Available via license: CC BY 4.0
Content may be subject to copyright.
1
INSYDE-content: a synthetic, multi-variable flood damage model
to household contents
Pradeep Acharya1,2, Mario Di Bacco3, Daniela Molinari2, Anna Rita Scorzini1
1 Department of Civil, Environmental and Architectural Engineering, University of L’Aquila, 67100 L’Aquila, Italy
2 Department of Civil and Environmental Engineering, Politecnico di Milano, 20133 Milano, Italy
5
3 Department of Civil and Environmental Engineering, University of Florence, 50139 Firenze, Italy
Correspondence to: Anna Rita Scorzini (annarita.scorzini@univaq.it)
Abstract. This paper introduces INSYDE-content, a novel, probabilistic multi-variable synthetic flood damage model
designed to analyze physical damage to household contents on a component-by-component basis. The model addresses a
critical gap in current modeling tools, which often overlook the significance of household contents in overall damage
10
assessments. Developed through an expert-based approach and grounded in the scientific and technical literature, INSYDE-
content leverages desk-based data to characterize model features, including uncertainty treatment arising from incomplete
input data. A sensitivity analysis and a benchmark test against observed losses demonstrate the robust performance of the
model and highlight the contribution of different features to damage mechanisms affecting house contents. While in this
study INSYDE-content is tailored for illustrative purposes to the hazard, vulnerability and exposure characteristics of
15
Northern Italy, the model is highly adaptable, allowing for its application to different regional contexts through appropriate
customization.
1 Introduction
Floods have been one of the major natural disasters around the world for centuries (Brázdil et al., 2006; Tanoue et al., 2016).
With the changing weather patterns associated to climate change, both the frequency and severity of floods have been on the
20
rise (Hirabayashi et al., 2013; IPCC, 2023). Several flood management approaches have been formulated and implemented
over the years to mitigate losses, with a shift from focusing solely on flood hazard control to a more holistic flood risk
management (Plate, 2002; Sayers et al., 2002; Gralepois et al., 2016; Disse et al., 2020). In this change, damage estimation
has become a key element in formulating effective policies for flood risk mitigation in both rural and urban sectors (Merz et
al., 2010; Marín-García, 2023).
25
Flood damage models allow quantifying the extent of damage in both relative and absolute terms. They can be classified as
univariable or multi-variable based on the number of variables involved in the model. A further classification distinguish
between empirical and synthetic models based on the adopted methodology for their development: empirical models use
historical data to relate vulnerability and hazard variables to damage, while synthetic models are founded on an expert
judgment that adopts a what-if approach for the definition of the expected damage (Merz et al., 2010; Jongman et al., 2012;
30
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
2
Dottori et al., 2016; Gerl et al., 2016; Martínez-Gomariz et al., 2021). Synthetic models have better flexibility in
transferability in both space and time if their modeling assumptions and data are properly adapted (Merz et al., 2010;
Scorzini et al., 2021; Di Bacco et al., 2024), while multi-variable models allow more reliable estimations (Schröter et al.,
2014; Wagenaar et al., 2018; Amadio et al., 2019; Paulik et al., 2023; Xing et al., 2023).
Among the various exposed assets, damage modeling for the residential sector has been the most extensively investigated
35
and developed, with numerous models available in the literature (Gerl et al., 2016). Nevertheless, most of these models are
devised to assess building damage only, often neglecting or adopting a simplified approach to household contents (i.e., those
items within the house that are not permanently installed in it, such as the furniture and appliances). The report of the US
Army Corps of Engineers, “Catalog of Residential Depth-Damage Functions” (Davis and Skaggs, 1992), is one of the early
efforts to address the difficulties of appraising flood damage to house contents, linked to the many economic, social, and
40
structural factors involved in the modeling. By inventorying the contents of sample houses and pricing them, the report
suggested that damage to contents can reach up to roughly 50% of the building damage, thus highlighting the importance of
house contents on the overall damage figure in inundation scenarios.
More recently, a few studies have attempted to address the research gaps on modeling flood damage to contents across
various spatial scales by adopting different approaches. Zhai et al. (2005) analyzed flood damages for the 2000 Tokai flood
45
in Japan, identifying inundation depth, duration and household income as key explicative variables of flood damage to both
buildings and contents. Similar results were found by Wahab and Tiong (2016), who proposed a multivariate flood damage
model for the residential sector in Jakarta, Indonesia.
Romali and Yusop (2020) developed an empirical multiple regression flood damage model for Kuantan, Malaysia,
incorporating residents’ occupation, household income and house types as explanatory variables. They found that content
50
loss was positively correlated with inundation duration along with household income and inundation depth. Shrestha et al.
(2021) used household surveys in Myanmar’s Bago region to create new local flood damage functions based on building
characteristics. Ahadzie et al. (2022) studied flood impacts on residential buildings in urban settlements across five flood-
prone regions in southern Ghana, with data showing significant damage to contents, such as furniture and clothes.
Mosimann et al. (2018) developed and cross-validated two linear regression models based on insurance data to estimate
55
building and content losses in Switzerland. Their findings revealed that content losses accounted for 21% to 36% of total
losses and even surpassed building losses in cases of low-severity damages.
Endendijk et al. (2023) developed empirical, multivariate vulnerability models to assess flood damage to buildings and
household contents in the Netherlands, incorporating the influence of flood mitigation measures on damage outcomes.
Nofal et al. (2020) addressed the issue of data scarcity and the challenges of deterministic models by developing synthetic
60
uni- and multi-variable component-based flood fragility functions. The method involved dividing the building into 65
components, with some of them representing household contents, such as chairs, desks, TVs, electric appliances, sofas,
kitchen cabinets, beds and mattresses. Each component was assessed across five predefined damage states ranging from DS0
(insignificant damage) to DS4 (complete damage). The proposed approach was applied on a hypothetical single-family
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
3
residential wood building, with contents distributed in it based on practical considerations. However, these basic assumptions
65
significantly limit the model’s real-world applicability, as they fail to account for the substantial variability in housing
characteristics and household contents.
Carisi et al. (2018) highlighted the challenge of developing comprehensive flood damage models for buildings and contents
in the Italian context. Using ex-post data from the 2014 Secchia flood in the Emilia-Romagna region, they created empirical
uni- and multi-variable flood damage models for buildings. Damage to contents was instead assessed indirectly using a
70
simple square root regression relationship to building damage. By overlooking the complex induction mechanisms leading to
damage and the related key factors, such as content distribution within the building, this approach, although practical,
significantly hinders the model’s ability to provide insights into content vulnerability in flood scenarios.
Overall, the literature highlights that modeling flood damage to contents presents a significant challenge due to the inherent
variability of exposed items within buildings and the complexity of describing their susceptibility to floods. This variability
75
increases across different regions, further complicating the spatial transferability of models. Nonetheless, current research in
this area remains limited, underscoring the necessity for continued exploration and refinement of methodologies and tools.
The present study addresses this gap by introducing INSYDE-content, a micro-scale, multi-variable synthetic flood damage
model for household contents. Similar to the original INSYDE model for buildings (Dottori et al., 2016), INSYDE-content
employs a component-wise, probabilistic, “what-if” approach, characterized by a flexible design that can be adjusted to the
80
local features and data availability of the region in which it is implemented, enhancing its applicability across different
geographical areas. The clarity and explicitness of input data and assumptions play a crucial role in models’ applicability.
The paper is structured to first outline the methodology for model development, exemplified for the context of Northern Italy
(Po River District, Figure S1 of the Supplement 1). Insights on the influence of the different input variables on household
contents losses and their relationships to building damage are then derived based on a probabilistic sensitivity analysis.
85
Finally, the model’s performance is evaluated using ex-post loss data from two flood events that occurred in the region.
2 Methodology
2.1 Model development
INSYDE-content adopts the general model framework proposed in the original INSYDE for buildings (Dottori et al., 2016),
where damages are first modelled component-wise in physical terms and then converted into monetary values using the full
90
replacement costs derived from reference price lists. The overall economic damage D to house contents in each building is
calculated by summing the replacement costs Ci for each of the n content items within the building that is expected to be
damaged:
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
4
where no.dam.itemsi represents the number of damaged items for each specific household content i and unitpricei the
95
corresponding unit replacement prices. The value of no.dam.itemsi depends on flood event features (such as inundation depth
and duration, flow velocity, and water quality, in terms of the presence of pollutants or sediments), as well as on the
vulnerability and exposure of the affected objects:
Based on practical considerations derived from household surveys, INSYDE-content includes 11 standard items: beds, sofas,
100
wardrobes, dining table setup (dining table and chairs), kitchen setup (lower and upper cabinets), TVs, washing machine,
oven and microwave oven, dishwasher and refrigerator.
Since information on the actual number and distribution of contents within buildings is typically not known at large spatial
scales, in model development it is necessary to introduce a probabilistic method that estimates their presence based on more
commonly available data, such as building features. This approach allows for a more accurate assessment of the exposed
105
items without needing for detailed, point-wise evaluations, which are often impractical due to the variability in individual
choices of households on content selection for their properties. To this aim, in the tailored INSYDE-content model for
Northern Italy (Po River District), empirical data derived from virtual surveys of buildings have been used to establish
reliable estimates of content distribution as a function of certain building features. The virtual surveys involve analyzing real
estate listings to extract key information from advertised posts, architectural drawings and photos detailing the buildings and
110
their contents (Scorzini et al., 2022). Given the potential inconsistency in data completeness and quality across real estate
platforms, only listings with complete information about the building and its contents have to be considered. To qualify for
selection, the advertisement has to include at least information on interior details, main geometrical attributes and
architectural layout and profiles, as well as a sufficient number of photos taken from various angles to cover most of the
areas within all rooms. The applied criteria in the analyzed case resulted in only 60 houses being deemed suitable for
115
analysis out of approximately 500 examined. Indeed, in many cases, only partial information could be obtained due to either
a lack of data or privacy issues. Additionally, geometrical features did not always align with secondary data, such as building
footprints available from other databases.
The information collected during this process can be classified into two categories: housing characteristics (e.g., location,
number of floors, building type, finishing level, inter-story height, footprint area, surface area, external perimeter, year of
120
construction) and house contents (e.g., number and size of beds and sofas, number of furniture pieces, placement height of
appliances). For more detailed examples, please refer to the Supplementary Material 2 included in this study.
For practical reasons, the raw data acquired from the survey need to be transformed into more standardized variables on a
component-by-component basis. For instance, different types of beds (single and double) have been represented by the single
standard variable called BEDLeq, where 1 BEDLeq equals 1 large double bed or 2 single beds. The same approach was
125
applied to other items, such as sofas, wardrobes and dining setups (for details, please refer to the Supplementary Material 1).
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
5
At this stage, it becomes possible to identify empirical relationships allowing for the assessment of the number of exposed
contents per building by using regression functions that relate the standardized variables for each item to the characteristics
of the building. A straightforward approach might involve correlating the number of specific items to the number of rooms
designated for them (e.g., beds to the bedrooms). However, since the number of rooms per building is often unknown in
130
flood risk assessments, it is essential to establish a relationship between the number of exposed items and more general
independent variables to enhance the practical applicability of the model. The most straightforward approach is to relate item
quantities to building size, expressed as either footprint area (FA) or surface area (SA). Footprint area refers to the total
ground area occupied by the building, while surface area is the total horizontal built-up area across all floors. For single-
story buildings, FA and SA are the same. For multi-story buildings, if each floor is occupied by separate households, the SA
135
for each unit corresponds to the individual floor area. In cases where a single household occupies multiple floors, SA
becomes more representative of the total contents, as it comprises the total floor area of the entire building (SA = NF·FA,
with NF indicating the number of floors).
For Northern Italy, a power regression function of SA was found effective in describing the number of exposed beds, sofas,
wardrobes and dining tables (Supplementary Material 1). For these elements, a stochastic component derived from a normal
140
distribution with a mean of 1 and a standard deviation of 0.2 was applied to account for the inherent variability in the
distribution of household contents. The other items were instead treated as constant functions based on practical judgment
supported by empirical observations: for instance, it is expected that each housing unit may contain only one kitchen setup.
The functions thus obtained can be then converted into step functions to more realistically represent the increments in the
number of exposed objects with changes in SA. This means that for a specific range of SA, the standardized variable related
145
to a certain content can assume only a physically sound constant value (e.g., the predicted BedLeq is expressed in increments
of 0.5 BedLeq to capture the increase in the number of beds corresponding to a single bed unit).
Once the number of exposed items (expi) is defined, it is possible to determine no.dam.itemsi through the identification of the
primary driving factors for damage induction for each content type, and the subsequent definition of the corresponding
damage. To this aim, INSYDE-content adopts a probabilistic approach based on the use of fragility functions. The model
150
assumes a binary damage state: an undamaged state (ds0) or a fully damaged state (ds1). For each content type, fragility
functions express the probability of reaching a fully damaged state, based on the event intensity measure(s) (IM). To
combine this probability with the actual occurrence of a damage state for the individual exposed elements, a random value
Pi, accounting for the survival probability of each item, is sampled from a uniform distribution between 0 and 1 and
compared to the damage probability derived from the fragility function for the corresponding content type. The random
155
nature of the implemented process serves to capture the inherent uncertainty in the damage mechanisms, reflecting the
intrinsic variability in content vulnerability to the same event intensity. Consequently, if Pi falls below the damage
probability calculated from the fragility function, it is considered fully damaged (ds1), otherwise, it remains undamaged
(ds0).
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
6
The formulation of the fragility functions is based on expert knowledge, practical experience, as well as available technical
160
and scientific documentation. Assigning the thresholds for the event feature driving the damage mechanism is not trivial and
the lack of relevant research in this field added further complexity to it. However, given the practical considerations
surrounding the damage process for contents, it is feasible to establish lower and upper bounds for the IM.
Tables 1 and 2 describe the building and event features used in the model, including their range of values and the
assumptions regarding the dependencies between the variables in case of missing information. Indeed, similar to the original
165
INSYDE for buildings, the model can automatically assign default values when certain input data are not available. This is
achieved by either leveraging the implemented relationships among the variables (e.g., for SA or hi) or by sampling from
user-defined distributions that reflect the characteristics of the region under analysis. In the version presented in this study,
the model incorporates the distributions proposed by Di Bacco et al. (2024), which are based on a combination of physically-
informed approaches and empirical survey data for northern Italy (reported for completeness in Figures S2-S3 of the
170
Supplementary Material 1).
Table 1: Building features in INSYDE-content for estimating exposed household items.
Variable
Description
Range of values
Default dependencies
NF
Number of floors [-]
> 0
FA
Footprint area [m2]
> 0
SA
Surface area [m2]
> 0
SA=f(SA, NF)
IH
Inter-floor height [m]
> 0
GL
Ground floor level [m]
≥ 0
BT
Building typology [-]
1: Detached House
2: Semi-Detached House
3: Apartment
FL
Finishing level [-]
0.8: Low
1: Medium
1.2: High
GU
Ground use [-]
1: Residential use
2: Other use (garage, storage, etc.)
GU=f(BT)
HU
Number of housing units [-]
≥ 1
HU=f(FA,BT)
Table 2: Event features in INSYDE-content.
Variable
Description
Range of values
Default dependencies
he
Inundation depth outside the building [m]
≥ 0
hi
Inundation depth inside the building (for each floor) [m]
[0:IH]
hi=f(he,GL)
d
Inundation duration [hours]
≥ 0
d=f(he)
s
Indicator for the presence of sediments [-]
0: No
1:Yes
q
Indicator for the presence of pollutants [-]
0: No
1:Yes
175
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
7
Additionally, for a more comprehensive analysis, this study also employed the non-region-specific synthetic distributions
proposed by the same authors (Figures S4-S5 in Supplementary Material 1). By covering a wider range of values, these
distributions facilitate a deeper investigation into the model’s sensitivity to input variables beyond the specific context of
northern Italy. The fragility functions and general assumptions are detailed in Supplementary Material 1, while Table 3
summarizes the variables that affect both the determination of exposed elements and the damage mechanisms in the model.
180
It is worth noting that that flow velocity does not appear among the event features considered in INSYDE-content (Table 2),
despite its potential relevance to content damage. The decision to exclude it as an input variable was guided by practical
considerations. Firstly, the impact of flow velocity on content damage should ideally be assessed based on water velocity
inside the building. However, this information is neither available in standard flood risk analyses nor are there reliable
methods to estimate it indirectly from external flow velocities (Dewals et al., 2023; Zhu et al., 2023). Furthermore, the
185
mechanism of flow intrusion into a building generally leads to a significant reduction in velocity, making its effect negligible
when assessing content damage. Conversely, when the external flow velocity is extremely high and capable of washing away
the entire building, flow velocity may become critical as it would lead to the complete destruction of both the structure and
its contents. However, this extreme scenario has not been included in the current version of INSYDE-content due to its rarity
in fluvial floods, particularly for masonry and reinforced concrete buildings. Additionally, the lack of well-established
190
criteria defining the threshold for washing away (based on a combination of water height and velocity) makes it challenging
to quantify a situation where content damage would be complete.
Table 3: Variables that affect the determination of exposed elements (EXP) and the damage mechanisms (i.e.,
fragility functions (FF)) in INSYDE-content.
Damage component
Variables affecting the component
Beds, wardrobes, dining setup
EXP: SA=f(FA,NF), BT, HU=f(FA,BT), NF, GU
FF: hi=f(he, GL), d, s, q
Sofas
EXP: SA, BT, FL, HU, NF, GU
FF: hi, d, s, q
Kitchen setup
EXP: HU, NF, GU
FF: hi, d, s, q
Washing machine, TV, oven, refrigerator
EXP: HU, NF, GU
FF: hi
Dishwasher, microwave oven
EXP: HU, FL, NF, GU
FF: hi
2.2 Model evaluation
195
2.2.1 Sensitivity analysis to the input variables
A sensitivity analysis of INSYDE-content was conducted to evaluate how the selected explanatory variables influence
damage estimation and to determine how the absence of certain input data contributes to uncertainty in these estimates. This
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
8
analysis was performed using two synthetic portfolios, each consisting of 250,000 buildings exposed to hypothetical
flooding scenarios, as described in Di Bacco et al. (2024). The first dataset represents more general, non-region-specific
200
inundation and building characteristics (referred to as the “extended dataset”, hereinafter), while the second focuses on the
specificities of northern Italy (Po River District, “Po dataset”, hereinafter). This approach enabled the identification of a
more general ranking of the variable importance within the model and allowed for subsequent observations of how this
ranking might change based on the specific characteristics of the region in which the model is implemented.
The sensitivity analysis followed these steps: first, INSYDE-content was applied to calculate damages for both building
205
portfolios, with all required model variables known. For each j-th building in the dataset, the estimated damage value in this
step was taken as reference value, D0. Then, one input variable was sequentially removed and replaced, for each building in
the datasets, with values sampled from the distributions given by Di Bacco et al. (2024). This process was repeated for each
variable, with damage recalculated each time (Di). The absolute difference in damage compared to the reference value was
recorded for each of the j buildings (| D0 – Di |j), facilitating the determination of the variance each feature contributes to the
210
model’s outcome.
2.2.2 Model validation
The model was validated using loss data from two historical flood events that occurred in Northern Italy: the 2002 flood in
Lodi and the 2010 flood in Caldogno. These events have been analyzed in previous studies regarding building damage
(Amadio et al., 2019; Molinari et al., 2020), but never for household contents. In this study, only buildings without
215
basements were considered, coherently with the model’s assumption that does not account for their presence. The validation
dataset includes 194 buildings for the Lodi event and 169 buildings for the Caldogno event, with total actualized losses to
year 2023 amounting to about 3.1 million euro for both cases.
The loss data were derived from the forms for damage quantification distributed by the authorities as part of the state’s loss
compensation process, which were filled in by affected citizens. These forms provided actual restoration costs, certified by
220
original receipts and invoices. The types of content damage eligible for compensation aligned with the 11 components
identified in the model formulation. The unit prices for content items were taken from the decree issued by the delegated
commissioner responsible for damage compensation following the Caldogno flood. These prices were then applied to the
Lodi case after being adjusted for inflation to the event date.
In addition to loss data, the dataset includes information on external water depth at building’s location. As regards the other
225
event features required by INSYDE-content, qualitative data from previous studies allowed to determinate an approximate
value for inundation duration equal to one day for both cases (Di Bacco et al., 2024). Accordingly, the sampling of d values
was obtained from a truncated normal distribution centred at 24 hours and spanning between 16 and 48 hours. The
information on the presence of fine-graded sediments allowed for assigning s=1 (yes) for both cases, while local data on
pollutants was available only for Lodi.
230
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
9
Concerning building features, the dataset includes footprint area (FA), number of floors (NF), building type (BT) and
finishing level (FL). For the remaining missing variables, the model was applied by leveraging the established relationships
among the variables and the built-in sampling process from pre-defined distributions, as outlined by Di Bacco et al. (2024).
To account for uncertainty arising from the sampling process of unknown inputs and from the implemented assumptions on
exposure and damage mechanisms, content losses were calculated probabilistically for each building over 1,000 iterations,
235
resulting in confidence intervals for the estimated values which were compared to reported losses.
3 Results and discussion
3.1 Examples of resulting damage functions
Figure 1 illustrates an example of traditional damage functions resulting from the application of INSYDE-content, broken
down by components and in terms of total damage. The functions correspond to a scenario involving both apartment and
240
single-family, detached or semi-detached residential buildings with a surface area of 200 m2, distributed across two floors,
with IH=3 m, GL=0 m and a high finishing level. In the total damage functions, the interquartile range is shown alongside
the median damage values to represent the variability induced by the probabilistic modeling of the damage mechanism. For
each water depth, damage values were calculated over 1,000 iterations by randomly assigning q as either 0 or 1 and
generating random samples of inundation durations based on water depth scenarios, with d assumed to be relatively short for
245
shallow inundations and extending from 6 hours to 2 days for depths greater than 2 meters. The trends shown in Figure 1
effectively highlight the assumptions on the damage mechanisms implemented into the model for the various items, as
described in the Supplementary Material 1. For instance, focusing on detached and semi-detached buildings, damage to beds
only occurs when the inundation depth exceeds 3 meters, consistent with the assumption that bedrooms are located on the
upper floors for this type of dwelling. Similarly, sofas, assumed to be in the downstairs living area, appear to be highly
250
vulnerable to flood, necessitating a full replacement even under very shallow water depths. The pattern for wardrobes and
their contents is more complex, as they include different types of furniture, ranging from small and decorative pieces in the
living room to full-height wardrobes in bedrooms, each with varying thresholds for damage onset. This results in the stepped
function shown in Figure 1, panel c) similar to the one for kitchens, where the steps represent the damage occurring to lower
and upper cabinets. Electrical components display a similar behavior across all items, with only different activation
255
thresholds based on their varying vulnerability. For apartments, the overall damage is higher compared to single-family
buildings with the same surface area, due to the presence of multiple housing units within the same space, each containing its
own furniture and appliances. To ensure clarity in Figure 1, panel b) focuses exclusively on the damage functions for
electrical components and kitchen setup in a single-family building, which correspond to those of a single housing unit of an
apartment building. For the other components, distinct patterns emerge between the two building types, reflecting different
260
assumptions regarding the distribution of contents across the floors. For example, in the case of apartments, a non-null
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
10
damage to beds can be recognized even on the first floor, with a stepped increase in damage between floors, as observed for
sofas.
Figure 1: Example of traditional damage functions resulting from the application of INSYDE-content for different
265
building types: a) total content damage; b) damage to kitchen and damage setup as well as electric appliances; c)
damage to beds, sofas and wardrobes to a detached/semi-detached building; d) as panel c), but for an apartment
building.
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
11
3.2 Sensitivity analysis to the input variables
270
This section reports on the sensitivity of damage estimation to the variables considered in INSYDE-content and on the
uncertainty arising from potential missing input data in model implementation. Figure 2 summarizes the findings by
illustrating the difference in computed damage when the model is applied to the reference portfolios of 250,000 buildings
and to their replicas obtained by replacing the values of one input variable at a time with a sampling from the predefined
distributions implemented in the model. The left and right panels of Figure 2 correspond, respectively, to the extended
275
synthetic dataset, which encompasses a broader range of values for the input features, and the Po dataset, tailored to the
characteristics of northern Italy. In Figure 2 features are ranked by the median value of | D0 – Di |j, where D0 represents the
reference damage for the j-th building and Di the corresponding value calculated after removing and resampling the i-th
variable from the predefined distributions developed by Di Bacco et al. (2024).
Dependent variables, such as SA or hi, are not shown in the figure as their effects are inherently accounted for through the
280
independent variables with which they are assumed to be linked. Similarly, the influence of sediments, represented by the
binary variable s, is not reflected in the figure; this is because the distributions generated by Di Bacco et al. (2024) for
representing the fraction of sediments on water volume always yield a non-null value in the analysis, which implies a
constant s=1 (i.e., presence of sediment, of any gradation) in INSYDE-content.
Overall, Figure 2 demonstrates a feature importance pattern that remains relatively consistent across the two tested datasets,
285
with the most influential variables (he, BT, FA, and d) consistently appearing among the top five in both cases. As expected,
higher values of | D0 – Di |j are observed in the extended dataset, where missing data sampling (especially for extensive and
intensive variables) can span a broader range of variability than in the Po scenario (Di Bacco et al., 2024).
Figure 2: Variable importance in INSYDE-content. Results obtained with sampling from the two different datasets
290
developed by Di Bacco et al. (2024): a) case for the extended synthetic dataset; b) case for the Po River District
synthetic dataset. Variables are ordered according to the median value of the absolute damage difference.
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
12
Inundation depth (he) emerges as the primary contributor to uncertainty in damage estimation, with median absolute damage
differences ranging from approximately 4,500 to 6,800 euro, confirming the well-known significance of such variable in
295
direct flood damage assessment for various types of assets, including household contents (Merz et al., 2013; Schröter et al.,
2014).
Building type (BT) plays a crucial role in determining the exposed value of contents by shaping the distribution of items
within the building (e.g., the assumption regarding bed placement in single- and multi-family buildings discussed earlier).
This effect is evident in the median | D0 – Di |j shown in Figure 2, which range from approximately 3,000 to 3,500 euro,
300
establishing BT as the second most influential feature in INSYDE-content, in contrast to its minimal impact on building
damage mechanisms (Dottori et al., 2016; Di Bacco et al., 2024). The minor influence of ground use (GU) is partly masked
by the effect of BT, as the model differentiates the use of the ground floor only in the case of apartment buildings, where it is
more common for the ground floor to be allocated for non-residential purposes (e.g., garages, communal areas, etc.).
Closely related to inundation depth is the role of ground level (GL), which influences the water depth inside the building and
305
subsequently affects damage. In the extended dataset, GL ranks as the third most important feature, with median absolute
damage differences of about 2,900 euro. On the other hand, in the Po dataset, the lower variability of GL makes its impact
less significant, resulting in an estimated damage variation of about 1,500 euro, which is comparable to the variations
observed for other variables such as FL, GU, and q.
Regarding the number of floors (NF), while the quantitative effects on damage calculations are similar in both datasets, with
310
a median | D0 – Di |j of approximately 1,700 – 2,000 euro, greater variability is observed in the extended dataset (with an
interquartile range of 5,400 euro, compared to 3,100 euro for the Po case), as a consequence of the higher probability of
damage affecting upper floors, given the broader range of inundation depths represented in this dataset.
In the Po dataset, a relatively stronger influence is observed for the two variables affecting the number of exposed contents -
either directly through footprint area (FA) or indirectly via finishing level (FL) (see Supplementary Material 1 for the
315
corresponding functions) - with these variables occupying the third and fifth positions, respectively. The relatively limited
impact of FL, as shown in Figure 2, can be attributed to the fact that the model considers its effect only for the estimation of
the number of exposed contents, without accounting for potential increases in their unit price. This modeling choice was
guided by practical considerations: first, for household contents, it is virtually impossible to establish a clear upper limit on
the economic value of an item as its luxury level increases; second, this approach aligns with the application of the model in
320
flood risk management contexts (e.g., under the European Floods Directive) or in government compensation schemes for
flood damage, where standard average costs are used, regardless of the actual quality of the damaged items.
3.3 Insights into content-to-building damage relationship
This section expands on the model’s performance by analyzing the relationship between content and building damage,
aligning with previous studies that aimed to establish connections between the two (Thieken et al., 2005; Carisi et al., 2018;
325
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
13
Mosimann et al., 2018). To this end, building damage was calculated using the INSYDE 2.0 model (Di Bacco et al., 2024)
and paired with content damage estimates for the two synthetic datasets used in the sensitivity analysis.
Figure 3a illustrates the results, plotting estimated content damage against building damage for both datasets and comparing
them with the equation proposed by Carisi et al. (2018), based on post-event observations from the 2014 Secchia flood in
Emilia Romagna, Italy.
330
INSYDE results reveal substantial variability in content damage for same levels of building damage, highlighting the
complex, multi-variable nature of damage mechanisms that cannot be fully represented by simple univariate functions.
Interestingly, the root function proposed by Carisi et al. (2018) aligns at a median level with INSYDE results, although in a
region with lower sample density. Conversely, INSYDE allows discerning distinct patterns in the relationship between
building and content damage, shaped by the combined effects of inundation duration thresholds and building characteristics
335
(such as BT and FL) triggering specific damage mechanisms to certain components, as previously described for INSYDE 2.0
by Di Bacco et al. (2024).
Furthermore, Figure 3b illustrates the content-to-building damage ratio (CBR) as a function of inundation depth, offering
insights into the limitations of using a univariate approach based solely on building damage, especially in case of shallow
inundation depths (< 0.5 m), where the higher vulnerability of contents can lead to CBR values exceeding 5. However, as
340
inundation depth increases, overall damage becomes predominantly driven by building damage, leading to a reduction in the
differences between the two approaches.
Median CBR values for INSYDE across the two datasets range from 0.26 (calculated on inundation durations exceeding 48
hours) to 0.36 and 0.42, respectively for the extended and Po River datasets at shorter durations.
345
Figure 3: a) Content to building damage calculated with INSYDE models on the two tested datasets and comparison
with the root function of Carisi et al. (2018); b) Content to building damage ratio (CBR) expressed as a function of
inundation depth with INSYDE models and Carisi et al. (2018).
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
14
These findings are consistent with post-event observations in Switzerland and Germany (Thieken et al., 2005; Mosimann et
350
al., 2018), which reported CBR values around 0.28-0.29, with higher values for lower-entity damages, as illustrated here in
Figure 3b. Nonetheless, the results presented in Figure 3 highlight the importance of detailed knowledge regarding the
vulnerability and exposure characteristics of potentially impacted assets to achieve accurate content damage estimations with
minimized uncertainty. The probabilistic sampling approach implemented in INSYDE-content addresses the practical
challenges of obtaining such detailed information in large-scale applications by effectively managing unknown missing data
355
through calibrated distributions of the input features, while also explicitly accounting for uncertainties associated with
predictions and thereby mitigating the false sense of certainty that, instead, often accompanies deterministic models.
3.4 Model validation
The validation outcomes are summarized in Table 4, presenting the statistics of damage estimates from INSYDE-content,
based on 1,000 replicates for each building with missing input features, alongside observed damage values. Figure 4
360
complements these findings with a detailed visualization of the differences between individual building-level estimations and
actual observations.
Table 4 indicates a good alignment between the total estimated damages and observed values. In Lodi the median of
calculated losses aligns closely with reported values, while for Caldogno reported losses approach the third quartile of
estimates. Beyond the overall convergence, Figure 4 highlights notable scatter at the individual building level, especially for
365
apartment buildings, where the absence of ground-level usage information (GU) introduces large variability in the estimates,
as the model randomly samples residential or non-residential use over the iterations, impacting both the exposed contents
and expected damage. However, it is worth noting that such building-level differences between estimates and observations
are a common outcome in flood damage validation exercises, especially for lower-magnitude damages, where models often
tend toward overestimation, as also described in previous studies on building damage (Merz et al., 2008; Molinari et al.,
370
2020; Pinelli et al., 2020). From another perspective, this discrepancy may be also attributable to poor representativeness of
claim data, particularly for lower entity losses, as highlighted by several authors (Molinari et al., 2020; Pinelli et al., 2020;
Wing et al., 2020; Museru et al., 2024). Such findings are even more expected for contents compared to buildings, given that
they exhibit much greater variability and less standardization than structural or non-structural building components.
Table 4: Results of the probabilistic validation of INSYDE-content for the case studies of Caldogno and Lodi:
375
statistics of total estimated damage versus reported damage.
Estimated damage [M€ 2023]
Observed damage [M€ 2023]
1st quartile
Median
3rd quartile
Caldogno
1.10
1.76
3.50
3.15
Lodi
1.75
3.30
6.87
3.06
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
15
Figure 4. Results of the probabilistic validation of INSYDE-content: a) Caldogno event; b) Lodi event. Median
computed damage (dot) and corresponding interquartile range (line) are plotted for each building against observed
380
damage (expressed in 2023 euro).
In addition to confirming INSYDE-content’s robustness in estimating total damages across the two events, this analysis
highlights the importance of incorporating input data uncertainty when validating and evaluating model performance.
Indeed, presenting results with well-defined uncertainty bounds not only increases model transparency and reliability, but
385
also prevents the potential for a misleading impression of certainty, a limitation often found in simpler deterministic models
(Pappenberger and Beven, 2006; Merz et al., 2015).
4 Conclusions
Modeling flood damage to household contents poses significant challenges due to the inherent heterogeneity of such items,
which contrasts to the more standardized geometry and features of buildings. This complexity is further exacerbated by the
390
need to account for a wide range of factors influencing damage mechanisms (including flood characteristics, the
vulnerability of both buildings and their contents, as well as broader economic aspects), demanding a more comprehensive
approach than traditional univariate damage assessments.
This study aimed to address such complexities by introducing INSYDE-content, a probabilistic, multi-variable flood damage
model, specifically designed for household contents. The model was developed through a structured, multi-phase process
395
including: (i) an extensive data collection on household contents and flood characteristics; (i) the probabilistic estimation of
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
16
the presence of household contents to reflect their variability and distribution across different building types; (iii) the
development of fragility functions for individual items based on their driving factors; and (iv) a price analysis to obtain
monetary damage estimations. Although the model presented in the paper has been tailored to the context of Northern Italy
(Po river district), its components and basic input data can be adapted or adjusted to the specific characteristics of other
400
regions, ensuring its generalizability.
Sensitivity analysis and validation exercises provided insights into the model’s performance. One of the key findings
emerging from this study is the critical importance of accounting for the multi-variable nature of flood damage to household
contents, with the primary factors being inundation depth (and, to a lesser extent, inundation duration) and the variables that
quantify and distribute exposed items within buildings, such as those representing building size, type and use. Additionally,
405
the study confirms that in cases of shallow inundation, content damage can exceed building damage, with content-to-
building damage ratios largely exceeding 1, thus highlighting the often-overlooked vulnerability of household contents in
flood risk assessments.
The probabilistic nature of INSYDE-content effectively addresses uncertainties arising from potential missing input data,
offering a more robust alternative to traditional deterministic models. By employing sampling techniques of input features,
410
the model can estimate damage even in the absence of certain data, providing information on estimation uncertainty while
maintaining transparency.
Additionally, the validation results confirmed the reliability of INSYDE-content, with its damage estimates aligning with
observed claim data from two flood events in Italy. From another perspective, the analysis performed in this study also
highlighted the importance of incorporating input data uncertainty when validating and evaluating model performance.
415
Presenting damage estimations with clearly defined uncertainty bounds not only enhance model transparency and reliability,
but also helps preventing a false sense of certainty, which is typical of simpler deterministic models. On the other hand,
comparing observed data within a plausible range of estimates allows to properly consider also the uncertainty that is
inherent in claim data.
By offering an integrated approach that combines the complex interactions between flood characteristics, building features
420
and household contents, while also addressing the challenges posed by data variability and uncertainty, INSYDE-content
then provides a robust and reliable tool for supporting more comprehensive flood risk assessments. Future work should focus
on expanding and customizing the input datasets for household contents across different geographical contexts, refining and
validating component fragility functions for a wider range of flood scenarios, and exploring the role of additional socio-
economic factors in shaping damage patterns.
425
Code availability
During the review process, the code of INSYDE-content is available at the following link:
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
17
https://drive.google.com/file/d/1T8XPF5UPALxJ0JGUYj2rodl7TStRP9Pz/view?usp=sharing and it will be made available
on Mendeley data upon final acceptance.
Author contribution
430
Pradeep Acharya: Data curation; Formal analysis; Investigation; Software; Visualization; Writing - review and editing.
Mario Di Bacco: Conceptualization; Methodology; Data curation; Investigation; Software; Writing - review and editing.
Daniela Molinari: Conceptualization; Investigation; Writing - review and editing.
Anna Rita Scorzini: Conceptualization; Methodology; Investigation; Visualization; Writing - original draft.
Competing interests
435
The authors declare that they have no conflict of interest.
Financial support
This study was partly carried out within the RETURN Extended Partnership and received funding from the European Union
Next-GenerationEU (National Recovery and Resilience Plan - NRRP, Mission 4, Component 2, Investment 1.3 - D.D. 1243
2/8/2022, PE0000005).
440
References
Ahadzie, D.K., Mensah, H., and Simpeh, E.: Impact of floods, recovery, and repairs of residential structures in Ghana:
insights from homeowners. GeoJournal, 87(4), 3133-3148, doi: 10.1007/s10708-021-10425-2, 2022.
Amadio, M., Scorzini, A.R., Carisi, F., Essenfelder, A.H., Domeneghetti, A., Mysiak, J., and Castellarin, A.: Testing
empirical and synthetic flood damage models: the case of Italy. Nat. Hazards Earth Syst. Sci., 19(3), 661-678, doi:
445
10.5194/nhess-19-661-2019, 2019.
Brázdil, R., Kundzewicz, Z.W., and Benito, G.: Historical hydrology for studying flood risk in Europe. Hydrol. Sci. J., 51(5),
739-764, doi: 10.1623/hysj.51.5.739, 2006.
Carisi, F., Schröter, K., Domeneghetti, A., Kreibich, H., and Castellarin, A.: Development and assessment of uni-and
multivariable flood loss models for Emilia-Romagna (Italy). Nat. Hazards Earth Syst. Sci., 18(7), 2057-2079, doi:
450
10.5194/nhess-18-2057-2018, 2018.
Davis, S.A. and Skaggs, L.L. Catalog of residential depth-damage functions used by the Army Corps of Engineers in flood
damage estimations. IWR Report 92-R-3, 140p. US Army Corps of Engineers, Water Resources Support Center, Institute for
Water Resources, Ft. Belvoir, Virginia, USA, 1992.
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
18
Dewals, B., Kitsikoudis, V., Mejía-Morales, M.A., Archambeau, P., Mignot, E., Proust, S., Erpicum, S., Pirotton, M., and
455
Paquier, A.: Can the 2D shallow water equations model flow intrusion into buildings during urban floods? J. Hydrol., 619,
129231, doi: 10.1016/j.jhydrol.2023.129231, 2023.
Di Bacco, M., Molinari, D., and Scorzini, A.R.: The value of multi-source data for improved flood damage modelling with
explicit input data uncertainty treatment: INSYDE 2.0. Nat. Hazards Earth Syst. Sci., 24(5), 1681-1696, doi: 10.5194/nhess-
24-1681-2024, 2024.
460
Disse, M., Johnson, T.G., Leandro, J., and Hartmann, T.: Exploring the relation between flood risk management and flood
resilience. Water Secur., 9, 100059, doi: 10.1016/j.wasec.2020.100059, 2020.
Dottori, F., Figueiredo, R., Martina, M., Molinari, D., and Scorzini, A.R.: INSYDE: a synthetic, probabilistic flood damage
model based on explicit cost analysis. Nat. Hazards Earth Syst. Sci., 16, 2577-2591, doi:10.5194/nhess-16-2577-2016, 2016.
Endendijk, T., Botzen, W.W., de Moel, H., Aerts, J.C., Slager, K., and Kok, M.: Flood vulnerability models and household
465
flood damage mitigation measures: An econometric analysis of survey data. Water Resour. Res., 59(8), e2022WR034192,
doi: 10.1029/2022WR034192, 2023.
Gerl, T., Kreibich, H., Franco, G., Marechal, D., and Schröter, K.: A review of flood loss models as basis for harmonization
and benchmarking. PloS One, 11(7), e0159791, doi: 10.1371/journal.pone.0159791, 2016.
Gralepois, M., Larrue, C., Wiering, M., Crabbé, A., Tapsell, S., Mees, H., Ek, K., and Szwed, M.: Is flood defense changing
470
in nature? Shifts in the flood defense strategy in six European countries. Ecol. Soc., 21(4), doi: 10.5751/ES-08907-210437,
2016.
Hirabayashi, Y., Mahendran, R., Koirala, S., Konoshima, L., Yamazaki, D., Watanabe, S., Kim, H., and Kanae, S.: Global
flood risk under climate change. Nat. Clim. Change, 3(9), 816-821. doi: 10.1038/nclimate1911, 2013.
IPCC: Summary for Policymakers. In: Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III
475
to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J.
Romero (eds.)]. IPCC, Geneva, Switzerland, pp. 1-34, doi: 10.59327/IPCC/AR6-9789291691647.001, 2023.
Jongman, B., Kreibich, H., Apel, H., Barredo, J.I., Bates, P.D., Feyen, L., Gericke, A., Neal, J., Aerts, J.C.J.H., and Ward,
P.J.: Comparative flood damage model assessment: towards a European approach. Nat. Hazards Earth Syst. Sci., 12(12),
3733-3752, doi: 10.5194/nhess-12-3733-2012, 2012.
480
Marín-García, D., Rubio-Gómez-Torga, J., Duarte-Pinheiro, M., and Moyano, J.: Simplified automatic prediction of the
level of damage to similar buildings affected by river flood in a specific area. Sustainable Cities Soc., 88, 104251, doi:
10.1016/j.scs.2022.104251, 2023.
Martínez-Gomariz, E., Forero-Ortiz, E., Russo, B., Locatelli, L., Guerrero-Hidalga, M., Yubero, D., and Castan, S.: A novel
expert opinion-based approach to compute estimations of flood damage to property in dense urban environments. Barcelona
485
case study. J. Hydrol., 598, 126244, doi: 10.1016/j.jhydrol.2021.126244, 2021.
Merz, B., Kreibich, H., and Apel, H.: Flood risk analysis: uncertainties and validation, Österreichische Wasser-und
Abfallwirtschaft, 60, 89-94, doi: 10.1007/s00506-008-0001-4, 2008.
Merz, B., Kreibich, H., Schwarze, R. and, Thieken, A.: Review article: “Assessment of economic flood damage”. Nat.
Hazards Earth Syst. Sci., 10(8), 1697-1724, doi: 10.5194/nhess-10-1697-2010, 2010.
490
Merz, B., Kreibich, H., and Lall, U.: Multi-variate flood damage assessment: a tree-based data-mining approach. Nat.
Hazards Earth Syst. Sci., 13(1), 53-64, doi: 10.5194/nhess-13-53-2013, 2013.
Merz, B., Vorogushyn, S., Lall, U., Viglione, A., and Blöschl, G.: Charting unknown waters - On the role of surprise in flood
risk assessment and management, Water Resour. Res., 51(8), 6399-6416, doi: 10.1002/2015WR017464, 2015.
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
19
Molinari, D., Scorzini, A.R., Arrighi, C., Carisi, F., Castelli, F., Domeneghetti, A., [...], and Ballio, F.: Are flood damage
495
models converging to “reality”? Lessons learnt from a blind test. Nat. Hazards Earth Syst. Sci., 20(11), 2997-3017, doi:
10.5194/nhess-20-2997-2020, 2020.
Mosimann, M., Frossard, L., Keiler, M., Weingartner, R., and Zischg, A.P.: A robust and transferable model for the
prediction of flood losses on household contents. Water, 10(11), 1596, doi: 10.3390/w10111596, 2018.
Museru, M.L., Nazari, R., Giglou, A. N., Opare, K., and Karimi, M.: Advancing flood damage modeling for coastal Alabama
500
residential properties: a multivariable machine learning approach. Sci. Total Environ., 907, 167872, doi:
10.1016/j.scitotenv.2023.167872, 2024.
Nofal, O.M., van de Lindt, J.W., and Do, T.Q.: Multi-variate and single-variable flood fragility and loss approaches for
buildings. Reliab. Eng. Syst. Saf., 202, 106971, doi: 10.1016/j.ress.2020.106971, 2020.
Pappenberger, F. and Beven, K.J. Ignorance is bliss: Or seven reasons not to use uncertainty analysis, Water Resour. Res.,
505
42(5), W05302, doi:10.1029/2005WR004820, 2006.
Paulik, R., Zorn, C., and Wotherspoon, L.: Evaluating the spatial application of multivariable flood damage models. J. Flood
Risk Manage., 16(4), e12934, doi: 10.1111/jfr3.12934, 2023.
Pinelli, J.P., Da Cruz, J., Gurley, K., Paleo-Torres, A.S., Baradaranshoraka, M., Cocke, S., and Shin, D.: Uncertainty
reduction through data management in the development, validation, calibration, and operation of a hurricane vulnerability
510
model, Int. J. Disaster Risk Sci., 11, 790-806, doi: 10.1007/s13753-020-00316-4, 2020.
Plate, E.J.: Flood risk and flood management. J. Hydrol., 267(1-2), 2-11, doi: 10.1016/S0022-1694(02)00135-X, 2002.
Romali, N.S. and Yusop, Z. Establishment of residential flood damage function model for Kuantan, Malaysia. International
Journal of GEOMATE, 19(71), 21-27, doi: /10.21660/2020.71.9144, 2020.
Sayers, P.B., Hall, J.W., and Meadowcroft, I.C.: Towards risk-based flood hazard management in the UK. Proceedings of
515
the Institution of Civil Engineers - Civil Engineering, 150 (5), 36-42, doi: 10.1680/cien.2002.150.5.36, 2002.
Schröter, K., Lüdtke, S., Redweik, R., Meier, J., Bochow, M., Ross, L., Nagel, C., and Kreibich, H.: Flood loss estimation
using 3D city models and remote sensing data. Environ. Modell. Software, 105, 118-131, doi:
10.1016/j.envsoft.2018.03.032, 2018.
Wagenaar, D., Lüdtke, S., Schröter, K., Bouwer, L.M., and Kreibich, H.: Regional and temporal transferability of
520
multivariable flood damage models. Water Resour. Res., 54(5), 3688-3703, doi: 10.1029/2017WR022233, 2018.
Scorzini, A.R., Dewals, B., Rodriguez Castro, D., Archambeau, P., and Molinari, D.: INSYDE-BE: adaptation of the
INSYDE model to the Walloon region (Belgium). Nat. Hazards Earth Syst. Sci., 22(5), 1743-1761, doi: 10.5194/nhess-22-
1743-2022, 2022.
Scorzini, A. R., Di Bacco, M., and Manella, G.: Regional flood risk analysis for agricultural crops: Insights from the
525
implementation of AGRIDE-c in central Italy. Int. J. Disaster Risk Reduct., 53, 101999, doi: 10.1016/j.ijdrr.2020.101999,
2021.
Shrestha, B.B., Kawasaki, A., and Zin, W.W.: Development of flood damage assessment method for residential areas
considering various house types for Bago Region of Myanmar. Int. J. Disaster Risk Reduct., 66, 102602, doi:
10.1016/j.ijdrr.2021.102602, 2021.
530
Tanoue, M., Hirabayashi, Y., and Ikeuchi, H.: Global-scale river flood vulnerability in the last 50 years. Sci. Rep., 6(1),
36021, doi: 10.1038/srep36021, 2016.
Thieken, A. H., Müller, M., Kreibich, H., and Merz, B.: Flood damage and influencing factors: New insights from the
August 2002 flood in Germany. Water Resour. Res., 41(12), W12430, doi: 10.1029/2005WR004177, 2005.
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
20
Wahab, R. and Tiong, R.: Multi-variate residential flood loss estimation model for Jakarta: an approach based on a
535
combination of statistical techniques. Nat. Hazard., 86, 779-804, doi: 10.1007/s11069-016-2716-z, 2017.
Wing, O.E., Pinter, N., Bates, P.D., and Kousky, C.: New insights into US flood vulnerability revealed from flood insurance
big data. Nat. Commun., 11(1), 1444, doi: 10.1038/s41467-020-15264-2, 2020.
Xing, Z., Yang, S., Zan, X., Dong, X., Yao, Y., Liu, Z., and Zhang, X.: Flood vulnerability assessment of urban buildings
based on integrating high-resolution remote sensing and street view images. Sustainable Cities Soc., 92, 104467, doi:
540
10.1016/j.scs.2023.104467, 2023.
Zhai, G., Fukuzono, T., and Ikeda, S.: Modeling flood damage: Case of Tokai Flood 2000. JAWRA J. Am. Water Resour.
Assoc., 41(1), 77-92, doi: 10.1111/j.1752-1688.2005.tb03719.x, 2005.
Zhu, Z., Gou, L., Liu, S., and Peng, D.: Effect of urban neighborhood layout on the flood intrusion rate of residential
buildings and associated risk for pedestrians. Sustainable Cities Soc., 92, 104485, doi: 10.1016/j.scs.2023.104485, 2023.
545
https://doi.org/10.5194/egusphere-2025-1413
Preprint. Discussion started: 12 May 2025
c
Author(s) 2025. CC BY 4.0 License.
