In Switzerland, there seems to be a growing awareness that just as overtopping rivers and lakes pose substantial flood risks for society, so too does flooding that takes place far away from watercourses. All across Europe, there are well-known examples of such inland flood events. In 1988, for instance, a devastating flood occurred in Nîmes, France e.g.,. In 2007, Hull, UK, was affected by flooding e.g.,. One year later Dortmund, Germany, experienced widespread flooding e.g.,. In 2011, the Danish capital Copenhagen was affected heavily by flooding e.g.,. The Swiss canton of Schaffhausen was affected severely in 2013 e.g.,. On the same day in 2014, the Dutch capital Amsterdam e.g., and Münster, Germany, experienced substantial flooding . These events in Europe share a common thread, which stems from their origin as inland floods, triggered by heavy precipitation, but are mostly unrelated to watercourses.

As the definition of such floods is not straightforward, we adopt the term surface water floods (SWFs) for now, use it for non-fluvial floods in general and discuss the terminology in Sect.  in detail. Inherently, SWFs are not constrained to areas close to watercourses but can occur practically anywhere in the landscape . Consequently, such floods are difficult to document, study and forecast e.g., and related data are scarce e.g.,. mention the lack of data and the impact on small spatial scales as possible explanations why relatively little scientific research has been dedicated to such SWF in comparison to fluvial floods. In contrast, gray literature covers the topic of SWFs rather extensively, which is reflected by the availability of many guidelines and manuals discussing how to prepare for and manage such floods, for instance for single objects in Switzerland or on communal or regional levels in Germany e.g., or France e.g.,. This might exemplify that the scientific flood risk community is indeed quite oblivious of resourceful gray literature . In any case, it indicates that the topic is a concern for the people, the responsible authorities and other stakeholders. In order to reduce the risk, an effective approach is to focus on the physical protection of exposed objects e.g.,. Although this strategy is certainly heading in the right direction, we have to be conscious about the basis on which current and future decisions concerning SWFs are made. Undoubtedly, the lack of quantitative data and studies hampers our process understanding . Therefore, the underlying crucial question is “how can we reduce losses from natural hazards when we do not know … when and where they occur?” .

Owing to vast river discharge time series, fluvial floods can be well predicted along gauged rivers . As there are no such data concerning SWFs , we must exploit other data sources in order to quantify the relevance of this flood type in space and time. Possible data sources include, but are not limited to, insurance claim records e.g.,, disaster databases e.g.,, press reports e.g., and interviews with or reports from affected people e.g.,. All data sources are probably subjected to a varying degree of a so-called ”threshold bias”, which refers to the bias introduced due to varying damage inclusion criteria . Disaster databases only list events that exceeded predefined loss and/or fatality thresholds . Similarly, damage data based on news reports are subjected to unknown thresholds, as damage is only reported if it is found to be interesting enough. As interview campaigns are more likely to be initiated after devastating flood events, such data are biased towards more extreme events, as well . Insurance claim records are likely affected the least by a threshold bias; as long as the related insurance policy stays the same, insured objects are not changing greatly over time and the deductibles are low or can be accounted for.

Damage claim records of insurance companies are therefore a profitable data source. Not surprisingly, they have been the base for several studies related to SWFs e.g.,. Unfortunately, insurance claim data are generally difficult to collect, since most insurance companies do not publish or provide loss data due to confidentiality issues . Furthermore, analyses based on such data are often impaired by the data's spatial or temporal aggregations. For instance, the limited usefulness of monthly aggregated data was demonstrated by , while pointed out some limitations of insurance data aggregated to administrative units, which do not have homogeneous topographical properties. As insurance companies usually do not assess and record detailed information for each damage claim, it is difficult to verify and differentiate the cause of each damage without at least knowing the explicit location of the damaged object. This is particularly important, as the corresponding data often cover different processes without explicit classification: for instance, had to exclude all damage records with dates that coincided with dates of known fluvial flood events to obtain a subset of SWF-related claims. chose a more elaborate method of applying a statistical filter based on the assumption that rainfall-related damage is clustered around wet days, while other causes of damage occur on any day throughout the year. Finally, even though many or even all buildings are insured against floods in several countries (e.g., in Sweden, as in ; or in the Netherlands, as in ), usually only a subset of all objects is covered by the obtained data records. This is due to the fact that the objects are usually insured by many different companies, each having a different (unknown) market share. In addition, these shares are generally not constant over time either but may fluctuate heavily over time and space, as exemplified by . These spatial and temporal changes need to be taken into account, which is often not trivial.

Luckily, most of these limitations are not applicable for damage claim records of the Swiss public insurance companies for buildings (PICBs). In Switzerland, PICBs are present in 19 out of the 26 cantons, whereas each company insures (almost) all buildings within the respective canton due to their monopoly position and because the insurance is generally mandatory for all house owners e.g.,. Beside other natural hazards, the insurance covers damages caused by floods, which includes both fluvial floods and SWFs. Data records of PICBs are, therefore, exceptionally interesting for analyzing floods in general and SWFs in particular. Most PICBs have shown a general interest about research on this topic and, thus, were willing to provide flood claim records including the address of each damaged object.

Based on these data, the first aim of this study is to provide a method with which each claim can be classified as being caused by SWFs or fluvial floods. Second, based on the classified claim records, we aim to answer the fundamental question of how relevant damage caused by SWFs is, as well as where and when such damage occurs in space and time. The underlying data set stems from 13 PICBs and covers 48 % of all buildings in Switzerland. Thus, the data set is representative of most of Switzerland, except for southern Switzerland (i.e., Western Inner Alps and Southern Alps). The analyzed data records all end in 2013 and extend back to at least 2004, but even up to 1981 depending on the corresponding PICB. As the PICBs, save a few exceptions, insure only property and not its contents, this study only considers damage to buildings. More specifically, this study is limited to direct tangible flood damages to buildings, i.e., monetary losses caused by the buildings' direct contact with flood water . Thereby, we acknowledge that these damages only constitute a portion of the total flood losses.

We have identified a lack of a common terminology concerning SWFs. Therefore, we dedicate the following Sect.  to a short overview of terms that are currently being used to address flood types that could be categorized as SWFs, as mentioned before. In Sect.  we describe the data in detail and introduce a method to differentiate SWF damage from fluvial flood damage. Thereafter, in Sect. , we present general characteristics of the number of claims and associated loss caused by SWFs in comparison to fluvial floods. Furthermore, we present the spatial and temporal characteristics of damage caused by SWFs in Switzerland during the last decades and discuss the results in Sect. . Finally, by providing concluding remarks, we conclude the study (Sect. ).

Flooding is a complex interlinked system, affecting many aspects of the physical, economic and social environments acting at different spatial and temporal scales . As such, flooding involves a wide range of interconnected hydraulic subsystems and processes . Therefore, the classification of such a complex process like flooding is not trivial, particularly in practice. At the same time, many of the terms used to address flood types or involved hydrological processes in relation to SWFs are either used ambiguously in the literature or not well-defined. To prevent terminological ambiguities, we first introduce relevant hydrological processes, which helps to distinguish SWFs and fluvial floods. Thereafter, we elaborate related flood terms for a clearer definition of SWFs and provide recommendations for these terms' future reference.

SWFs are characterized by overland flow and ponding, which can be defined as follows. As precipitation reaches the land surface, different runoff generation mechanisms determine whether water starts to pond and whether overland flow is generated e.g.,. The water may then take several routes towards the stream channels , as depicted in Fig. . The flow path along the land surface is sometimes ambiguously referred to as “surface runoff” but is better defined by the widely-used term “overland flow” . However, in the literature, this distinction is inconsistently made, whereas either of the terms or even both are used. We adopt the term overland flow and, thereby, mean the transport of water downhill at the land surface as thin sheet flow or anastomosing braids of rivulets and trickles until the water reaches or is concentrated into recognizable streams .

The propagation and accumulation (i.e., ponding) of overland flow can be considered as a flood, which in the glossary of is defined as “the overflowing of the normal confines of a stream or other body of water, or the accumulation of water over areas that are not normally submerged”. As long as the water is directed towards a watercourse, but has not yet reached it, the flood can be regarded as a SWF, as defined later. Thus, the notable difference between a SWF and a fluvial flood is that in the former case, water is making its way towards a watercourse, whereas in the latter case flooding stems from a watercourse (Fig. ).

As outlined previously, different flood terms are used in relation with SWFs. For a better distinction of these terms, we discuss each term and give recommendations about their future reference. A summary of the terms is presented in Table .

“Pluvial floods” are caused by intense rainfall that, for whatever reason, cannot be drained by natural or artificial drainage systems, thereby ponds in local depressions or propagates along the surface as overland flow , before it possibly, but not necessarily, reaches or is concentrated into regular watercourses. The term pluvial flood is often used synonymously with SWF although, according to , SWFs have a broader meaning. Namely, in addition to pluvial floods as defined above, the term SWF also includes flooding from sewer systems, small open channels, culverted watercourses or flooding from groundwater springs . Therefore, SWFs can be regarded as the most general definition of rainfall-related (pluvial) floods. For future studies, we recommend using these two terms distinctively, depending on the corresponding context.

The term “muddy flooding” is well-established and refers to floods that are formed by muddy runoff from agricultural fields that damage adjacent properties downslope . Here, the term is mentioned to point out that this flood type is implicitly included by the definition of SWFs and pluvial floods.

The term “flash floods” is used quite ambiguously in the literature . Traditionally, it refers to fluvial floods triggered by short, intense and local storm events e.g.,. However, the term may include other causes as well . Moreover, the term has increasingly been used in relation to pluvial flood types see. Apparently, the term is often used in this context by publications in German using the translated term Sturzflut see. For future reference, we recommend adopting the term flash flood only in the traditional sense and use the applicable term, i.e., pluvial flood or SWF, for all other cases.

The terms “urban” or “intra-urban” are mainly used as a specifier of the geographical extent of a flood or the main focus of the corresponding study see. If applicable, the use of this term as a specifier in combination with other flood terms can be recommended, since the corresponding flood type is thus better defined. However, we suggest refraining from the isolated usage of the term, as in “urban flood” for instance, since the flood type is thereby not unequivocally defined. In case the term is intentionally used in such a broad context, we recommend mentioning this explicitly.

Finally, we deem it necessary to introduce a further distinction for a better understanding of this study's results. Namely, it is important to note that the term “flood” is sometimes implicitly used in the hydrological sense but sometimes also in the context of “damaging floods” . In the former case, any inundation of land is considered, while in the latter case the flood necessarily interacts with the societal system causing adverse effects . Thus, our results represent only damaging floods, as this study is based solely on the exploitation of damage data. Note that this distinction is visualized in Fig. .

The compiled data set is based on flood damage claim records from 14 different PICBs. In addition, we obtained similar records from Swiss Mobiliar, a cooperative insurance company (CIC). The corresponding data records were solely used to support the parametrization of the classification scheme. Thus, they were not part of the data analyses, as elaborated in more detail in Sect. .

As mentioned before, each PICB holds a monopoly position and, thus, insures virtually every single building within the respective canton against various natural hazards including flooding. Therefore, damage caused by water entering the building envelope at the surface is insured, while damage associated with direct intrusion of groundwater or backwater from the sewer, as well as flooding from dams or other artificial water structures, is generally excluded. As a consequence, water-related damage covered by PICBs is caused by either SWFs or fluvial floods, whereas the insurance companies themselves do not differentiate the two processes . Therefore, similar to other studies e.g.,, the data have to be classified first. However, in contrast to the aforementioned studies, the claim records were provided in a spatially explicit way, enabling a classification based on each claim's geographical context.

Following the data processing procedure depicted in Fig. , we first describe the compiled data set and the harmonization and geocoding thereof (Sect. ). Then we introduce a method to differentiate claims associated with SWFs and fluvial floods (Sect. ) and, thereafter, we discuss the necessary normalizations of the data (Sect. ). Note that the classification scheme is described as generally as possible to make its application to other contexts and countries as straightforward as possible. However, it could not be prevented that the classification scheme is adapted to some national characteristics, in particular concerning the properties of the considered Swiss flood maps. The specific input data for each data processing step listed in Fig.  are described in detail in Table .

After presenting the validation's results (Sect. ), we quantify, characterize and compare the damage caused by SWFs with damage caused by fluvial floods (Sect. ). In the following, we present the spatial distribution of SWF damage (Sect. ) and show how the damage evolved within the last 20 years (Sect. ).

Note that in the following damage claims classified as A or B, i.e., (most) likely surface water floods, are regarded as damage caused by SWFs, if not stated otherwise. Analogously, damage claims classified as D or E, i.e., (most) likely fluvial floods, are counted as damage caused by fluvial floods. Claims of class C, i.e., fluvial flood or surface water flood, are not counted for one or the other flood type unless total values are presented.

The key to the exploitation of the insurance data with regards to SWFs lies in the classification of the damage claims (beside the provision of the data in the first place). The classification scheme, as introduced in this study, is based on the geographical location of each damage with respect to known fluvial flood zones and the hydrological network. On the one hand, this obviously requires spatially explicit damage data. On the other hand, it provides a reproducible, objective and, most importantly, an independent classification. These characteristics are important, as the following examples highlight. had to exclude damage that occurred on the same day as known fluvial floods in order to distinguish pluvial from fluvial flood claims. However, our results show that fluvial flood damage almost always coincides with damage caused by SWFs. Consequently, excluding damage occurring on the same day as fluvial floods likely introduces a bias. Another example is the statistical model applied by in order to differentiate rainfall-related damage clustered around wet days from non-rainfall-related damage occurring throughout the year. Thereby, the classification of each claim is not independent anymore but depends on how many other damage claims occurred on the same day.

Although the classification scheme presented in this study has striking advantages, it has the following shortcoming: as overland flow propagates over the land surface, it may eventually reach a watercourse (see Fig. ). Areas alongside watercourses, where the overland flow joins the river, may be a flood hazard zone. If so, all claims in that specific area are classified as a fluvial flood, even though the claim might have been caused by incoming overland flow. However, a more qualified classification entails likely event-specific, time-consuming manual assessments. In fact, it is extremely difficult to disentangle the different flood types, even more so for events in the more distant past and if no data with that particular focus are available. In contrast, for claims that are located far away of any watercourse, it is very unlikely that they are affected by watercourses at all. Therefore, our method renders a lower boundary of claims associated with SWFs, in essence. In reality, the numbers are likely higher, but, as mentioned before, disentangling the flood types within their overlapping domains is difficult.

By applying the classification scheme to the harmonized damage data, we could show that SWFs caused almost the same number of damage claims as fluvial floods. Thereby, our results confirm anecdotal evidence that indicated similar numbers. For instance, one of the few quantitative studies about SWFs in Switzerland reported that at least half of the flood damage claims in the canton of Aargau were caused by overland flow . However, the study is comprehensible in terms of neither the applied methods nor the underlying data and covers only a small part of Switzerland. Thus, for the first time, we can present sound evidence about the relevance of SWFs in Switzerland based on a large data set including more than 30 000 damage claims covering 15 years and 48 % of all Swiss buildings.

Despite the remarkably high number of damage claims caused by SWFs, our results show that SWFs only account for roughly one-quarter of the total loss, which is in line with results from the pilot study i.e.,. Nevertheless, the associated yearly loss is highly significant, as the following numbers exemplify: the median of total yearly losses to buildings caused by fluvial floods within the considered regions is even slightly lower (5.0 mio CHF yr-1) than the median of SWF losses (5.9 mio CHF yr-1) based on the data set D2 covering the period of 1999–2013 (see Table ). However, the mean yearly loss of fluvial floods is more than 3 times the loss caused by SWFs (i.e., 31.3 mio CHF yr-1 versus 10.1 mio CHF yr-1, respectively). The difference between the maximum yearly losses caused by each flood type is even more pronounced: while the maximum loss of SWFs amounts to 38.3 mio CHF yr-1 in 2005, fluvial floods caused 234.3 mio CHF yr-1 in the same year, which corresponds to a factor of roughly 6.

These observation concerning annual flood losses are supported by the characteristics of the individual losses. Their exploration (Fig. ) expressed that the range of loss per claim is much narrower for SWFs than for fluvial floods. As SWFs are expected to be associated with significantly lower flow depth than fluvial floods, this might be one of the main reasons for the lower associated loss, since water depth is among the most significant single impact parameters for structural damage to residential buildings e.g.,. Interestingly, the median loss of each claim associated with fluvial floods is also rather low, although significantly higher than the median loss of claims related to SWF. However, the highest losses per claim are caused by fluvial floods during the most severe events within the study domain (Fig. ). As during extreme events, larger areas are affected and the associated shares of objects inundated by large water depths are higher , higher losses per claim can be expected. Along the same lines, report that the most severe events contribute to more than half of the estimated total loss and found an even higher share for flood losses in the whole of Europe. Undoubtedly, loss ratios are higher during more extreme events . Although this probably also holds true for damage caused by SWFs, such damage certainly seems less influenced by the severity of the event (see Figs.  and ). Consequently, SWFs may rarely cause the total destruction of a building, and associated loss ratios may, thus, mostly be well below 1.

As outlined in the introduction, this study is limited to direct tangible loss to buildings. Therefore, the absolute loss values are low in comparison to other loss estimations that include other losses, as well. For instance, report a mean financial loss of 317.2 mio CHF yr-1 between 1972 and 2007, which is roughly 7 times higher than the mean of all flood losses to buildings, as represented by our data set. For one, the data published by cover the whole of Switzerland and consider a longer period. More importantly, however, these estimates also include damage to infrastructure, forestry and agricultural land, in addition to damage to buildings and their content. Therefore, the associated losses are inherently higher than the numbers presented in this study. This exemplifies that one has to be careful when comparing values from different data sources . Moreover, it highlights the fact that damage to buildings are associated with just a small fraction of the total loss caused by SWFs for the society. Nevertheless, these data serve well for assessing the relevance of SWF damage in Switzerland, especially when considering relative values.

The spatial distribution of damage caused by SWFs can be deceiving: obviously, an area with a higher building density will likely result in a larger number of damage claims compared to an area that is less populated (Fig. b versus c). Therefore, it is important to have a look at relative values, as well (Fig. a). Thereby, the effect of higher values caused by a denser number of buildings is considered. However, the relative values are quite sensitive in sparsely populated areas. A damaged house with virtually no other houses in the vicinity will produce a high relative value or a low value if the same is not affected. In contrast, in more populated areas, the relative value will not change much in case a building is more or less damaged. Thus, to obtain a complete picture, the relative and absolute values should be considered alongside the building density. In that way, the most exposed areas can be identified, like the two highlighted areas in the Western Plateau that are associated with high relative and absolute numbers of damage claims (Fig. ).

Furthermore, it is important to keep in mind that in case an area has no registered damage, it does not necessarily mean that the area has not been affected by a floods at all. It just indicates that either no buildings were in the vicinity of the flooded area or the buildings were properly protected against such floods. Therefore, damage records can only indicate floods that lead to some sort of damage and never to the occurrence of floods in the hydrological sense, as discussed in the introduction (see Fig. ). However, understanding the characteristics of damaging floods can open the stage to understand the process in a broader context, as well.

The temporal distribution of claims related to SWFs exhibits a distinct seasonality (Figs.  and ). Similar to the flood losses reported by , most damage clearly occurs in summer, with a few exceptions. Therefore, thunderstorms associated with short but intense rainfall are certainly an important driver of SWF damage. Nevertheless, long duration rainfall events are also responsible for a large share of SWF damage claims, highlighted by the most severe events that are mostly associated with long duration precipitation. In contrast, much fewer damage claims are caused in spring and fall, and virtually no damage claims are caused in winter. Damage to buildings in winter can likely be attributed to rather local events coinciding with conditions promoting overland flow generation such as rain on frozen soils. Overall, these observations have important implications for assessing the hazard of SWFs. In particular, simply focusing on high-intensity rainfall events may lead to an underestimation of the risk of SWFs.

Although the time series is relatively short, the data do not exhibit any increasing trends of SWF damage in the period of 1993–2013. Obviously, the general increase of absolute loss in time, which can be found in our data as well, is eliminated when the data are normalized. Thus, as suggested for instance by , the increase in loss can be mainly attributed to the socioeconomic development. However, we did not consider further aspects that could have an influence on such trends, such as a change in vulnerability . Moreover, insurance or local governmental policies that might have changed over time were not taken into account either. Nevertheless, it is important to note that increasing absolute losses are most likely attributable not to climate change but to socioeconomic factors e.g.,. Consequently, the major associated risks related to SWF damage is not climate change but the increased exposure due to population growth and increasing wealth. This has implications for decision and policy makers, as well as for insurance companies and similar stakeholders.

Indeed, the flood processes are a complex interlinked system, as stated. In fact, the insurance data illustrated that damage caused by SWFs occur (almost) always alongside claims caused by fluvial floods (see Fig. ). Be it a short and intense thunderstorm or a long duration event, rainfall is the main trigger of every SWF and (almost) every fluvial flood. Understandably, if there is enough rainfall to cause a SWF, it may as well cause or at least contribute to a fluvial flood once part of the water reaches the next watercourse. Undoubtedly, severe events that include hundreds or thousands of damage claims entail a combination of flood processes, while, of course, some local events may be associated with a single flood process only.

In this study, we have presented a simple and pragmatic approach of how spatially explicit insurance data records can be exploited to investigate damage caused by SWFs. The method provides a robust lower estimate of SWF damage. Using the presented percentile values (Table ), the method is applicable for classifying any claim in Switzerland except in the Western Inner Alps and Southern Alps, where data were lacking. For these regions, appropriate values could be approximated. Moreover, the method is transferable to other regions and countries but has to be adapted to locally available flood hazard maps.

There seems to be a consensus among practitioners and experts that SWFs are responsible for a large share of all flood damage. However, this perception stems not from quantitative research but rather from single case studies or practical experience and, thus, lacks evidence. With the study at hand, we are able to quantify the striking relevance of SWFs in Switzerland based on a sound data basis, regionally representing 39–100 % of all buildings over a period of 15 years. The data reveal that SWFs cause nearly as many damage claims as fluvial floods. In contrast, SWFs account for roughly one-quarter of the direct tangible losses, driven by lower losses per SWF claim. This hints at the different processes' characteristics with generally low flow depths associated with SWFs, opposed to both low and high, static and dynamic, flow depths during fluvial floods that are additionally sensitive to the severity of an event.

The most affected areas are clearly the Western Plateau, in both relative and absolute terms, followed by the Eastern Plateau and the Jura Mountains. The more mountainous regions, i.e., the Northern Alps and the Eastern Inner Alps, are affected less. Notably, there are also large differences between the spatial distribution of damage within each region. By relating the absolute number of damage claims to the number of buildings in the vicinity, the effect of varying building densities can be considered. Nevertheless, in sparsely populated areas the relative numbers are sensitive and, thus, less robust due to the particularly low building densities. Furthermore, not all regions are affected by SWFs to the same extent throughout the year. However, in all regions most of the damage occurs in summer, save a few exceptions.

In general, the spatial and temporal distribution of SWF damage is complex. Different factors might be responsible for high damage within certain areas or during certain periods. For instance, the meteorological forcing differ spatially and temporally, the predisposition due to unfavorable soils or land use practices play a role, past human interventions such as the installation of drainage and the removal of small natural rivulets can have an influence, but also slightly differing practices by the insurance companies or different rules applied for buildings to be built might be relevant. Undoubtedly, we stand at the beginning of better understanding SWFs in Switzerland and also on an international level. Meanwhile, a common terminology is the base to strengthening and extending the science within this field across the countries' borders.

This study highlights the fact that SWFs are a highly significant flood process in Switzerland. Unlike for fluvial flood hazards, there is no publicly available up-to-date information about the hazard of SWFs, in spite of the process's obvious relevance. Since SWFs can occur practically anywhere in the landscape, it is paramount to have detailed information about local SWF hazards. Such information can help to make well-founded decisions by all different stakeholders, e.g., planning and installing appropriate property protections by house owners, applying measures to reduce overland flow generation on agricultural fields by local farmers, providing surface retention ponds by municipalities or amending regulations to prevent SWF damage by the federal government. However, as a first priority, SWFs in general and the influencing factors of SWFs in particular should be further studied and, ultimately, better understood.

As a first step in this direction, we propose that SWFs should not be regarded as an isolated process by itself. A better way is probably to extend our focus from rivers and lakes alone to hidden rivulets, covered drains, the sewer system, impervious areas, agricultural fields and headwaters, which all contribute to the generation of SWFs. Therefore, we should regard overland flow and ponding as an integrated part of our catchments. In this manner we may start to understand the complex interlinked flood processes better in the future.