100086 Daniel B. Wright, Ph.D., Disaster Risk Management Consultant** OVERVIEW This technical note provides an overview for authorities who wish to conduct flood hazard and risk assessments and who must develop a step-by-step plan for carrying out the assessment that is appropriate and feasible in the local context. It is important to keep in mind, however, that many aspects of flood hazard and risk assessment requires specific expertise and experience. It is not advisable to conduct these assessments if your team lacks this experience, and this technical note does not provide specific guidelines, which can vary dramatically depending on local and regional conditions. 1. Introduction location, and types of buildings and other assets that could be damaged. Poorly-conducted Floods are amongst the most frequent and hazard and risk assessments can lead to poor risk destructive type of disaster, causing significant management decisions, from insufficient damage and disrupting livelihoods throughout protection to the wasting of scarce finances on the world. A wide range of flood risk unneeded protection. Well-conducted flood management can reduce this destruction, and hazard and risk assessments, on the other hand, managing flood risks requires the estimation of can provide valuable support for a range of flood hazards and the impacts that they can decisions such as land-use master planning, cause. Proper estimation of risk is challenging design of infrastructure, and emergency and requires careful consideration of a number response preparation. of factors, including watershed properties such as size, topography, and land use, the types and Different types of floods can be found in Figure characteristics of storms that produce rainfall 1. Before a hazard assessment is carried out, it and flooding in the region, and the number, is necessary to determine which types of floods are most common or destructive in the area, * This document has been produced under the guidance and supervision of Fernando Ramirez-Cortés and Oscar A. Ishizawa, Senior Disaster Risk Management Specialists, as part of the Technical Notes developed under the World Bank LCR Probabilistic Risk Assessment Program (CAPRA). ** Technical review by Maria Carolina Rogelis, Senior Consultant on Flood Risk Assessment. because in most cases the selection of hazard and focuses on fluvial floods (i.e. floods in river risk modeling methods will vary depending on systems), the general concepts are applicable for the type of flood. Detailed specifications for the other types of floods shown in Figure 1. flood hazard assessment can be found in (FEMA, 2003 (a)). While this technical note Figure 1 Types of flood hazards (adapted from (Díez-Herrero, Huerta, & Isidro, 2009)). The goal of flood hazard assessment is to 2. Flood Hazard Assessment understand the probability that a flood of a particular intensity will occur over an 2.1. Overview extended period of time. Hazard assessment aims to estimate this probability over periods of A natural hazard is a potentially damaging years to decades to support risk management physical event, phenomenon or human activity, activities1. Intensity usually refers to the which may cause the loss of life or injury, combination of flood depth and horizontal flood property damage, social and economic extent; although other intensity measures such as disruption, or environmental degradation flow velocity and flood duration can also be (UNISDR, 2004). “Potentially damaging” important depending on the situation. means that there are elements exposed to the hazard that could, but need not necessarily, be This relationship between the probability of a harmed (Gouldby & Samuels, 2005). flood and its intensity gives rise to the concept of return period, represented by the symbol T 1 This contrasts with early warning systems, which to facilitate emergency response actions. aim to issue warnings over periods of minutes to days 2 and expressed in terms of years. A T-year flood 2.2.1. Statistical Discharge Frequency Analysis is the flood intensity that has a probability of 1/T of being exceeded in a given year. This This approach relies on the existence of long probability is called the exceedance probability. records of accurate river discharge For example, there is an exceedance probability measurements. Usually, the highest recorded of 1/10 (0.10) that in a given year there will discharge record from each year is used in the occur a flood larger than the 10-year flood analysis. For example, if there are 30 years of intensity. Flood hazard assessments usually aim daily discharge measurements available at a to estimate the flood intensity for a range of particular measuring station, then 30 data points exceedance probabilities, for example from 0.1 are used in the discharge-frequency analysis, to 0.001. It is important to point out that this each one corresponding to the largest daily definition of return period is contrary to what the discharge observation from one of the 30 years term “10-year flood” or “100-year flood” would of record. These data points are referred to as seem to imply, (i.e. the intensity of a flood that annual discharge maxima. Once these data would occur once every ten or one hundred points have been identified, the analyst fits years). This is a major source of confusion and several statistical distributions (for example: misunderstanding of the definition of return log-normal, log-Pearson, or generalized extreme periods and can result in improper estimation or value) and selects the distribution that most perception of hazard and risk. accurately describes the data. Figure 2 illustrates this type of analysis performed using discharge 2.2. Estimation of design discharge measurements from a station in Puerto Rico. A partial list of tools for performing statistical This section focuses on the methods used to analysis tools can be found at (FEMA, 2012). It estimate design discharge, or the rate of the flow must be emphasized that the proper application of water through the river or floodplain. The two and interpretation of statistical procedures most frequently-used approaches are outlined requires substantial experience and specialized here—discharge-frequency analysis and knowledge. rainfall-runoff modeling. While other methods are sometimes used, they typically supplement, Important considerations in discharge rather than replace, one or both of these two frequency analysis: approaches.  The discharge records must be of good quality. Proper measurement of discharge Discharge-frequency analysis approaches can be requires maintenance of equipment to used to estimate peak discharge, or the provide for continual automatic monitoring maximum flow rate of water that passes a certain of water levels, as well as verification using location during a flood. Rainfall-runoff field measurements of flow and river cross modeling can also be used to estimate peak sectional profiles at a range of flow discharge and usually also a design conditions at least several times per year hydrograph, or an estimate of the flow rate past (Buchanan & Somers, 1969). a certain location over a period of time. Figure  The discharge records must be sufficiently 3 demonstrates the relationship between a long to estimate the return periods that are discharge hydrograph and peak discharge. A required for the flood hazard analysis. There detailed review of riverine and coastal flood is no single guideline, but records should be hazard assessment procedures, tools, and data at least 10 years in length to perform any sort can be found in (Prinos, 2008). of frequency analysis (Interagency Advisory 3 Committee on Water Data, 1982), and longer (Villarini, Smith, Baeck, Sturdevant-Rees, & if estimates of low exceedance probabilities Krajewski, 2010). are desired.  The temporal resolution of the discharge record needs to be fine enough to measure the important properties of floods in the river. For example, daily discharge measurements are adequate on large rivers but not for steep mountain watersheds or in small urban basins, where automated measurements at time intervals of 1-5 minutes may be needed.  The upstream area cannot have undergone significant changes in terms of land use such as urbanization, agricultural development, or deforestation over the period of the discharge record. If significant changes have Figure 3 Example hydrograph and peak discharge taken place, the results of standard statistical value from a heavy rainfall event on March 27, 2012, analyses will not be valid. More advanced measured at the U.S. Geological Survey discharge gage statistical methods exist that attempt to station on Rio Grande de Manatí near Morovis, Puerto Rico. account for these challenges, but still have significant limitations and are generally not A design discharge obtained from discharge- recommended for decision-making frequency analysis is usually only valid in the vicinity of the measurement station, particularly Figure 2 Left—Plot of annual maxima discharge observations from 1965-2011 for the U.S. Geological Survey discharge gage station on Rio Grande de Manatí near Morovis, Puerto Rico. Right —comparison of statistical models based on the generalized extreme value and lognormal distributions for the annual maxima discharge observations on Rio Grande de Manatí near Morovis. 4 if there are important tributaries or other level of the user. A list of most commonly discharge sources upstream or downstream of accepted models for flood applications can be the station. Consequently, a relatively large found at (FEMA, 2014). It is extremely number of measurement stations are needed to important to consider these factors when adequate estimate design discharges over a large selecting a rainfall-runoff model. While some of or complex river system. these models are very simple and can involve simple hand calculations, most are complex, 2.2.2. Rainfall-runoff modeling computer-based, and require specialized knowledge to use correctly. The two main In many situations, discharge measurements are classes of rainfall-runoff models are: either nonexistent or of insufficient quantity or quality to be able to conduct a discharge  Lumped: Lumped models treat the frequency analysis as described in Section 2.2.1. watershed as a single unit. Calculations In such situations, one of a broad class of tools are performed using simplified, spatially averaged processes. The resulting discharge estimate only applies to the watershed outlet (the most downstream modeled point of the river network). Well-known lumped models include TR- 55 and other unit hydrograph-based methods.  Distributed: Distributed models use spatially varying input data for processes such as precipitation, infiltration, interception, interflow, infiltration, and base flow estimating discharge or other variables. This kind of model demands more data than lumped models but are more flexible and can be more accurate. Distributed models are often intended for Figure 4: Diagram showing the general components of use at a particular range of scales, such grid-based rainfall-runoff model. Adapted from as small urban watersheds (SWMM, http://www.endlessland.org/ (accessed 6 January 2014). GSSHA, Vflo, OpenLISEM) or large known as rainfall-runoff models (also referred river basins (VIC). Most distributed to as hydrologic models) can be used to convert models, if properly used, can provide estimates of extreme rainfall into design discharge estimates at various locations discharge estimates and design hydrographs. To along the river network. do so, they must represent the movement of Some popular models, such as HEC-HMS, water across the landscape (a process known as discretize the watershed into a number of runoff) and into the river channel. Some of the subwatersheds, each of which is then basic hydrologic processes that these models represented using a lumped rainfall runoff consider are shown in Figure 4. Many different model. The results of the individual lumped rainfall-runoff models exist, each with certain models can thus be combined to estimate advantages and disadvantages depending a range watershed response at various points. In this of factors such as application, geographic way, some of the advantages of lumped and setting, and data availability, and knowledge distributed models can be combined. 5 results are presented as intensity-duration- In addition to the selection of a proper rainfall- frequency (IDF, also sometimes referred to as runoff model, the suitability of this approach to intensity-frequency-duration) curves. Example flood hazard assessment depends on availability IDF curves are shown in Figure 5. Ideally, IDF of high-quality local or regional rainfall datasets curves are provided for multiple durations and and other information to characterize the for a wide range of return periods. If the desired watershed such as topographic maps or digital rainfall duration or return lies between two IDF elevation models (DEMs), land cover and soil curves, the proper rainfall intensity can be information, and the location and properties of linearly interpolated from the published curves. river channels and other water bodies. It is also These curves can facilitate rainfall-runoff crucial that measurements of rainfall and either analyses, but have important limitations, flood discharge or flood extent are available for especially in large watersheds and for long- one or more past floods so that the model can be duration rainfall (note, for example, that the verified and adjusted to account for local rainfall durations shown in Figure 5 only extend conditions. The process of verification against to 60 minutes). More robust alternatives are local measurements is known as validation; the available but not widely used (see, for example process of adjusting various components of the (Wright, Smith, & Baeck, 2014)).. model so that the simulated results better match Once a rainfall intensity of a given return period local measurements is known as calibration. Once a model has been calibrated and validated, it can be used for estimating design flows. While more sophisticated approaches exist, the usual starting point is to conduct a rainfall frequency analysis on annual rainfall maxima. The rainfall frequency analysis procedure is similar to that described for discharge frequency analysis in Section 3.2.1, and is subject to many of the same general challenges. One important additional component in rainfall frequency analysis is that Figure 5: Example IDF curves the rainfall duration must also be selected. (http://onlinemanuals.txdot.gov/, accessed 28 May Proper selection of rainfall duration is very 2014). important because floods in small urbanized or is estimated, it is usually assumed that, when mountainous watersheds result from extreme used as input to the rainfall-runoff model, the rainfall lasting several minutes to several hours, resulting simulated discharge has the same while in large river systems, flooding can result return period. For example, if the 12-hour from rainfall lasting several days to several duration 100-year rainfall intensity is estimated weeks. Therefore, selecting the improper for the region containing the watershed and then rainfall duration can lead to poor estimates of used as input to the watershed rainfall-runoff flood discharge. Guidance on the selection of model, the simulated peak discharge output is rainfall duration can be found in most assumed to be an adequate estimate of 100-year introductory hydrologic engineering texts (see, peak discharge. This assumption is not strictly for example, (McCuen, 2005)). valid for several reasons and more advanced Rainfall frequency analyses have already been methods can be used to avoid it, but it is conducted for some areas. In these cases, the 6 generally accepted in standard flood hazard short-duration rainfall that causes floods in practice (Wright, Smith, & Baeck, 2014). steep mountain watersheds or in small urban There are several important considerations when areas (Schilling, 1991; Berne, Delrieu, estimating design discharge using rainfall-runoff Creutin, & Obled, 2004), but can be models: adequate for larger rivers.  The rainfall records must be of good quality.  Depending on the region and the selected Proper rainfall measurement requires proper rainfall-runoff model, additional detailed installation as well as frequent maintenance information is needed. This information can of equipment and its surroundings (Curtis, include land cover data, the location and 1996; Sieck, Burges, & Steiner, 2007). characteristics of river channels, and  The rainfall records must be sufficiently long information to characterize soils and to estimate the return periods that are groundwater flow. required for the flood hazard analysis. There  It is difficult to perform rainfall-runoff is no single guideline, but records should be modeling in watersheds that have significant at least 10 years in length to perform any sort regulation (i.e. manmade controls of river of frequency analysis (Interagency Advisory flows via reservoirs and other hydraulic Committee on Water Data, 1982), and longer infrastructure). Some rainfall-runoff models if estimates of very low exceedance provide the capability to include these probabilities are desired. effects, to varying levels of detail. Even if a  In addition to having sufficiently long selected model has this capability, however, rainfall records of the appropriate duration, it it is rare that the modeler has information is usually necessary to have multiple rain regarding how this infrastructure will be gages over the region or watershed because operated during flood conditions. there can be significant spatial variations in  If flood hazard and risk estimates are needed rainfall intensity within individual storms. at many points in a large watershed, it will Multiple gages allow for the characterization be necessary to perform multiple simulations of this variation and for the interpolation of with different rainfall properties (i.e. rainfall to fill in gaps between gages. There different rainfall durations and intensities). are no simple guidelines for how many rain This is because the rainfall duration and gages are needed nor how these gaps should intensity that causes floods in the main river be filled. Accounting for spatial variation in will be different than the rainfall duration rainfall is particularly challenging in and intensity that causes floods in the smaller mountainous areas due to topographic upstream tributaries (Wright, Smith, & influences on storm motion and rainfall Baeck, 2014). generation. When available, rainfall estimates from weather radar or satellites can The challenges associated with rainfall-runoff be used to assess spatial variations of modeling point to the necessity of trained rainfall, but significant experience is needed hydrologists, preferably with previous to properly use such datasets for flood experience in the region and with the specific applications. rainfall-runoff model that has been selected.  The temporal resolution of the rainfall record needs to be fine enough to measure rainfall 2.3. Hydraulic modeling and floodplain at the durations that cause flooding in the mapping watershed. For example, hourly rainfall measurements are insufficient to estimate the 7 Once a design discharge (either a peak discharge dimensional (3D) models, which in the past value or hydrograph) has been estimated, it is these were rarely used for flood hazard transformed into an estimate of flood water assessment due to their complexity and cost. elevation, known as flood stage, and velocity However, the use of 3D models such as Delft3D using a hydraulic model (also called a is increasingly common, particularly for hydrodynamic model). In the past, these models simulating coastal flooding due to storm surge. were often small-scale physical models, but now Hydraulic models can be run in two different they are almost always computer-based ways, depending on the application and the simulation software. There are many hydraulic nature of the input design discharge. If the models that vary significantly in complexity and design discharge is a peak discharge estimate, data requirements. A list of commonly-used then the models must be run in a steady flow models can be found at (FEMA, 2014). An mode. This is the least intensive mode from a experienced modeler can select the most computation standpoint. While steady flow appropriate model based on the location, the mode is frequently used in flood hazard needs of the risk assessment, and the available assessment, it may not capture the complicated data. The two most common classes of flow dynamics in complex terrain such as urban hydraulic models are: floodplains. If the design discharge is a  1-Dimensional (1D): 1D models are complete hydrograph, then the models may be simplified models that characterize the run in unsteady mode, in which discharge rates terrain using a series of cross sections. At and water levels across the model area can vary each cross section, the flow depth and over time. Unsteady simulations can require velocity perpendicular to the cross section is significant computational resources in some computed. These models are well suited for cases, particularly when 2D or 3D models are areas where the direction of flow is well used. defined. The best-known 1D model is HEC- Accurate hydraulic modeling requires detailed RAS from the U.S. Army Corps of information regarding the river channel and the Engineers. HEC-RAS is free to download at floodplain. Some hydraulic models, such as the http://www.hec.usace.army.mil/software/he quasi-2D LISFLOOD-FP, rely on a single c-ras/. digital terrain model (DTM) consisting of  2-Dimensional (2D): 2D models calculate regularly-sized square grid “cells” in which the the flow both parallel and non-parallel to the channel and the floodplain are both represented. main flow. They are useful for modeling areas of complex topography such as wider floodplains or broad estuaries but require high quality data and can require long computation times. Examples of 2D models include TELEMAC 2D, SOBEK 1D2D, and Flo2D. Because of their greater sophistication, most 2D models are not freely available. Some models, such as LISFLOOD-FP, are called quasi-2D, and combine some of the benefits of 2D models with some of the simplicities of 1D models. There are also 3- 8 detailed procedure for commissioning a LiDAR survey can be found in (FEMA, 2003 (b)). Geometric information of river- or lake beds or the seafloor, called bathymetry, requires field surveys or LiDAR and can be difficult to collect for large or deep water bodies. In some cases, bathymetric data are not necessary for flood hydraulic modeling in rivers because the channel features can be adequately represented by idealized rectangular or trapezoidal shapes. This decision should be made only by an experienced hydraulic modeler. While the required accuracy of the DTM varies depending on the specific location and level of detail required by the study, vertical precision should be 1-2 meters or better. In some cases, vertical precision must be 5-10 centimeters. Consequently, freely available topographic Figure 6 LiDAR DTM. Different colors denote datasets based on satellite observations such as different elevation ranges. Black areas indicate the SRTM or ASTER GDEM are not adequate for outlines of buildings or submerged areas. From hydraulic modeling in many settings. The http://lias.cis.rit.edu/ (accessed 6 January 6, 2014). required horizontal resolution depends heavily Most 2D models also use make use of data from on the specific application and on local such a DTM, but require conversion from a conditions, and can vary from approximately 1 regular square grid to a triangular mesh prior to meter up to 100 meters or more. simulation. As previously mentioned, most 1D In addition to topographic and bathymetric models such as HEC-RAS make use of multiple information, hydraulic models require the cross-sections which are perpendicular to the specification of the hydraulic roughness, or predominant flow direction and which resistance to flow, of the river channel and the incorporate both river channel and floodplain floodplain. The roughness is low in bedrock geometry. This geometry can come from ground river channels that are free of vegetation or surveys or from aircraft-based instruments. debris and high in a debris-filled river channel or Given the difficulty of conducting precise a floodplain that has thick vegetation or ground-based topographic surveys over large buildings. Estimates of hydraulic roughness are areas, aircraft-based instruments such as LiDAR usually based on expert judgment using visual (Light Detection and Ranging) are the most surveys or aerial photography. Manmade common method for collecting floodplain structures such as bridges, culverts, and levees topographic information for large-scale flood or geological features such as rock outcrops that modeling activities. For shorter river sections, can impede flood flows can be very important however, field surveys are still commonly used. for hydraulic modeling and usually require field Example results of a LiDAR survey are shown surveys. in Figure 6. It should be noted that the collection and preparation of these data can be quite As with rainfall-runoff modeling, calibration expensive and requires specialized expertise. A and validation are crucial steps in hydraulic 9 modeling. Usually this consists of adjusting In the case of 2D models, the flood extent, depth, model parameters (typically hydraulic and velocity can be visualized directly in the roughness) while simulating past floods in an model output. For 1D models, it is usually attempt to recreate observed water elevations necessary to project the estimated flood water and flood inundation extents. It is important, levels onto a DTM to estimate flood extent and therefore, that flooded locations and maximum depth. This DTM is often the same as that used water levels be recorded during post-flood field to configure the model. If not, the two DTMs damage surveys or using aerial or satellite must share a common elevation datum and the photography (Gaume & Borga, 2008; Abhas, same aforementioned precision and horizontal Bloch, & Lamond, 2012). resolution requirements apply. Once the hydraulic model is properly Historical flood information from major flood configured, calibrated, and validated, it can be events that can also be used to produce flood used for flood extent mapping. An example of a hazard maps. These maps can be the result of 100-year flood extent map is shown in Figure 7. carefully-conducted field surveys, aerial In this example, only flood extent is shown. In photography, or more recently, satellite-based situations where flow velocity is high, both flood imagery. In order to estimate the frequency of depth and velocity maps should be produced. flood hazard, flood extent maps from multiple Fortunately, all hydraulic models readily past floods are available. These maps do not calculate both water level and velocity. exist in many locations and thus, in practice, flood extent maps and imagery are more commonly used for hydraulic model calibration. Figure 7 Example 100-year flood extent map for River Eden in Carlisle, UK. From http://web.sbe.hw.ac.uk/frmrc/index.htm (accessed 6 January 2014). 10 3. Vulnerability Assessment regional experience. In general, finer-scale vulnerability assessment will yield more accurate 3.1 Overview estimates of flood risk but at greater cost. The other major component of flood risk, aside 3.2 Exposure from flood hazard, is flood vulnerability. The goal of vulnerability assessment is to The flooded area shown in a flood hazard map understand how a system will be affected by usually does not display the flooded houses, floods. Examples of possible systems could factories, etc. nor their respective characteristics. include physical structures such as houses or Exposure analysis, therefore, aims to examine bridges that could be damaged or destroyed, a the economic assets and activities covered by the business or service whose supply chain could flood (Kang, Su, & Chang, 2005). Exposure is a face interruption, or a community that could geospatial mapping of the types of assets of suffer fatalities, property losses, and negative interest relative to the flood hazard (i.e. flood health impacts in the aftermath of a flood. extent). In addition, exposure information should include at least some basic characteristics of the There are different classes of vulnerability. assets in question. For example, a risk Some of these classes are briefly discussed assessment of residential properties would below, but this document focuses mainly on require exposure on information on the locations physical vulnerability—meaning the of the properties, the type, the number of floors, vulnerability of the built environment to floods. the floor area, etc. The focus on physical vulnerability stems from the fact that it is the most obvious and easily quantified vulnerability class, and because in many situations it constitutes a large share of total flood vulnerability. For simplicity, flood risk studies and management efforts often only consider one or several of these classes. Regardless of the vulnerability class under consideration, mapping is a central element of any assessment. Expertise in mapping tools such as Geographic Information Systems (GIS) and Global Positioning Systems (GPS) is therefore critical. Despite its importance, vulnerability assessment for floods and other natural disasters has received less attention than has hazard assessment (Changnon, 2003). There have been few efforts to standardize vulnerability measurement and estimation techniques, particularly in developing countries and for non-economic measures. Depending on the context, however, non-economic considerations Figure 8 Example land use map for flood exposure in can be extremely important. Because of this, it Cologne, with flood prone areas superimposed (http://www.viewsoftheworld.net/, accessed 28 May is important to involve flood vulnerability 2014). experts, preferably with previous local or 11 The scale at which flood exposure mapping curves (also referred to as damage functions or should be conducted will vary depending on the vulnerability functions), such as the examples needs of a particular flood risk assessment. shown in Figure 9 and Figure 10. Figure 9 shows Detailed mapping in urban areas can be a damage curve for a structure. This curve ranges conducted at the scale of individual buildings from 0% (no damage) to 100% (totally using handheld GPS or with aerial or satellite destroyed), describing the relationship between photography. For large floodplains, it may be damage and flood characteristics such as water preferable to model vulnerability at the scale of depth or velocity. larger administrative units, such as lots or census tracts, or using existing land-use maps. In the case of flood mitigation works such as An example of a coarse exposure map based on levees, the resistance and effectiveness of land use can be seen in Figure 8. Regardless of infrastructure can be represented by fragility the chosen method, the data will be stored curves (also referred to as fragility functions). entered into a geospatial database using a GIS Figure 10 shows a fragility curve for a flood interface. control levee, in which the probability of failure ranges from 0.0 at zero water depth to 1.0 at the 3.3 Physical Vulnerability maximum height of the levee (at which point the levee fails due to overtopping). Between these An important consequence of flooding is the two water levels, the probability of failure is damage to physical structures such as buildings, greater than 0.0 but less than 1.0 due to different bridges, roads, and public utilities. Damage can failure modes such as rotational failures, piping, be defined as the amount of money needed to scour, etc. (Allsop, Kortenhaus, & Morris, 2007). restore the area back to its original condition It should be noted that damage can still occur to before the disaster (Kang, Su, & Chang, 2005) a flood mitigation structure even if the structure and can be caused by lateral pressure forces, itself does not fail. For example, a levee may velocity forces, uplift, erosion of foundations, need to undergo repairs after a flood due to gradual weakening due to waterlogging, and piping or scour damage, even if the levee is not other effects (Kelman & Spence, 2004). overtopped. These repair costs could be represented as a function of height (and/or flood duration) by an additional damage curve. Damage and fragility curves are typically assigned to particular asset classes or construction types. For example, separate damage functions should be developed for residential buildings, industrial buildings, etc. Likewise, brick houses will require separate damage functions than adobe or than wood houses. Similarly, masonry flood retaining walls will require different fragility curves than earthen Figure 9 Example flood damage curve based on levees. Determining these functions is a major observations. The blue line is a best-fit to the challenge, as there can be significant variations observations. The gray area is an estimate of in the construction type and quality even within uncertainty (the 95% confidence bounds). the same structure class. In addition, there are The impacts of flood forces on structures need usually very few measurements of damage based to be understood and represented. These on local construction materials and practices on impacts are typically described by damage which these functions can be based. Structural analysis and expert opinion can help in the 12 development of damage functions (Schultz, to structures and building contents. Possible Goulby, Simm, & Wibowo, 2010). Estimates of indirect economic impacts include: economic losses will usually be very sensitive to the selected damage functions.  Lost business activity due to interrupted supply of utilities such as water or electricity.  Lost business activity due to interruption of supply chains caused by large floods.  Spoiling of agricultural products due to a damaged transportation network.  Lost wages for employees due to any of the above impacts. A detailed review of types of economic damage, several estimation techniques, and remaining challenges can be found in (Merz, Kreibich, Schwarze, & Thieken, 2010). Social vulnerability can be another important vulnerability class. Floods can cause death directly via drowning, physical trauma, or secondary effects such as the failure of water and sanitation services, the spread of waterborne Figure 10 Example fragility curve for a section of diseases and decreased nutrition (Abhas, Bloch, flood control levee. & Lamond, 2012). Even when death does not result, floods can cause injury, psychological The most obvious direct economic impact is the trauma, negative health impacts, and stress cost associated with repairing or replacing stemming from loss of housing or employment. damaged buildings and infrastructure. These These impacts range in scope from individuals impacts can be estimated by combining damage and families to entire communities. These functions with estimated replacement costs. impacts are oftentimes concentrated amongst the The replacement costs of contents such as a poor, who tend to live in flood-prone areas, have business’s inventory or a family’s possessions little access to flood warnings and evacuation can also be significant and should be included services, and generally have fewer resources and in the analysis. services to draw upon to recover from a flood. Because social vulnerability has many different 3.4 Nonphysical Vulnerability dimensions, there are many metrics that can be used to quantify it. A review of potential social In addition to the physical vulnerability of vulnerabilities to flooding and some ways of buildings and infrastructure, there are other measuring these vulnerabilities can be found in important vulnerability classes. (Tapsell, Penning-Rowsell, Tunstall, & Wilson, The most commonly-considered nonphysical 2002). Health impacts of floods are reviewed in vulnerability is economic loss. There are (Few, Ahern, Franziska, & Kovats, 2004; Ahern, multiple ways in which floods can have Kovats, Wilkinson, Few, & Matthies, 2005). economic impacts, mostly related to the Social vulnerability is oftentimes not included in interruption of various activities that then lead quantitative risk assessments, however, since it to negative economic consequences. These can be very difficult to quantify. This omission indirect economic impacts are often much more can lead to significant underestimation of overall challenging to estimate than are direct impacts risk. 13 Floods can have also have significant impacts on the natural environment. These 4. Risk Assessment environmental impacts can sometimes be beneficial, because the natural environment has adapted to the occurrence of floods over Once a flood hazard and vulnerability assessment thousands of years. For example, rivers can has been completed, it is relatively simple to carry large amounts of sediment during floods, arrive at an estimate of flood risk. An example which can help to rebuild delta regions, fish of the relationship between the different steps is spawning areas, and agricultural soils. Floods shown in Figure 11. Firstly, a statistical or help to replenish groundwater supplies. rainfall-runoff model is used to estimate design Seasonal floods signal fish and other organisms discharge for various exceedance probabilities, to reproduce or to migrate. In areas where shown in the upper-right. Then these design flooding is caused by or exacerbated by human discharges are converted into flood elevations activities, however, the environmental impacts (also known as flood stage), as shown in the of floods can be very negative. This is because upper-left, using a hydraulic model and human impacts such as urban or agricultural floodplain mapping. The lower-left shows the development can alter the characteristics of damages associated with different flood stages. floods, making them more intense and Finally the economic risk is shown in the lower- destructive than they would have been in the right, represented by monetary damage for absence of human influence. For example, various exceedance probabilities. A similar urban and agricultural development tends to procedure could be used for estimating social, increase the onset speed of floods because water structural, or economic risks, as long as the moves more quickly across lots, streets, and different steps can be properly quantified. drainage systems. The faster-moving water can carry more sediment, thus increasing scouring effects on structures and causing unnaturally large buildups of sediment in certain areas and intense erosion and landslides in others. Deforestation can further exacerbate these problems by limiting the landscape’s ability to retain soil and absorb water. Flood risk management projects can restore flood characteristics to more natural conditions, or at least minimize the additional negative environmental impacts. Environmental impact experts and ecologists can assist in the identification of possible positive and negative environmental impact and with the management of these impacts. 14 Figure 11 Conceptual diagram of how discharge estimates are converted into stage and economic damage estimates at a particular exceedance probability. that exceedance probability curves and average The result of the final step in Figure 11 (the annual losses could also be developed for non- lower-right) is an example of what is called an structural damages. One could, for example, exceedance probability curve. Another compute an exceedance probability curve for example exceedance probability curve is shown average annual inventory lost due to floods or an in Figure 12 This curve is very useful for average annual number of people displaced due decision-making purposes because it allows the to floods. estimation of a number of useful quantities. The most important of these quantities is the average It is clear that completely “correct” risk annual loss (also known as expected annual calculations can only be accomplished when the loss), the average loss to occur per year over a probability and the magnitude of the loss can be long time period. The average annual loss is estimated with complete accuracy. This is computed by finding the expected value of the impossible in practice due to the limited amount loss across all return periods. The exceedance and accuracy of information. There are many curve can also help guide decisions to reduce situations in which flood hazard, vulnerability, or risks against very rare and intense floods. Note both are very difficult to quantify due to a lack of sufficient information. In such cases, one can 15 either attempt to collect additional information to estimate hazard and vulnerability, or attempt to make risk management decisions without this 5. Uncertainty in Flood Hazard and information. Making flood risk management decisions with no or poor risk information can Risk Assessment result either in too little or too much protection. Too little protection means that citizens or Even when good information is available, it will economic assets face continued exposure to never capture all of the details of flood hazard flood impacts, while too much protection means and vulnerability. A major challenge in flood that money has been unnecessarily spent on hazard and risk assessment is to understand the unneeded protection. Procedures for evaluating uncertainties that exist at every stage of the the costs and benefits of flood risk management process, and to decide how to incorporate these measures can be found in (Medina, 2006) and uncertainties into subsequent risk management (Johnson, Hansen, Warren, Reynolds, Foley, & decisions. Fulton, 1988). For example, the estimation of design discharges, whether done using statistical methods or rainfall-runoff models, always depends on the use of multiple assumptions, incomplete datasets, and imperfect models. This will lead to errors in design discharges, which will in turn lead to errors in the water levels estimated using a hydraulic model, which will combine with imperfections in the hydraulic model to affect the predicted spatial extent of flooding. Likewise, there are considerable uncertainties in vulnerability assessment. For example, variations in the characteristics of individual buildings cannot be captured using a single damage function, but it is impractical to create accurate damage functions for each individual home or business that might be affected. The inability to characterize building-level flood Figure 12 Example exceedance probability curve. impacts will then translate into errors in resulting estimates of damage and economic losses. 16 Even the most careful analyses performed by leading experts cannot avoid such uncertainties. Though these uncertainties cannot be eliminated, their importance and impact on the decision-making process can and should be examined. In fact, flood hazard and risk experts put considerable efforts into understanding the various uncertainties and how they can affect risk estimates. An example is shown in Figure 13. These maps are taken from a demonstration by the Flood Risk Management Research Consortium (http://web.sbe.hw.ac.uk/frmrc) and show how the 0.01 exceedance probability (100-year) flood extent can vary for different levels of uncertainty. As can be seen in the figure, uncertainties in inputs to the floodplain mapping process can result in large variations in estimated flood extent. These variations can translate into significant differences in estimated economic and other types of flood damage and loss. These differences imply significant uncertainty in the decision-making required to reduce these damages. Uncertainties should be carefully considered when considering risk management investments. (Rogelis, 2012) provides a more detailed explanation of the various sources of uncertainty and how they can be incorporated Figure 13 Example of how consideration of hazard into in flood hazard and risk assessment. estimation uncertainties can affect predicted 100-year flood extent. From http://web.sbe.hw.ac.uk/frmrc/index.htm (accessed 6 January 2014). 17 6. References (2012). Numerical Models Meeting the Minimum Requirement of the National Flood Insurance Program - Current Nationally Accepted Statistical Models. [Online] Retrieved from: Abhas, K., Bloch, R., & Lamond, J. (2012). Cities http://www.fema.gov/national-flood-insurance- and Flooding: A Guide to Integrated Urban Flood program-flood-hazard-mapping/numerical-models- Risk Management for the 21st Century. Washington, meeting-minimum-requirement-2. [Accessed 14 D.C., USA: The World Bank. January, 2014]. Ahern, M., Kovats, S. R., Wilkinson, P., Few, R., & Federal Emergency Management Agency (FEMA). Matthies, F. (2005). Global Health Impacts of (2014). Numerical Models Meeting the Minimum Floods: Epidemiologic Evidence. Epidemiologic Requirements of the National Flood Insurance Reviews , 27, 36-46. Program - Current Nationally Accepted Models. [Online] Retrieved from: Allsop, W., Kortenhaus, A., & Morris, M. (2007). http://www.fema.gov/national-flood-insurance- Failure Mechanisms for Flood Defence Structures. program-flood-hazard-mapping/numerical-models- meeting-minimum-requirements. [Accessed 14 Berne, A., Delrieu, G., Creutin, J.-D., & Obled, C. January, 2014]. (2004). Temporal and spatial resolution of rainfall Few, R., Ahern, M., Franziska, M., & Kovats, S. measurements required for urban hydrology. (2004). Floods, health and climate change: a Journal of Hydrology , 299 (3-4), 166-179. strategic review. Tyndall Center for Climate Change Research. Buchanan, T., & Somers, W. (1969). Discharge measurements at gaging stations. Techniques of Gaume, E., & Borga, M. (2008). Post-flood field Water-Resources Investigations, book 3, chap A8. investigations in upland catchments after major flash U.S. Geological Survey. floods: proposal of a methodologyand illustrations. Journal of Flood Risk Management , 1, 175-189. Changnon, S. D. (2003). Measures of Economic Impacts of Weather Extremes. Bulletin of the Gouldby, B., & Samuels, P. (2005). Language of American Meteorological Society , 83, 1231-1235. Risk-Project Definitions. HR Wallingford. Curtis, D. (1996). Inadvertent Rain Gauge Interagency Advisory Committee on Water Data. Inconsistencies and Their Effect on Hydrologic (1982). Guidelines for determining flood flow Analysis. California-Nevada ALERT Users Group frequency, Bulletin 17B. Reston: U.S. Dept. of the Conference , 4. Interior, USGS, Offfice of Water Data Coordination. Federal Emergency Management Agency (FEMA). Johnson, N. B., Hansen, W. J., Warren, J., Reynolds, (2003 (a)). Guidelines and Specifications for Flood F. R., Foley, C. O., & Fulton, R. L. (1988). National Hazard Mapping Partners. [Online] Retrieved Economic Development Procedures Manual Urban: from: http://www.fema.gov/media-library- Flood Damage. Institute for Water Resources. Fort data/20130726-1543-20490-1120/frm__gsv1.pdf. Belvoir: U.S. Army Corps of Engineers. [Accessed 14 January, 2014]. Kang, J., Su, M., & Chang, L.. (2005). Loss functions Federal Emergency Management Agency (FEMA). and framework for regional flood damage estimation (2003 (b)). Guidelines and Specifications for Flood in residential areas. Journal of Marine Science and Hazard Mapping Partners- Appendix A: Guidance Technology, 193-199. for Aerial Mapping and Surveying. [Online] Retrieved from: http://www.fema.gov/media- Kelman, I., & Spence, R. (2004). An overview of library-data/20130726-1543-20490- flood actions on buildings. Engineering Geology , 73 2653/frm__gsaa.pdf. [Accessed 9 July, 2014]. (3-4), 297-309. Federal Emergency Management Agency (FEMA). 18 McCuen, R. (2005). Hydrologic Analysis and Center. Washington, D.C.: US Army Corps of Design, Pearson Prentice Hall. Engineers. Medina, D. (2006). Benefit-Cost Analysis of Flood Sieck, L., Burges, S. J., & Steiner, M. (2007). Protection Measures. Technical Memorandum, Challenges in obtaining reliable measurements of CH2M Hill. point rainfall. Water Resources Research , 43, W01420. Merz, B., Kreibich, H., Schwarze, R., & Thieken, A. (2010). Assessment of economic flood damage. Tapsell, S. M., Penning-Rowsell, E. C., Tunstall, S. Natural Hazards and Earth Systems Science , 1697- M., & Wilson, T. L. (2002). Vulnerability to flooding: 1724. health and social dimensions. Phil. Trans. R. Soc. Lond. A , 360, 1511-1525. Prinos, P. (2008). Review of Flood Hazard Mapping. European Community Sixth Framework Programme United Nations Office for Disaster Risk Reduction for European Research and Technological (UNISDR). (2004). Living with risk: a global review Development. FLOODsite. of disaster reduction initiatives. 429 p. Rogelis, M. (2012). Advances in Probabilistic Flood Villarini, G., Smith, J. A., Baeck, M. L., Sturdevant- Hazard Assessment. CAPRA. Washington D.C.: Rees, P., & Krajewski, W. F. (2010). Radar analyses The World Bank. of extreme rainfall and flooding in urban drainage basins. Journal of Hydrology , 381 (3-4), 266-286. Schilling, W. (1991). Rainfall data for urban hydrology: what do we need? Atmospheric Wright, D. B., Smith, J. A., & Baeck, M. L. (2014). Research, 27 (1-3), 5-21. Flood Frequency Analysis Using Radar Rainfall Fields and Stochastic Storm Transposition. Water Schultz, M. T., Goulby, B. P., Simm, J. D., & Resources Research. Wibowo, J. L. (2010). Beyond the Factor of Safety: Developing Fragility Curves to Characterize System Reliability. Engineer Research and Development 19 Funded by: © 2015 The World Bank. This work is a product of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work.