Enhancement of large-scale flood damage assessments using building-material-based vulnerability curves for an object-based approach

In this study, we developed an enhanced approach for large-scale flood damage and risk assessments that uses 10 characteristics of buildings and the built environment as object-based information to represent exposure and vulnerability to flooding. Most current large-scale assessments use an aggregated land-use category to represent the exposure, treating all exposed elements the same. For large areas where previously only coarse information existed such as in Africa, more detailed exposure data is becoming available. For our approach, a direct relation between the construction type and building material of the exposed elements is used to develop vulnerability curves. We further present a method to differentiate flood 15 risk in urban and rural areas based on characteristics of the built environment. We applied the model to Ethiopia, and found that rural flood risk accounts for about 22% of simulated damages; rural damages are generally neglected in the typical landuse-based damage models particularly at this scale. Our approach is particularly interesting for studies in areas where there is a large variation in construction types in the building stock, such as developing countries. It also enables comparison across different natural hazard types that also use material-based vulnerability, paving the way to the enhancement of multi-risk 20 assessments.


Introduction
Globally, floods are one of the main natural hazards in terms of socioeconomic impacts, causing billions of dollars of damage each year.For example, between 1980 and 2013, global flood damages exceeded $1 trillion, and resulted in ca.220,000 fatalities (Dottori et al., 2016).Reducing disaster risk, such as from flooding, is globally very high on the political agenda.For example, it is an important aspect of both the Sendai Framework for Disaster Risk Reduction (UNISDR, 2015) and the Warsaw International Mechanism for Loss and Damage Associated with Climate Change Impacts (UNFCCC, 2013).
To achieve this reduction in risk at the global scale requires methods to quantitatively assess global flood risk (Mechler et al., 2014).Here, flood risk is defined as a function of three components: hazard (e.g.flood extent and depth), exposure (assets Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.and people exposed), and vulnerability (factors that determine the susceptibility of the exposed assets to the hazard) (UNISDR, 2015).
Global flood risk assessments are increasingly used in decision-making and practice, and have been useful for identifying flood risk hotspots (e.g. Ward et al., 2015).In an ideal situation, such flood risk assessment models could use detailed, highresolution data for all locations around the globe (Jonkman, 2013).In practice, data and resources required for such models rarely exist, and therefore global flood risk models have been developed.Current global flood risk models often use resolutions between 30" x 30" and 0.5° x 0.5° to assess the exposed assets (e.g.Alfieri et al., 2013;Arnell and Gosling, 2016;Ward et al., 2013).Recently, much effort has been put into improving global risk models, mainly by improving the hazard component (e.g.Dottori et al., 2016;Ikeuchi et al., 2017;Sampson et al., 2015;e.g. Trigg et al., 2016).However, much less attention has been given to improvements in the representation of exposure and vulnerability, despite the fact that their overall contribution to uncertainty is large (de Moel and Aerts, 2010).
In large-scale assessments, i.e. regional to global levels, exposure is generally represented based on aggregated land-use categories, especially in regions where only limited data are available, such as Africa (de Moel et al., 2015).Whilst using such data provides a useful first assessment of large-scale damages and risk (e.g.Feyen et al., 2011;Hall et al., 2005;Ward et al., 2013), more detailed information of the exposed objects could improve these assessments.Vulnerability is mostly represented using stage-damage functions, also known as vulnerability curves, which describe the relationship between potential damages of the exposed elements for different levels of the hazard (usually water depth).For large-scale studies, a vulnerability curve is generally developed for each of the aggregated land-use categories used to represent exposure (Ward et al., 2013).
Whilst aggregated land-use categories may be a suitable option to represent exposure if data are limited, they cannot reflect the (spatial) heterogeneity within each land-use category (Wünsch et al., 2009).For instance, large-scale flood risk models usually focus on an 'urban' category that aggregates exposed elements of various types (e.g.houses, infrastructure, shops, green areas etc.) into one land-use class (Ward et al., 2015).Since an aggregated land-use category like 'urban' is coupled to one 'urban' vulnerability curve, these curves generalise the relationship between flood depth and damage across all of the diverse exposed element types within that category.Without a more direct relation between these types of exposed elements and the impact of flood waters, large uncertainties exist in the simulated damages (de Moel and Aerts, 2010).More detailed information on the specific land use, its extent, and the vulnerability of the exposed elements could improve large-scale assessments, for example by using high-resolution remote sensing products (Goldblatt et al., 2018;Myint et al., 2011) or information as used in local scale flood damages studies at an object level (individual buildings, businesses, infrastructure objects, etc.) (de Moel et al., 2015;Merz et al., 2010).In our approach, we therefor utilize information about the composition of an area's building stock and the characteristics of exposed objects, particularly construction types and materials.Applying these object-based information, which is not to be confused with object based image analysis in remote sensing, is contrasting to the common land-use-based approach in large scale flood risk assessments.The literature distinguishes flood vulnerability of buildings according to different structural factors (such as building type, quality, height, and material), as well as occupancy type (such as residential, commercial, industrial, etc.).The latter is a commonly used factor for determining the vulnerability (de Ruiter et al., 2017), with much fewer studies relating potential losses to the structural factors.Reasons for this are the paucity of information and the huge effort it takes to obtain information on the damage incurred by individual objects and the structural components (Wahab and Tiong, 2016).Studies or models that do include information on these factors are mostly based on surveys and were therefore only feasible on smaller scales.FLEMOps (Thieken et al., 2008) is an example of a model that uses survey data on flood damages in Germany, and includes factors such as building type and quality.The study by de Villiers et al. (2007) is one of the few assessments (see also World Bank, 2000) within Africa, but uses size and content value of houses to determine flood damage and does not go into detail on structural features.Studies that focus on construction type and building material to assess the flood damage show that these characteristics, together with ground floor elevation and number of floors, are important features in determining the vulnerability of different building types to floods (e.g.Godfrey et al., 2015;Neubert et al., 2008;Schwarz and Maiwald, 2008;Zhai et al., 2005).
Compared to risk assessments in the earthquake domain where they are essential components (de Ruiter et al., 2017), or in local-scale studies focusing on physical vulnerability to debris flows (Papathoma-Köhle et al., 2017), construction types and building materials have only played a minor role as indicators for flood vulnerability.Large-scale flood risk assessments could be improved by using object-based characteristics to represent exposure and vulnerability, particularly in developing countries with a diverse structural building stock.
Recently, a building exposure dataset has been developed for several African countries as part of the Building Disaster Resilience program for the World Bank's Africa Disaster Risk Financing Initiative by ImageCat (ImageCat et al., 2017).
ImageCat uses a stratified sampling technique that infers the number of buildings in a region from census data and then uses image processing tools to identify development patterns (Hu et al., 2014).The construction practices are then characterized through a review of the literature, interviews, review of VHR images, in situ video, and in some cases site visits (Silva et al., 2018).This characterization of development patterns is used for dasymetric mapping of building counts to a 15" grid.
Estimates are supplemented with total estimates of floor area, and replacement values based on construction practices observed in each development pattern (Huyck and Eguchi, 2017).Compared to the methods employed in current large-scale flood risk models, the information about the built environment of an area and its characteristics as provided in such datasets enables a differentiation between the exposed objects in terms of vulnerability to flood waters and exposed value.
Furthermore, a greater level of detail opens up the possibility to address the issue of distinguishing urban and rural flood risk.This is commonly neglected in land-use-based flood risk assessment, due to the focus on higher value urban damages.
Moreover, land-use classification studies have difficulties in assessing urban and rural differences.This is because the resolution in previous land-use and land-cover products was not sufficient to identify smaller settlements, and the characteristics of urban and rural areas are very different and can be difficult to grasp in land-use classification studies (Dijkstra and Poelman, 2014).Internationally there is no agreed way to distinguish urban from rural areas.For example, according to the national census in Ethiopia, localities of 2,000 or more inhabitants are considered urban, whereas the urban definition for Niger only includes capitals of departments and districts (UNSD, 2016).Another traditional distinction is that urban areas provide a different way of life and usually a higher living standard (UNSD, 2017).Compared to developed countries, the building stock in rural areas of developing countries is often constructed from more traditional and less expensive building materials, which makes them more vulnerable to flooding.In this regard, urban settlements in the context of this study are defined as geographic units with built-up area that are more developed and have a higher built-up density than rural settlements.
The aim of this paper is to develop an approach for assessing large-scale flood risk in urban and rural areas using objectbased data from ImageCat to represent exposure, and to develop vulnerability curves for different building classes.The approach draws upon common practices in earthquake risk assessments, and relates damage by flood waters more directly to the vulnerability of buildings based on the building materials.We test the suitability of this approach for the case of Ethiopia, comparing our results with those using a more traditional large-scale flood damage modelling approach, examining how the increased detail influences risk estimates.

Data and Methods
The approach used in this study is composed of the following main four steps, and shown in Figure 1: 1) development of vulnerability classes and curves for different construction types and building materials based on a literature review of previous studies; 2) classification of an object-based exposure dataset using input data from ImageCat; 3) derivation of maximum damage values and 4) risk assessment by combining the aforementioned vulnerability and exposure with hazard data.
Each of these steps is described in more detail in the following subsections.

Vulnerability classes and curves
As a first step (Figure 1), an extensive literature review was conducted to develop flood vulnerability classes and associated curves based on construction types and building materials (Table 1).An increasing number of studies investigate multiparameter damage models (e.g.Chinh et al., 2016;Wagenaar et al., 2018), but given the large amount of data required to apply such models, we here only consider studies on river floods that apply stage-damage curves.For the class and curve development, we use studies from different regions that have focused on the vulnerability of individual construction types or building materials, and which are preferably based on actual event data.Some additional studies, often more qualitative in nature, were used to further refine the flood vulnerability classifications of the different building materials and construction types (e.g.Kappes et al., 2012;Laudan et al., 2017;Neubert et al., 2008;Zhai et al., 2005).Apart from reviewing the literature, experts with a structural engineering background were consulted to confirm the coherence of the final classification and vulnerability curves.In all of these studies, the construction type or (dominant) building material is clearly specified, and is either the only From the literature described in Table 1, we identified information to develop the stage-damage curve for each of these vulnerability classes.The stage-damage curves in most of the studies are concave, increasing steeply at low water depths (especially for the more vulnerable materials), and with a decreasing slope at higher water depths.This overall concave shape was differentiated into curves for each of the four vulnerability classes, shown in Figure 2, using information on threshold levels (e.g. the water depth at which most damage incurred) from the studies in Table 1.We distinguish curves that go up to 2.5m and 5m (for buildings with 1-and 2-floors) as flood levels rarely reach higher levels.Housing built through informal channels dominate in Africa (World Bank, 2015), and self-constructed buildings using inexpensive materials and traditional manufacturing techniques are very common (Alagbe and Opoko, 2013;Collier and Venables, 2015).Buildings of class I and II belong to this group and are assumed to be one floor only as multiple story buildings require higher quality materials and hiring a professional construction crew.The four vulnerability classes are described below: Class I are non-structured buildings created by mud, (non-cemented) adobe blocks and other traditional materials or informal buildings.Buildings in this class can dissolve and collapse easily when impacted by flood waters.These are the most vulnerable to flooding.Literature shows that mud walls will collapse when flooded by about a meter of water (Maiti, 2007), and submersion tests illustrate that most adobe bricks completely dissolve when submerged for 24 hours (Chen, 2009).
These buildings are assumed to be one floor only.
Class II consists of wood frame buildings.Theoretically, these are far less vulnerable to collapsing than class I, but if wood frames become wet, they often have to be replaced, or finishing needs to be removed for drying (and replaced afterwards).In a study carried out in Germany, Buck (2007) showed that damages can be ~35%-50% higher for wood frame homes than for masonry/concrete homes.However, the value of the wooden buildings in Ethiopia is much lower and they seem to be predominantly present in rural areas with more informal, less durable building material.Therefore, we decided to let the curve progress up to damage factor 1 at 2.5 m (i.e.damage can reach full building value, unlike masonry and concrete constructions).Buildings that are based on wood construction types can account for a large proportion of overall building stock in some countries (e.g.USA and Ethiopia).The quality of these constructions and the building's value can vary considerably.For large-scale assessments outside of Africa, adjustment towards a greater flood resistance is recommended.
Class III are unreinforced masonry/concrete buildings.These buildings are more vulnerable than those in class IV (reinforced masonry/concrete).This is related to the fact that unreinforced walls are less able to resist pressure of water exerted on walls.However, damage is assumed to be less than class II, as masonry and concrete are less likely to need Class IV represents reinforced masonry/concrete and steel buildings.These buildings are basically standard in most western countries and large cities in Africa.Overall, they constitute the most resistant class to flooding.Many studies (e.g.Buck, 2007;Li et al., 2016;Maiti, 2007) show that vulnerability curves for these types of buildings do not go up to a damage factor of 1, as some elements do not need replacement after a flood (e.g.foundation or carrying walls).This is similar to the values from Dutta et al. (2003) and HAZUS-MH (Scawthorn et al., 2006), who show examples of curves that go up to 0.6-0.7.
Therefore, it is chosen to let this curve go up to 0.65.Both masonry and concrete are put in the same class.

Object-based exposure data
In step 2 (Figure 1), we reclassify the objects identified in the ImageCat database into the four vulnerability classes, and distinguish between urban-and rural areas.The exposure data developed by ImageCat are available on a 15" x 15" grid for several African countries.Each grid cell contains building counts for different construction types, as well as the total floor area and total building value of the cell's building stock.In total, 22 construction types are differentiated in the ImageCat data.Most large-scale flood assessments focus on urban areas as due to the availability of data and high potential damages.In countries with large differences between urban and rural living standards, such as developing countries, it can be expected that the share of more vulnerable buildings (i.e.class I and II) is higher in rural areas compared to urban areas (e.g.Fiadzo, 2004).To account for these differences, we classify each cell as urban or rural.If more than 50% of the ImageCat objects in a cell belong to vulnerability class I or II, the area is assumed to be predominantly rural.
To check the assumption that the share of class I and II buildings in developing countries is higher in rural areas compared to urban areas, we examined these shares in the PAGER dataset (Jaiswal and Wald, 2008).PAGER is a global building inventory at the country level, which is often used in earthquake research.PAGER provides information at a country level on the construction types that make up the total urban and rural building stock.First, we reclassified the PAGER construction types into the four flood vulnerability classes used in our study (similar to Table 2).Then, we calculated the percentage of buildings in PAGER's total urban and rural building stocks that are categorised as class I and II (Figure 3).In low and lower middle income countries, the percentage of buildings in class I and II is indeed much higher in rural areas (36%) than in urban areas (10%).The chosen threshold to identify rural areas in the ImageCat dataset (>50%) is larger than the average share we find in PAGER (Figure 3).This means that cells identified as rural using the ImageCat data information about the built environment with the chosen threshold are quite likely to indeed be rural.In remote sensing or land-use studies, accuracy assessments determine a process' accomplishment of classifying an image (e.g.satellite data, aerial photos).Such an assessment requires reference values that represent the ground truth of the region of interest.Preferably these values are from ground collected data or hand-labelled high-resolution imagery validated by multiple interpreters (e.g.Goldblatt et al., 2018;Miyazaki et al., 2011).With these options out of the scope of this study, we examine the similarity between existing land-use products and classified areas in our approach.Compared to a strict accuracy assessment this holds the limitation of comparing already classified products.However, by benchmarking the classified ImageCat data against established and recently published products, we provide an assessment of how well areas are identified in comparison.To this end, we reviewed the quality of the urban-rural ImageCat map by visual comparison with satellite imagery and by overlap with other classification products, visually and by quantifying the agreement between classified areas of the ImageCat data and other products (section 3.1).Two comparisons are made, one for urban and rural areas, and one for only urban areas.Similar to an accuracy assessment, we express the performance of this overlap by calculating common comparison metrics from a confusion matrix such as overall accuracy, kappa coefficient, and producer's and user's accuracy for the sampling cells as described in Supplementary figure 1. Overall accuracy and kappa coefficient are metrics indicating the general agreement between the reference and comparison dataset.The latter two refer to the accuracy of individual classes of which the producer's accuracy describes the probability that, for example, an urban pixel is correctly classified, and the user's accuracy that a pixel classified as urban is actually urban.
For Ethiopia, the comparison maps are from several global land-use datasets as there are no other maps on national scale available for the country.For the reference map, the ImageCat data are assigned the reference categories 'urban', 'rural', and 'other land use' for cells outside of settlements.From the comparison maps, GHS-SMOD is the only other product that also considers rural settlements, allowing for a comparison of both urban and rural classifications.For GHS-SMOD, built-up areas are combined with population grids to differentiate between three settlement classes: urban centres, urban clusters, and rural (Pesaresi and Freire, 2016).In order to compare to the ImageCat reference, the GHS-SMOD's urban centre and cluster cells were reassigned into a single urban class and rural cells were kept as is.
More products are available that provide a classification limited to urban areas, but largely overlook rural areas, such as: GRUMP (CIESIN, 2011), MOD500 (Schneider et al., 2009), the Global Urban Footprint (GUF) (Esch et al., 2017), and HBASE (Global Human Built-up And Settlement Extent) (Wang et al., 2017).GRUMP and MOD500 are widely used land cover/use datasets, with GRUMP being a 30" x 30" grid of urban extent and MOD500 based on MODIS satellite data with a 500m x 500m resolution.GUF represents built-up area based on satellite imagery with a 12m x 12m spatial resolution.
HBASE is a 30m x 30m Landsat derived dataset of the extent of built-up area and settlements.All these products are used in the second comparison, in which only the 'urban' classified ImageCat settlements remain in the reference map and all cells outside of these settlements are reassigned to 'other land use'.From GHS-SMOD, the urban centre and cluster cells are again combined, but rural GHS-SMOD areas are excluded in this assessment.
Both the urban-rural and the sole urban classification comparisons between the ImageCat data and the other products follow a class defined stratified random sampling scheme, meaning that per class 10,000 sample points were randomly placed over the cells in each reference class.As the original maps do not all share a common geospatial model, they were reprojected to a 15" x 15" raster, using the WGS-84 datum.The results of the assessments can be found in section 3.1.

Maximum damage values
In step 3 (Figure 1), we determine the maximum damage of buildings in each vulnerability class.For a coherent set of input values, we use depreciated country-specific structural maximum damage estimates per square meter as provided by the JRC report of Huizinga et al. (2017), in which residential construction costs are estimated per country using a non-linear relationship between construction costs and GDP per capita.This maximum damage value needs to be further differentiated between the four different vulnerability classes used in our study, and then multiplied by an estimate of the building footprint area per cell.This is achieved by applying the following formula for each cell: Where D i is total structural maximum damage in a given cell (i), S is structural maximum damage per square metre in Ethiopia, N is the number of buildings belonging to vulnerability class k and cell i, A is the object area, meaning the building footprint for each vulnerability class k and cell i, and F is the maximum damage adjustment factor for vulnerability class k.
A and F are derived as follows: Building footprint area (A) Our assumptions on the number of floors are derived from information in the ImageCat country data descriptions.Since buildings of construction types belonging to vulnerability class I or II rarely exceed one floor, we assumed them to have one floor in both urban and rural areas.The construction of class III and IV buildings with more than one floor requires a higher skill level, mainly found in professional construction workers available in urban areas.Considering these characteristics, most class III buildings can be assumed to be one floor in rural areas.However, as most buildings in urban areas have more than one floor, we assumed class III buildings in urban areas to have two floors.Class IV buildings are assumed to be multiple floors in all areas.The resulting building footprints for Ethiopia can be found in Table 3.

Maximum damage adjustment factor (F)
The maximum damage values of Huizinga et al. (2017) are depreciated country-specific structural maximum damage estimates, averaged across various building types.Therefore, we differentiated these into maximum damage values for the four different vulnerability classes used in our study.Huyck and Eguchi (2017) provides estimates of replacement costs for different structures, based on factors such as input material and whether the structure is owner-built or engineered using professional contractors (Huyck and Eguchi, 2017).We used these to calculate the average replacement costs for each of the four vulnerability classes, for example the average for vulnerability class I is about 95 $/sqm.In order to apply comparable maximum damage values based on a coherent dataset, these average costs per vulnerability class are then put in ratio to the country-specific values from Huizinga et al. (2017), resulting in adjustment factors (F) for each vulnerability class (see Table 4) to arrive at maximum damage estimates.A detailed example of the maximum damage value can be found in Supplementary figure 2. The overall Ethiopian building stock is according to the ImageCat data comprised of over 16.8mln buildings.With the described approach, the total value exposed in urban areas amounts to about $250bln compared to almost $30bln in rural areas.

Risk assessment
To calculate the damage, we combine the new exposure and vulnerability data described above, with existing hazard maps derived from the GLOFRIS global flood risk model (WRI, 2018).These maps show inundation extent and depth at a horizontal resolution of 30'' x 30'' for different return periods.The original setup of GLOFRIS is described in Ward et al. (2013) and Winsemius et al. (2013).The maps used in this study are those developed for the current time-period in Winsemius et al. (2015), which have been further benchmarked against observations and high-resolution local models in Ward et al. (2017).In doing so, we estimate damage for the return periods 2, 5, 10, 25, 50, 100, 250, 500 and 1000 years.
We expressed flood risk using the commonly used metric of expected annual damage (EAD).This is estimated as the integral of the flood damage curve over all exceedance probabilities (e.g. Ward et al., 2013).A source of uncertainty in flood risk assessment is the level of incorporated flood protection.Here, we use the modelled protection standard for Ethiopia

Urban-rural identification
The results of our classification of ImageCat cells for Ethiopia into urban or rural are shown in Table 5, along with summaries of data from other data sources.For rural areas, our result (7.2%) is similar to that of GHS-SMOD (6.4%), which is the only other data source among the products that has a specific value for rural areas.The area in Ethiopia categorized as urban or built-up is relatively low in all data sources which is in accordance with Ethiopia being one of the least urbanized countries in Sub Saharan Africa, with the share of urban population being only between 11% and 16% (Schmidt and Kedir, 2009).

Visual comparison
Our urban-rural classification is shown spatially in the example of Figure 4, in which we compare different land-use products for an area near the City of Awasa.The urban and rural areas identified in GHS-SMOD and our classified ImageCat data show a more detailed and differentiated representation of the settlements than the coarse resolution GRUMP and MOD500 products.All products overlap in the location of main urban areas, although their extent varies.Locations of built-up areas with medium extent, for example in GUF, are hardly or not detected in HBASE, MOD500, and GRUMP, but are also seen with GHS-SMOD and our ImageCat classification.
Using our classification method, some smaller settlements are labelled urban with the ImageCat data, because their building stocks have high shares of class III and IV buildings, whilst GHS-SMOD classifies them as urban clusters or rural.Examples are the areas around Shashemene (see circled examples in Figure 4a).By visual inspection of Google Earth data, these seem to be areas of urban-rural transition.They have a more densely built environment than rural areas and a higher number of class III and IV buildings, which leads to the urban label in our method.Areas where cells from the ImageCat data get classified as rural are also rural in GHS-SMOD or to some extent urban clusters due to a higher population density in the surrounding cells.However, the overlap of these settlements is more about the general area and less regarding a cell by cell

Flood risk assessment
Modelled flood damages for the different return periods and risk for urban and rural areas are shown in Table 7.For 2-year return periods the damages are always zero as it is assumed that these flood events would not cause overbank flooding.As can be expected, the damages in urban areas are higher, as it is a more densely concentrated built-up environment and the value of the buildings is higher.On the other hand, the majority of exposed buildings are in rural areas.To illustrate, about 88,000 buildings in urban areas are exposed to a 100-year flood event, compared to more than four times as many rural buildings.The overall flood risk in Ethiopia (i.e.expected annual damage, EAD), is about $213.2mln/yr;78% of the total EAD is in urban areas.Whilst the rural EAD is below the EAD in urban areas, it is still high in absolute terms ($46.7mln/yr).This demonstrates that neglecting damages to rural buildings in large-scale assessments may lead to a severe underestimation of total damage values.Furthermore, the flood damages in urban and rural areas have to be considered in the context of the coping capacity of the population in the respective areas.The flood vulnerability of people below the poverty line is higher, as a larger proportion of their wealth could be affected during a flood event (Winsemius et al., 2018).While this is also true for the urban poor, the livelihoods of rural people are more susceptible where services and infrastructure are limited (Komi et al., 2016).

Comparison with Aqueduct
Compared to a traditional land-use-based model, the total simulated damages in our approach are somewhat higher, but similar in magnitude.For example, the new version of the GLOFRIS model used for the Aqueduct Global Floods tool (WRI, 2018) applies the same inundation data as used in this study, but follows the common approach of using land-use-based exposure and vulnerability data, resulting in EAD for Ethiopia of $182mln/yr.The results from our approach contain much more detail on the exposed elements and their vulnerability and allow us to examine damage in urban and rural areas.
Damage in urban and rural areas cannot be distinguished in GLOFRIS as it uses HYDE data (Klein Goldewijk et al., 2011) to represent exposure, which represents the urban built-up fraction per grid cell.Moreover, Figure 5 compares the land use exposure map using classified ImageCat data and HYDE for the example of Addis Ababa.As for the rest of the country, it demonstrates that datasets like the ImageCat exposure data can provide more spatial detail than the commonly used exposure maps such as HYDE used in land-use based flood risk models.Settlement extent and outlines are more distinctive, resulting in an overall better representation of affected settlement areas in the risk assessment of our approach.Further comparison with reported losses can be found in supplementary section 1.

Sensitivity analysis
Given the uncertainty in the input datasets and methods used in our approach, we perform a one-at-a-time sensitivity analysis to assess how the simulated EAD is affected by our assumptions on the: To assess the sensitivity of the results to the assumed values for maximum damage, we used the 90% confidence interval of estimated construction costs for residential buildings reported by Huizinga et al. (2017).These state that construction costs can be between 28% lower and 53% higher than the estimates used in this paper.For sensitivity to the threshold used in the urban/rural classification, we used thresholds of 20% and 80% for classifying urban areas, instead of the 50% used in the earlier analysis.Object areas can be very diverse between and within countries and depend on the characteristics of the housing market.For example, the Centre for affordable Housing Finance in Africa yearbooks include some indication on the average house size and price per country.However, the used sample sizes for example are very small and the average value covers only the minimum size that formal developers in urban areas are prepared to build, therefore neglecting self-built  Results of the sensitivity analysis are summarised in Table 9.Clearly, the flood risk estimate is very sensitive to the applied maximum damage values, as the EAD scales linearly with maximum damage changes.The results also show the EAD to be sensitive to the applied vulnerability curve.Using the single curve from GLOFRIS leads to a higher total estimate of risk by 41%.Therefore, the correct estimation of maximum damage values and improved representation of vulnerability are important considerations for large-scale flood risk modelling.Our approach improves the incorporation of vulnerability in the risk assessment by differentiating the built environment into classes that characterise the vulnerability of a building stock even on large scales.The EAD is very insensitive to the threshold used in the urban/rural classification.Even with the wide range of thresholds used in the sensitivity analysis, influence on the urban-rural distribution is minimal, confirming that the urban and rural built environment in Ethiopia is very distinct in terms of what materials and construction types are applied.
In our sensitivity analysis, the assumptions made on the object areas have little influence on the EAD, with overall slightly lower EAD when using alternative footprint sizes.Even though the effect is small here, it must be noted that these are estimated sizes and in reality building layouts are very diverse.

Figure 1
Figure 1 Flowchart for large-scale flood risk assessment using object-based data with a building-material-based vulnerability approach.
indicator, or one of the primary indicators, for the description of the flood vulnerability.Four vulnerability classes can be identified from this literature, of which each class consists of similar construction types and building materials with comparable behaviour towards flooding.The four classes are: (I) non-structured buildings built of materials such as mud and adobe or informal buildings; (II) wooden buildings; (III) unreinforced masonry/concrete buildings; and (IV) reinforced masonry/concrete and steel buildings.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.replacement after being flooded compared to wood.Therefore, a curve between class II and class IV was created for both one and two floors.

Figure 2
Figure 2 Stage-damage curves for four building-material-based vulnerability classes.For class III and IV the one and two floor curve are denoted by (a) and (b).

Figure 3
Figure 3 Average percentage of urban and rural buildings belonging to vulnerability classes I and II for different income groups and Africa according to PAGER for countries with different urban-rural inventory.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.As data on the footprint of different building types are not directly available, we estimated these based on floor area and number of floors derived from the ImageCat data.ImageCat provides estimates of floor areas for each construction type, based on sampling of building footprints, OSM data, interviews with local contractors and experts and literature review(Huyck and Eguchi, 2017).The country data descriptions also provide information on the typical number of floors, based on sampling.For each construction type, we divided the average floor area from the ImageCat data with the number of floors, and calculated the footprint area per class (A) as the average from the construction types belonging to each class.
(a) structural maximum damage values; (b) threshold used in the urban/rural classification; (c) object area; and (d) stage-damage curves.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.houses.Furthermore, no differentiation between building types or constructions is given(CAHF, 2017).For the sensitivity analysis, instead of calculating the footprint areas from average floor areas across the construction types per vulnerability class, we used the most frequent floor area size per type in the ImageCat data.The building footprint sizes most affected by this are those for classes II and III (see Supplementary table 4), as the size decreased with 5 to 11m 2 .The sensitivity regarding the vulnerability curves is analysed by applying only one vulnerability curve, thus neglecting the differentiation our model makes toward material-based vulnerability.To this end, we selected the residential stage-damage curve used in GLOFRIS, for which the degree of damage progresses slightly below the class III one floor curve.

Table 1 Overview of studies used to derive construction type and building material-based vulnerability classes and curves. The four classes are: (I) non-structured buildings created by mud, adobe or informal buildings; (II) wooden buildings; (III) unreinforced masonry/concrete buildings; and (IV) reinforced masonry/concrete and steel buildings
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscriptunderreviewfor journal Nat.Hazards Earth Syst.Sci.Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.Table1summarises the studies used to derive construction type and building material-based vulnerability classes and curves.
Table2shows how these can be reclassified into the four vulnerability classes used in our study.In the Ethiopian data nine of these types occur.Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.

Table 7 Simulated flood damages (in Million $ 2016) to building structures in urban and rural areas of Ethiopia, for different return periods (RP) .
Table8shows the damages per return period for the different vulnerability classes.These results show that most of the damage in rural areas results from damage to buildings of class I, which are buildings with the highest vulnerability.In urban areas, the largest share of the damage results from damage to buildings of class IV; these are the buildings with the highest exposed values.In addition, this class also makes up a large share of the exposed urban buildings, about 57,000 for a 100year flood event which is more than twice as many buildings of class III.In total more than 464,000 buildings are simulated to be affected for events with this return period, but most are in rural areas with the majority belonging to class I (58.3%)Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2019-32Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 12 March 2019 c Author(s) 2019.CC BY 4.0 License.