http://www.regeringen.se/artiklar/2015/11/utredning-for-att-starka-klimatanpassningen-isverige/ (in Swedish, accessed 2016-11-09). [29] Blahut, J., Poretti, I., de Amicis, M. & Sterlacchini, S. (2012). Database of geo-hydrological disasters for civil protection purposes. Natural Hazards, 60 (3), 1065-1083. [30] Merz, B., Kreibich, H., Thieken, A. & Schmidtke, R. (2004). Estimation uncertainty of direct monetary flood damage to buildings. Natural Hazards and Earth System Science, 4, 153-163. [31] Rözer V., Müller M., Bubeck P., Kienzler S., Thieken A., Pech I., Schröter K., Buchholz O., Kreibich H. (2016). Coping with Pluvial Floods by Private Households. Water 2016, 8, 304. [32] Zhou Q., Mikkelsen P.S., Halsnæs K., Arnbjerg-Nielsen K. (2012) Framework for economic pluvial flood risk assessment considering climate change effects and adaptation benefits. Journal of Hydrology 414–415 (2012) 539–549 [33] Bouwer, L.M., Crompton, R.P., Faust, E., Höppe, P. & Pielke Jr., R.A. (2007). Confronting disaster losses. Science, 318 (5851), 753. [34] Highfield, W.E., Norman, S.A. & Brody, S.D. (2013). Examining the 100-Year Floodplain as a Metric of Risk, Loss, and Household Adjustment. Risk Analysis, 33 (2), 186-191. [35] Kang, J., Su, M., Chang, L. (2005).Loss functions and framework for regional flood damage estimation in residential area. Journal of Marine Science and Technology, 13 (3), 193-199. [36] Zevenbergen, C., Cashman, A., Evelpidou, N., Pasche, E., Garvin, S. & Ashley, R. (2010). Urban Flood Management. [37] Leitao, J.P., Almeida, M.d.C., Simoes, N.E. & Martins, A. (2013). Methodology for qualitative urban flooding risk assessment. Water Science and Technology, 68 (4), 829-838. [38] Fatti, C.E. & Patel, Z. (2013). Perceptions and responses to urban flood risk: Implications for climate governance in the South. Applied Geography, 36, 13-22. [39] Kazmierczak, A. & Cavan, G. (2011). Surface water flooding risk to urban communities: Analysis of vulnerability, hazard and exposure. Landscape and Urban Planning, 103 (2), 185197.

24

[40] Douglas, I., Alam, K., Maghenda, M., Mcdonnell, Y., Mclean, L. & Campbell, J. (2008). Unjust waters: Climate change, flooding and the urban poor in Africa. Environment and Urbanization, 20 (1), 187-205. [41] Kelman, I. & Spence, R. (2004). An overview of flood actions on buildings. Engineering Geology, 73 (3-4), 297-309. [42] Fuchs, S., Heiss, K. & Hübl, J. (2007). Towards an empirical vulnerability function for use in debris flow risk assessment. Natural Hazards and Earth System Science, 7 (5), 495-506. [43] Grothmann, T. & Reusswig, F. (2006). People at risk of flooding: Why some residents take precautionary action while others do not. Natural Hazards, 38 (1-2), 101-120. [44] Fuchs S., Thöni M., McAlpin M.C., Gruber U., Bründl M. (2007.) Avalanche hazard mitigation strategies assessed by cost effectiveness analyses and cost benefit analyses—evidence from Davos, Switzerland. Nat Hazards (2007) 41:113–129. [45] Hurford, A.P., Parker, D.J., Priest, S.J. & Lumbroso, D.M. (2012). Validating the return period of rainfall thresholds used for Extreme Rainfall Alerts by linking rainfall intensities with observed surface water flood events. Journal of Flood Risk Management, 5 (2), 134-142. [46] Nordea. (oct 2012). Mortage loan. General conditions (in swedish). [Online] Available from: http://www.nordea.se/Images/39-13748/bolan-allmanna-villkor.pdf 2014-12-10. [47] Grahn, T. & Nyberg, R. (2014). Damage assessment of lake floods: Insured damage to private property during two lake floods in Sweden 2000/2001. International Journal of Disaster Risk Reduction, 10 (PA), 305-314. [48] Karagiannidis, A., Karacostas, T., Maheras, P. & Makrogiannis, T. (2009). Trends and seasonality of extreme precipitation characteristics related to mid-latitude cyclones in Europe. Advances in Geosciences, 20, 39-43. [49] Coulthard, T.J. & Frostick, L.E. (2010). The Hull floods of 2007: implications for the governance and management of urban drainage systems. Journal of Flood Risk Management, 3 (3), 223-231. [50] Clark, G.E., Moser, S.C., Ratick, S.J., Dow, K., Meyer, W.B., Emani, S., Jin, W., Kasperson, J.X., Kasperson, R.E. & Schwarz, H.E. (1998). Assessing the vulnerability of coastal communities to extreme storms: The case of reverse, MA., USA. Mitigation and Adaptation Strategies for Global Change, 3 (1), 59-82.

25