Quantifying the potential benefits of risk-mitigation strategies on future flood losses in Kathmandu Valley, Nepal

. Flood risk is expected to increase in many regions worldwide due to rapid urbanization and climate change if adequate risk-mitigation (or climate-change-adaptation) measures are not implemented. However, the exact benefits of these 10 measures remain unknown or inadequately quantified for potential future events in some multi-hazardflood-prone areas such as Kathmandu Valley, Nepal, which this paper addresses. This study examines the present (2021) and future (2031) flood risk in Kathmandu Valley, considering two flood-occurrence cases (with 100-year and 1000-year mean return periods) The analysis involves modeling two flood-occurrence cases (with 100-year and 1000-year mean return periods) and usinging four residential exposure inventories representing the current (2021) urban system (Scenario A) or near-future (2031) development trajectories 15 (Scenarios B, C, D) that Kathmandu Valley could experience. The results predict findings reveal substantial mean absolute financial losses (€ 473 million and € 775 million in repair/reconstruction costs) and mean loss ratios (2.8% and 4.5%) for the respective flood-occurrence cases in current times if the building stock’s quality is assumed to have remained the same as in 2011 (Scenario A). Under a “no change” pathway for 2031 (Scenario B), where the vulnerability of the expanding building stock remains the same as in 2011, mean absolute financial losses for the 100-year and 1000-year mean return period flooding 20 occurrences would respectively increase by 16% and 14% over those of Scenario A. However, a minimum (0.20 m) elevation of existing residential buildings located in the floodplains and the implementation of flood-hazard-informed land-use planning for 2031

This study contributes to the efforts required to quantify the benefits of appropriate mitigation strategies on growing flood risk 100 in urban areas for informing and promoting risk-sensitive decision-making (e.g., Cremen et al., 2022;Galasso et al., 2021a).
The work explicitly investigates the effect of various risk-mitigation strategies (i.e., elevating buildings, flood-hazard-informed land-use planning, building retrofitting, and building-code enforcement) on flood-induced financial losses in Kathmandu Valley, Nepal. The methodology integrates is a scenario-based flood loss estimation approach, using 100-year and 1000-year mean return period flood occurrence maps and four potential present (2021) and future (2031) exposure and vulnerability 105 scenarios, focusing only on residential buildings. Note that the impact of climate change is not explicitly considered within this work. The results can be relevant to various stakeholders, providing a clear quantitative description of the potential flood risk and its mitigation in Kathmandu Valley that can be leveraged for decision-making on investments in risk-reduction programs.

110
Quantifying flood risk requires modeling hazard (flood extent and flood depths), exposure (locations and characteristics of population and buildings), and vulnerability (the extent to which hazard affects exposed assets). Figure 1 provides a scheme that summarizes the methodology implemented in this study. The following subsections present further details of the study area, as well as the methods and data used for the analysis.

Study area
This study focuses on Kathmandu Valley, Nepal, which is surrounded by the Himalayan mountains and lies within the Bagmati river basin. The valleyKathmandu Valley occupies a total area of 721 km 2 , extending from 27°49'4" to 27°31'42" latitude and 120 85°11'19" to 85°33'57" longitude. Kathmandu Valley consisting of encloses three districts (Bhaktapur, Kathmandu, and Lalitpur), which comprise five municipal areas and several municipalities and rural municipalities (formerly named village development committees, or VDCs) (Mesta et al., 2022b), Figure 1. The built-up areas in Kathmandu Valley are estimated to be 202 km 2 for 2021 and are expected to increase to 307 km 2 by 2031 (Mesta et al., 2022b). (Mesta et al., 2022b) Figure 2 provides a physical map of Kathmandu Valley, showing elevation (available at https://earthexplorer.usgs.gov/, last accessed 125 December 2022) and the river network (available at https://openstreetmap.org/, last accessed December 2022). Figure 3 shows the administrative division of Kathmandu Valley and its built-up areas in 2021 and 2031 (Mesta et al., 2022b).  (Mesta et al., 2022b)

135
We use the high-resolution Fathom-Global model , which accounts for both fluvial and pluvial inundation. This global model uses the Multi-Error-Removed Improved-Terrain (MERIT) digital elevation model (Yamazaki et al., 2017) and MERIT Hydro (Yamazaki et al., 2019) as topography and hydrography datasets, respectively. These data provide the most accurate representation of ground surface elevation and location of the rivers at the global scale, which is critical to building robust flood models (Fathom, n.d.). Fluvial inundation is simulated in all river basins with upstream 140 catchment areas larger than 50 km 2 . At the same time, pluvial flooding is captured for all catchment sizes by simulating rainfall directly onto the modeled topography. The model considers a 2D shallow-water formulation to explicitly simulate flood wave propagation and a regionalized flood frequency analysis  to derive river discharge. Fathom-Global provides 90 m-resolution maps of flood extents and flood depths for multiple mean return periods (from 1:5 year to 1:1000 year). Note that hazard data of finer resolutions (e.g., 10 m or lower) are recommended for capturing the highly localized nature of flood 145 hazard (e.g., for representing small streams accurately) and associated risks at the urban scale (e.g., Afifi et al., 2019;Nofal and van de Lindt, 2021). In addition, urbanization effects on flood hazard (i.e., the replacement of natural ground with impermeable surfaces, changes to drainage or irrigation systems, and deforestation can increase runoff during precipitation eventsas a result of increased runoff during rainfall events) are not explicitly accounted for by the Fathom-Global model and are therefore neglected in our analyses. However, the primary purpose of this study is to test different exposure/vulnerability 150 scenarios using a common flood hazard input that is open and easily accessible; developing bespoke fine-resolution flood hazard models for the study area is not within the scope of this work.
We consider two cases of flooding occurrence in Kathmandu Valley. The first case is based on the Fathom-Global undefended flood map with a 100-year mean return period (i.e., 1% annual exceedance probability) and is intended to represent current flood hazard. Decision makers frequently use this type of map (e.g., to identify flood risk zones in the United States) (Ludy 155 and Kondolf, 2012;Federal Emergency Management Agency (FEMA), 2010). The second flood-occurrence case approximately reflects a situation in which flooding is exacerbated due to climate changemore severe and is based on the Fathom-Global undefended flood map with a 1000-year mean return period. The flood maps are resampled to 30 m using the nearest neighbor method to match the spatial resolution of the exposure maps (Díaz-Pacheco et al., 2018). We combine individual flood maps into aggregated hazard maps that represent fluvial-pluvial flooding for each mean return period by taking 160 their maximum depths in line with the method of Tate et al. (2021), who mosaiced fluvial and pluvial flood grids to generate an aggregated flood hazard map for the United States. The fluvial-pluvial hazard maps for each considered mean return period are presented in Figure 4; the individual flood maps are available online through the METEOR project (https://maps.meteorproject.org/map/flood-npl/, last accessed December 2022)2. Hereafter, we describe the flooding-occurrence cases using only the terms "100-year" and "1000-year", omitting the description "mean return period" for brevity. Overall, the aggregated flood 165 maps are largely dominated by the effects of pluvial flooding: in both 100-year and 1000-year aggregated flood maps, around 15% of the flooded areas are exposed to both types of flooding, 84% are only exposed to pluvial flooding, and less than 1% are only exposed to fluvial flooding. It should be noted that fluvial flooding generally results in low-velocity flows dominated by hydrostatic pressure, while pluvial flooding often features higher flow velocities (Gentile et al., 2022); these differences in velocity characteristics (Gentile et al., 2022) could be important for estimating flood damage in areas with steep terrain (Nofal 170 and van de Lindt, 2022). However, we use only flood depth as the intensity measure in this study, since it is widely used for flood loss estimation (e.g., Federal Emergency Management Agency (FEMA), 2022; Nofal and van de Lindt, 2022), and (Nofal and van de Lindt, 2022) flood velocities are more difficult to record than flood depths, requiring hydraulic simulations (e.g., Kreibich et al., 2009).

185
The proposed scenarios portray different conditions for Kathmandu Valley in terms of urban growth, the prevalence of varying building typologies, and the implementation of DRR measures (see Table 1). Seven building typologies are included in the considered exposure scenarios: adobe (A), brick/stone masonry with mud mortar (BSM), brick/stone masonry with cement mortar (BSC), wood-frame (W), current-construction-practice reinforced concrete (RC-CCP), well-designed reinforced concrete (RC-WDS), and reinforced masonry (RM). These typologies have been previously used by Chaulagain et al. (2016Chaulagain et al. ( , 190 2015 as well as Mesta et al. (2022a)

195
The specifications of each scenario are primarily provided in Mesta et al. (2022a); any deviations in these details for this specific study are documented in Sections 2.32.1 to 2.32.3. Note that we do not consider the presence of basements for any building typology since they do not seem to be a common feature of buildings in the valley (except for a minor proportion of RC buildings). For instance, an extensive post-earthquake damage assessment conducted by the National Society for Earthquake Technology-Nepal (NSET, 2016) described the presence of basements in modern high-rise RC buildings; however, 200 buildings with six or more stories only represent 1% of all buildings in Kathmandu Valley. Suwal et al. (2017) identified the presence of basements in 31% of RC buildings in the valley, but their study was limited to only 64 buildings.

Scenario A (population and buildings for 2021)
In this study, we group the BSM and BSC typologies under an individual building typology, titled BSM/BSC, since using either mud mortar or cement mortar does not alter building flood resistance within the specific vulnerability models used in 205 this study (see Section 2.43). Using the same reasoning, we group RC-CCP and RC-WDS under one building typology labeled RC-CCP/WDS. We determine the proportions of the building typologies per VDC municipality based on the 2011 census data (type of outer wall, type of foundation), as described in Mesta et al. (2022a). The exact height (number of stories) of each building is uncertain and is therefore randomly sampled using typology-specific empirical distributions or single values (not provided in Mesta et al., 2022a) that are derived from data collected for more than 20,000 buildings after the 2015 Gorkha 210 Earthquake by the NSET (2016). These distributions/values are defined as follows: two stories for A and W, between one and four stories (in the respective ratio 0.35:0.40:0.15:0.10) for brick buildings (i.e., BSM, BSC, RM), and between one and five stories (in the respective ratio 0.1:0.1:0.45:0.25:0.1) for concrete buildings (i.e., RC-CCP/WDS). We disaggregate the exposure data to match the 30 m spatial resolution of the urban map containing the 2021 built-up areas (see Figure 3).

215
Future exposure for Scenario B is simulated using the proportions of different building typologies and the empirical distributions/values of building heights defined in Scenario A. We disaggregate the exposure data into the urban map containing the 2021 and 2031 built-up areas (see Figure 3). Scenario C introduces two DRR measures that can reduce flood risk. The first DRR measure assumes that every existing building in a given floodplain is elevated by 0.2 m, a flood-risk mitigation measure proposed by Du et al. (2020) to reduce 220 flood losses in Shanghai, China. This elevation considers the construction of a 0.2m thick concrete platform above the ground floor, which constitutes a realistic (technically feasible) strategy to potentially reduce flood damage, especially in areas that experience low or moderate flood depths. Greater building elevations (as considered by Du et al., 2020) could prevent larger losses but would reduce household comfort (i.e., result in an excessive decrease in the floor-to-ceiling height). The second DRR measure consists of flood-hazard-informed land-use planning (i.e., the restriction of future urbanization in flood-prone

Modeling flood vulnerability
Since no specific flood vulnerability functions are developed for the study area, we adopt the depth-damage functions of the 265 global flood depth-loss model developed by the Joint Research Center (JRC) of the European Commission (Huizinga et al., 2017). More sophisticated analytical flood fragility and vulnerability functions, which propagate uncertainties in the hazard-dependent failure of building components and associated repair/replacement costs (e.g., Nofal et al., 2020;Nofal and van de Lindt, 2021), would require detailed component-level vulnerability information that is not available for this study.
The JRC vulnerability functions were developed for distinct continents and building occupancies (e.g., residential, commercial, 270 industrial). They express flood depth in meters and losses as mean loss ratios (i.e., financial losses as a percentage of the building replacement cost). We select the JRC vulnerability function for residential buildings in the Asian continent as our baseline function. We modify it to consider specific features of Kathmandu Valley's building stock. Firstly, we set a 100% maximum damage to be 100% for A and W, and 60% maximum damage for all brick (BSM/BSC, RM) and concrete (i.e., RC-CCP, RC-and WDS) typologies, following JRC recommendations. The 60% maximum damage threshold used for some 275 typologies reflects the assumption that a flood cannot damage major water-resistant structural components, which represent a substantial portion of building construction costs (Huizinga et al., 2017). This assumption is in line with other studies such as Both brick (BSM/BSC, RM) and concrete (i.e., RC-CCP, RC-WDS) typologies have higher durability compared with A and 290 W, and low permeability and represent the most flood-resistant buildings (e.g., Balasbaneh et al., 2019;Li et al., 2016).
Although the flood vulnerability may vary slightly between brick and concrete typologies (e.g., URM buildings are less able to resist the pressure of flood water exerted on walls than RM and RC buildings; Englhardt et al., 2019), these differences are not accounted for in the JRC vulnerability functions and thus are not included in this study.
The height-adjustment procedure of Gentile et al. (2022) assumes that the building replacement value is directly proportional 295 to the number of stories, which may not be strictly valid (e.g., electrical and mechanical equipment are usually installed on the ground floor and mixed occupancy buildings can have commercial areas on the ground floor, increasing the relative replacement values of this story). Furthermore, the construction costs considered in this study exclude content costs, such that the resulting financial losses may be underestimated. However, it should be emphasized that the exact losses for a given exposure scenario (absolute or relative) are not strictly of interest in this study. Instead, we focus on producing comparable 300 loss outputs for all exposure scenarios that are based on consistent assumptions. In this way, we aim to investigate how risk changes across the exposure scenarios in a relative sense. It is also worth noting that the uncertainty in the vulnerability model may strongly affect the loss estimation, particularly in terms of loss variability for a given mean return period. However, such uncertainty may be neglected if mean loss quantities are considered for comparison across different scenarios, as in this study.
305 Figure 53. Flood vulnerability functions for the different considered building typologies and their associated range of heights Moreover, to avoid overestimating losses, we account for the difference between the ground level (above which flood depth is reported in the hazard maps) and the ground-floor level (above which flood depth is measured in the vulnerability functions).
We set this difference at 0.2 m, as suggested in previous studies on flooding vulnerability for residential buildings (e.g.,

310
Dabbeek Maqsood et al., 2014;Olsen et al., 2015) and after consulting construction blueprints of buildings in the study area. Furthermore, we use the procedure detailed in Mesta et al. (2022b) to classify populations per VDC municipality as low, middle, or high income for facilitating socioeconomic disaggregation of financial losses. The classification is based on three variables (i.e., access to mobile/telephone services, mass media communication, and means of transportation) recorded in the 2011 Census, which are treated as proxies for economic wealth. The Census data are aggregated at the municipality 315 level; therefore, any variability in the population's income level within each municipality is not (and cannot be) assessed. The classification is quantile, such that the three income categories contain an equal number of municipalities. We assume that the population's income level did not vary between 2011 and 2021 and would remain unchanged in 2031, given the lack of available data to make confident projections. This assumption is at least partially supported by previous work from Cutter and Finch . (Cutter and Finch, 2008), who suggested that the social vulnerability of a community, which is influenced by its 320 underlying socio-economic and demographic characteristics (e.g., income level, gender, age), is not expected to vary significantly over timeframes similar to those considered in this study.   disproportionately occur in inundated areas. In Scenarios C and D, restricting future urban growth within the floodplain reduces the proportion of flood-exposed buildings by 6.7%.   VDCsmunicipalities, between 10% and 20% for 50 VDCsmunicipalities, and between 20% and 30% for 10 365 VDCsmunicipalities. In contrast, the proportions of buildings within the floodplain for Scenarios C and D are equal to or less than 10% for 79 VDCsmunicipalities, between 10% and 20% for 22 VDCsmunicipalities, and between 20% and 21% for 3

Distribution of buildings in the floodplain
VDCsmunicipalities, reflecting the benefits of constraining future urbanization to non-inundated areas. Figure 85 presents the municipalityVDC-level spatial distribution of buildings within the 1000-year floodplain. For Scenario A, proportions of buildings within the floodplain are equal to or less than 10% for 49 VDCsmunicipalities, between 10% and 370 20% for 43 VDCsmunicipalities, and between 20% and 28% for 12 VDCsmunicipalities. Corresponding Scenario B proportions are equal to or less than 10% for 31 VDCsmunicipalities, between 10% and 20% for 50 VDCsmunicipalities, and between 20% and 40% for 23 VDCsmunicipalities. Corresponding proportions for Scenarios C and D are equal to or less than 10% for 75 VDCsmunicipalities, between 10% and 20% for 24 VDCsmunicipalities, and between 20% and 25% for 5 VDCsmunicipalities.

Losses
390 Figure 69 (panels a, b) presents the mean loss ratios associated with the 100-year flooding occurrence, disaggregated by district and income level. From panel a, we observe some variability in the mean loss ratios by district. The Bhaktapur district exhibits the largest mean loss ratios for all the scenarios, which is explained by its disproportionate share of exposure in inundated areas (there are only minor differences in the prevalence of different building typologies between districts, and the three districts are dominated by brick and concrete building typologies that are assumed to have the same level of flood vulnerability; 395 see Section 2.3.) For instance, in Scenario A, the percentage of buildings in the floodplain is 14.9% in Bhaktapur, 14.5% in Kathmandu, and only 9.8% in Lalitpur. Proportions of buildings that experience flood depths below and above 2.0 m respectively are 72%-28% in Bhaktapur, 84%-16% in Kathmandu, and 66%-34% in Lalitpur. Similar results are observed for Scenario B, because the overall proportion of buildings within different flood-depth ranges of the floodplain remains largely identical (see Table 2), and the building stock's vulnerability is not changed. While DRR measures implemented in Scenarios 400 C and D reduce the mean loss ratios compared with those of Scenario B, the relative differences in mean loss ratios between districts are not particularly affected.
From Figure 69 (panel b), we identify some variability in the mean loss ratios by income level. All scenarios produce the highest mean loss ratios for the high-income population, which reflects their disproportionate share of buildings in inundated areas (there are also minor differences in the prevalence of building typologies between income groups, but the three income 405 groups are dominated by brick and concrete typologies). For instance, in Scenario A, the proportion of buildings in the floodplain is 15% in high-income VDCsmunicipalities, 11% in middle-income VDCsmunicipalities, and 12% in low-income VDCsmunicipalities. Proportions of buildings that experience flood depths below and above 2.0 m, respectively, are 81%-19% for high-income VDCsmunicipalities, 75%-25% for middle-income VDCsmunicipalities, and 84%-16% for low-income VDCsmunicipalities. Scenario B shows similar results to Scenario A due to its similar proportions of buildings within different 410 flood-depth ranges of the floodplain (see Table 2) and its identical quality of building stock. In addition, the benefits that result from the building elevation strategy and flood-hazard-informed land-use planning proposed in Scenario C are larger for the low-income population than for the other income groups; the mean loss ratio decreases from Scenario B to C by 44% for the low-income VDCsmunicipalities, by 28% for the middle-income VDCsmunicipalities, and by 22% for the high-income VDCsmunicipalities. There are two main reasons for this trend. On the one hand, low-income VDCs municipalities contain 415 the largest proportion of flood-exposed buildings in areas with flood depths below 1.0 m, where the effects of the elevation strategy are more noticeable (as implied by the steep initial slopes of the vulnerability curves presented in Figure 53). In Scenario B, the proportion of buildings that experience flood depths below 1.0 m is 63% for low-income VDCsmunicipalities, 50% for middle-income VDCsmunicipalities, and 51% for high-income VDCsmunicipalities. On the other hand, the proportions of new Scenario B buildings located in the floodplain are higher across low-income VDCs municipalities (34% in 420 total) than across middle-income (24%) and high-income (11%) VDCsmunicipalities. Furthermore, the benefits from the multihazard (i.e., flooding and seismic) risk-mitigation measures integrated within Scenario D are slightly better than those from the single-hazard-focused Scenario C: between Scenarios B and D, the mean loss ratio drops by 45% for the low-income VDCsmunicipalities, by 29% for the middle-income VDCsmunicipalities, and by 23% for the high-income VDCsmunicipalities.
425 Figure 96 (panels c, d) presents the mean loss ratios associated with the 1000-year flooding occurrence, disaggregated by district and income level. Similar to the 100-year flood case, there is an implicit relationship between the mean loss ratios and the extent of exposure in inundated regions; the same general trends for mean loss ratio across districts and income levels are observed for the more severe flooding occurrence. Bhaktapur district exhibits the highest mean loss ratios for all exposure scenarios, followed by Kathmandu and Lalitpur. All exposure scenarios result in the highest mean loss ratios for the high-430 income population, followed by the middle-income and low-income populations. The largest benefits from the risk mitigation strategies are also associated with the low-income population.  by more than € 74 million (+16%) in Scenario B, decrease by more than € 63 million (-13%) in Scenario C, and rise by more than €52 million in Scenario D (+11%), relative to Scenario A. For the 1000-year flooding occurrence, mean absolute losses increase by nearly € 108 million (+14%) in Scenario B, decrease by more than € 66 million (-9%) in Scenario C, and rise by more than € 130 million in Scenario D (+17%), relative to Scenario A. The relative increase in mean absolute financial losses for Scenario B is due to the presence of more assets in the floodplain. In other words, Scenario B demonstrates that a larger 450 population can easily lead to greater flood losses when risk mitigation is neglected. In contrast, the relative decrease in mean absolute financial losses for Scenario C shows that, despite a growing population, elevating existing buildings and implementing flood-hazard-informed land-use planning could significantly reduce flood losses in the future. However, it should be noted that risk-mitigation actions implemented in Scenario C would still leave the building stock highly vulnerable to earthquakes, and thus do not completely address multi-hazard risk in the valley, which is left to Scenario D. The relative 455 increase in mean absolute financial losses in Scenario D is associated with the larger replacement value of its building stock (due to the structural retrofitting and building code enforcement measures implemented), highlighting a tension between shortterm (pre-hazard occurrence) costs and long-term benefits (i.e., after the occurrence of hazard events) associated with holistic DRR measures. In summary, Scenario D demonstrates that, despite a growing population, adequate DRR measures that aim to improve the building stock's quality (for better sustaining both flood and earthquake damage) as well as incentivize 460 urbanization away from flood-sensitive areas can limit (but not reduce) flood losses in the future.
Absolute cChanges to the mean loss ratios provide additional interesting findings. In Scenario A, the mean loss ratios associated with the 100-year and 1000-year flooding occurrences are 2.8% and 4.5%, respectively. In Scenario B, as future urbanization continues occurring in both inundated and non-inundated areas and there are no changes in the building stock's quality, the mean loss ratios only show minimum variations compared to Scenario A. In Scenario C, elevating buildings and 465 the promotion of flood-hazard-informed land use produce a significant decrease in the mean loss ratios, which drop to 2.01% and 3.5% (27% and 23% smaller than in Scenario A), respectively. Due to additional improvements in the building stock's quality in Scenario D, the mean loss ratios drop further to 1.99% and 3.4% (28% and 24% smaller than in Scenario A, respectively). By comparing the mean loss ratios from both Scenarios C and D relative to Scenario A, we notice that seismicrisk mitigation interventions by themselves do not contribute much to reducing flood risk in the valley due to the low 470 replacement rate (<5%) of non-flood-resilient buildings (i.e., A, W) with flood-resilient buildings (i.e., RM) as a result of the seismic upgrading process. However, it is important to remember that Scenario D represents a much more robust approach to multi-hazard risk mitigation than Scenario C. Mean loss ratio 4.5% -0.17% -1.0% -1.1% Figure 710 and Figure 11 presents additional insights on the municipalityVDC-level spatial distribution of the mean loss ratio for the 100-year and 1000-year flooding occurrences, respectively. The alignment between the proportion of buildings located in inundated areas and the mean loss ratios is clear when the maps from Figure 107 and Figure  485 Figure 12 illustrates the absolute changes to the municipality-level mean loss ratios for the 100-year flooding occurrence, considering Scenario A as a baseline. In Scenario B, the mean loss ratios show small absolute variations (between -1.0% and +1.2%) compared to Scenario A, since future urbanization continues occurring in both flooded and non-flooded areas.
Nevertheless, tThe relative effects of the building elevation strategy and the flood-hazard-informed land-use planning proposed not contribute much to reducing flood risk. Figure 13 presents the absolute changes to the municipality-level mean loss ratios for the 1000-year flooding occurrence considering Scenario A as a baseline. In Scenario B, the mean loss ratios exhibit some absolute variations (between -1.0% and +2.3%) relative to Scenario A, which are larger than in the 100-year flood case; in other words, the consequences of not controlling future urbanization in flood-prone areas can increase with the severity of the considered flooding occurrence. T. Similar observations can be made regarding the VDC-level spatial distribution of the mean 500 loss ratio for the 1000-year flooding occurrence, depicted in Figure 8. The effects of the flood-specific DRR measures implemented in Scenario DC are as follows: absolute reductions in mean loss ratios for Scenario C relative to Scenario A decrease ranges in 12 VDCs betweenby 2.0-5.67.8% in 9 municipalities, in 36 VDCs betweenby 1.0-2.0% in 32 municipalities, and are less than 1.0% in the remaining in 33 VDCs by 0.5-1.0%, and in 22 VDCs by 0%-0.5%63 municipalities, relative to Scenario A or B. The benefits of the combined DRR measures in Scenario D are comparable to those in Scenario Cthe 505 following: mean loss ratios decrease in 13 VDCs by 2.0-7.9%, in 35 VDCs by 1.0-2.0%, in 35 VDCs by 0.5-1.0%, and in 20 VDCs by 0%-0.5%, relative to Scenario A or B.

525
The main results of this study provide a clear description of the current and potential near-future flood risk in Kathmandu Valley, suggesting that decision-makers of today have a unique opportunity to positively influence the risk of tomorrow, through their choices on implementing policies that control future risk drivers (e.g., Cremen et al., 2022b). However, we acknowledge that different sources of uncertainty and limitations of the data and methods used can influence the accuracy of the results obtained, which is now discussed.

530
In this study, we characterize the flood hazard using global maps with a coarse resolution (i.e., 90 m), which may not capture the highly localized nature of flood hazard (e.g., associated with small streams). While finer resolution hazard maps (e.g., 10 m or lower) are generally preferred for conducting regional flood risk assessments, the spatial resolution of the hazard model must also be consistent with the resolution of the exposure model used. We characterize exposure in the valley using urban maps with a spatial resolution of 30 m; therefore, our analyses would not benefit from hazard maps of a finer resolution. In 535 addition, some authors (e.g., Fatdillah et al., 2022;Zhang, 2020) report that using finer-resolution digital elevation models (DEM), which would be needed to produce finer-resolution flood hazard maps, can result in larger simulated flooded areas and losses compared to coarser-resolution DEM; however, other authors (e.g., McClean et al., 2020) suggest the opposite, indicating that flood risk may be exaggerated using flood maps based on global coarse DEM. These ambivalent findings suggest that the advantages of using finer-resolution flood maps for regional flood risk assessments, in fact, require careful 540 evaluation for each specific context. Another limitation of the flood maps employed in this study is that they do not capture the effects of urbanization on flood hazard (i.e., the replacement of natural ground with impermeable surfaces, changes to drainage or irrigation systems, and deforestation can increase runoff during precipitation events). The use of physics-based flood simulations that include future urban footprints would address this issue (e.g., Jenkins et al., 2022) but may entail a significant computational cost.

545
Uncertainties and limitations associated with the exposure and vulnerability models also affect the loss outputs. Due to the absence of a reliable database for Kathmandu Valley containing exact building footprints (and relevant attributes such as building typology and height), we construct our exposure model by downscaling data collected from census and surveys (mainly at the municipality level) into the built-up areas of the valley (i.e., dasymetric mapping). While exposure disaggregation techniques are widely used in regional risk assessments (e.g., Geiß et al., 2022;Dabbeek et al., 2020), it is 550 recommended to use original exposure models that are refined from the outset, since the accuracy of damage and loss estimates are highly sensitive to that of the exposure data. The accuracy of the exposure model may be of particular importance for flood loss assessments, given the potentially significant localized variability of flood hazard (i.e., flood depths can abruptly change even between closely-spaced locations). Moreover, the loss accuracy strongly depends on the quality of the vulnerability curves. In this study, we modify existing continental-based vulnerability curves to include relevant characteristics of the local 555 building stock in Kathmandu Valley (e.g., building typology and height). However, it is difficult to ascertain how much (if any) uncertainty and/or accuracy is effectively improved with these modifications.
The design and implementation of risk-mitigation strategies also face several challenges. For instance, policies that restrict future urbanization within floodplains rely on the accuracy of spatial designations made within flood maps. While flood maps provide a good basis for floodplain management, regulation, and mitigation-e.g., in the USA, 100-year flood maps are used 560 to identify Special Flood Hazard Areas where the National Flood Insurance Program's floodplain management regulations must be enforced (Ludy and Kondolf, 2012; Federal Emergency Management Agency (FEMA), 2010))-it is essential to acknowledge that different sources of uncertainty (e.g., climate change impacts, uncertainty in the hydrological/hydraulic models, etc.) can affect the resulting floodplain delineation (Zahmatkesh et al., 2021). Consequently, populations outside the designated floodplains may still be at risk of flooding and should be made aware of this.

565
A comprehensive sensitivity analysis could be conducted to investigate the impact of the aforementioned limitations on the results (e.g., Bernhofen et al., 2022). However, since the main focus of this study is to investigate relative risk changes across different sets of DRR-related actions, the exact values of absolute losses are not of particular interest or relevance.

570
This study has examined the present (2021) and future (2031)  The key findings of this study are as follows. First, rResults reveal that a significant proportion of the current building stock is 575 located within the 100-year and 1000-year floodplains (14% and 15%, respectively), which may lead to significant losses.
However, an appropriate combination of DRR measures (i.e., building elevation and flood-hazard-informed land-use planning) can substantially limit mean absolute financial losses and reduce relative versions of these losses (i.e., expressed as a proportion of associated replacement costs) in the future, compared to equivalent current levels. Second, tThis study reveals that highincome populations are exposed to the highest mean loss ratios across both flooding-occurrence cases due to having the largest 580 proportions of buildings in the floodplain. This contrasts with the trend in income versus earthquake-related losses identified for the same region in previous work (Mesta et al., 2022a), where low-income populations exhibited the highest seismic risk.
This discrepancy illustrates that risk-mitigation measures can have varying effects for different hazards; therefore, DRR plans should be appropriately tailored for a specific region or sub-region and account for multiple hazards. Kathmandu Valley's building stock is highly vulnerable to earthquakes due to the prevalence of URM buildings (particularly in low-income 585 municipalitiesVDCs), such as adobe and brick/stone masonry. However, this feature of the building stock does not make it particularly susceptible to flood damage (except in the case of adobe houses, which are made of mud), which is why a multihazard approach to DRR that also considers earthquake vulnerability strengthening measures has little effect on the mean loss ratios (and even results in increased mean absolute financial losses) in this study. Instead, the flood risk is mainly controlled by the extent to which populations are located in the floodplain. Considering that hazard intensities vary spatially and that 590 flooding and earthquake-induced ground shaking can affect different proportions of buildings in a given municipality VDC, combinations of individual DRR measures should be investigated to find the optimal DRR solution for a given municipalityVDC. Third, Furthermore, this study demonstrates that DRR initiatives uniformly targeting flood risk across different income levels produce the largest benefits for low-income populations. These findings are relevant because the benefits of mitigation measures are currently not well understood/quantified by various stakeholders in Nepal In summary, 595 this work provides important insights for decision-makers on how effective risk-informed policy making can limit future flood risk compared to current levels, particularly for low-income populations.
While this paper is focused on two levels of flooding occurrence, future research could analyze further scenarios to provide more robust results. Nonetheless, we do not expect the general trends identified in this study to significantly differ for other flood occurrence cases. Fine-resolution local hazard models (if and when available) could be used to more accurately quantify 600 flood hazard (and the associated risk), explicitly including the effect of building footprints, climate change, etc. Moreover, updated census information (when available) could be employed to adjust present and future exposure estimations. In addition, the accuracy of the characterization of physical vulnerability could be improved through appropriate modifications to the selected vulnerability functions in line with local construction practices. Future research could also investigate the effectiveness of other possible flood-related DRR actions (e.g., ring dike, wet-proofing, dry-proofing, nature-based solutions, relocation).
In summary, this paper addresses the essential need to communicate the growing flood risk in Kathmandu Valley and 610 potentially encourage local (or even Nepal-wide) risk-mitigation efforts. The adopted methodology can be easily extended to other geographical contexts to quantify the impacts of other (multiple) natural hazards on the present and future built environment, providing decision makers with an adequate understanding of the risk consequences of particular actions and the importance of particular risk mitigation/adaptation strategies Galasso et al., 2021).

615
Data availability. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.The flood hazard maps are available online through the METEOR project (available at https://maps.meteorproject.org/map/flood-npl/, last accessed December 2022). The urban maps for Kathmandu Valley are available online through a public repository (available at https://doi.org/10.5281/zenodo.7406981, last accessed December 2022). OtherThe datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

620
Author contributions. C.M., G.C., and C.G. conceived and designed the research. C.M. drafted the written content of the manuscript, performed the calculations, and developed the figures. All authors reviewed the manuscript.
Competing interests. The authors declare no competing interests.