Paper No. JAWRA-08-0042-P of the Journal of the American Water Resources Association (JAWRA). © 2008 American Water Resources Association. Discussions are open until August 1, 2009.
Assessing Effectiveness of National Flood Policy Through Spatiotemporal Monitoring of Socioeconomic Exposure1
Article first published online: 29 DEC 2008
© 2008 American Water Resources Association
JAWRA Journal of the American Water Resources Association
Volume 45, Issue 1, pages 237–252, February 2009
How to Cite
Patterson, L. A. and Doyle, M. W. (2009), Assessing Effectiveness of National Flood Policy Through Spatiotemporal Monitoring of Socioeconomic Exposure. JAWRA Journal of the American Water Resources Association, 45: 237–252. doi: 10.1111/j.1752-1688.2008.00275.x
- Issue published online: 27 JAN 2009
- Article first published online: 29 DEC 2008
- Received March 6, 2008; accepted July 13, 2008.
- flood policy;
- 100-year floodplain;
- geospatial analysis;
- risk assessment;
- North Carolina;
- Top of page
- Study Area
- Literature Cited
Abstract: After a century of evolving flood policies, there has been a steady increase in flood losses, which has partly been driven by development in flood prone areas. National flood policy was revised in 1994 to focus on limiting and reducing the amount of development inside the 100-year floodplain, with the goal of decreasing flood losses, which can be measured and quantified in terms of population and property value inside the 100-year floodplain. Monitoring changes in these measurable indicators can inform where and how effective national floodplain management strategies have been. National flood policies are restricted to the spatial extent of the 100-year floodplain, thus there are no development regulations to protect against flooding adjacent to this boundary. No consistent monitoring has been undertaken to examine the effect of flood policy on development immediately outside the 100-year floodplain. We developed a standardized methodology, which leveraged national data to quantify changes in population and building tax value (exposure). We applied this approach to counties in North Carolina to assess (1) temporal changes, before and after the 1994 policy and (2) spatial changes, inside and adjacent to the 100-year floodplain. Temporal results indicate the Piedmont and Mountain Region had limited success at reducing exposure within the 100-year floodplain, while the Coastal Plain successfully reduced exposure. Spatially, there was a significant increase in exposure immediately outside the 100-year floodplain throughout North Carolina. The lack of consistent monitoring has resulted in the continuation of this unintended consequence, which could be a significant driver of increased flood losses as any flood even slightly higher than the 100-year floodplain will have a disproportionately large impact since development is outside the legal boundary of national flood policy.
- Top of page
- Study Area
- Literature Cited
Flood Losses and Policies
During the 20th century, floods caused seven million fatalities and US$250 billion in damages worldwide (Linnerooth-Bayer et al., 2001; Cohen and Werker, 2004); and regardless of efforts to limit flood impacts, the frequency of flood events and losses has continued to increase (van Aalst, 2006; Berz, 2006). Despite vast resources, technology and capital, the United States (U.S.) has not been immune to the global trends of increasing flood losses. After nearly a century of nationally focused flood policy efforts, floods remain the most costly natural hazard in the U.S. with average annual losses of US$2.4 to 4 billion (Mileti, 1999; Cutter and Emrich, 2005). Between 1992 and 2001 alone, flooding accounted for 90% of all natural disasters in the U.S. (GAO, 2005) and contributed to 900 deaths and US$55 billion in damages (GAO, 2004). Currently, seven to eight million households are exposed to significant flood risk and represent considerable economic liabilities (Burby, 2002; Riggs, 2004). As socioeconomic exposure to floods has risen, there has been a concurrent increase in the intensity and frequency of extreme precipitation events (CRED, 2008). This coupling of increasing exposure and flood frequency has contributed to the increasing costs of floods at a rate of 3.45% annually (Pielke et al., 2002; Cartwright, 2005).
Despite billions of dollars spent to decrease flood losses in the U.S., there has been a steady increase in flood losses that is partially driven by the persistent movement of people into flood prone areas. This trend will likely continue with projected population growth (Mitchell, 2006). Reducing future flood impacts requires refocusing on the implementation of floodplain management and policies. Unfortunately, the effectiveness of current floodplain policies has not been monitored or assessed frequently over time, leading to uncertainty regarding what policies are most effective at reducing the potential for future losses.
National flood management strategies in the U.S. have evolved over the past 90 years. Initially, flood policy was preventative, using structural flood control measures to contain flooding. This was followed by the National Flood Insurance Program (NFIP) in 1968 as a reactive policy to spread damage costs among those at significant risk for a flood event. However, extensive flood losses suffered in the 1993 Mississippi River Basin Flood resulted in formal reviews of the effectiveness of national flood policies. These reviews culminated in the Galloway Report, which found that the optimal strategy for reducing flood losses was to “limit or even reduce infrastructure on floodplains” (IFMRC, 1994). Accordingly, U.S. national policy entered the “mitigation era” (Godschalk et al., 1999) and began a proactive effort to limit floodplain development and remove preexisting infrastructure. In 1994, the NFIP was accordingly reformed to further encourage obtaining and maintaining flood insurance. The reformed NFIP also created provisions and increased funding for the Hazard Mitigation Grant Program (HMGP) and the Flood Mitigation Assistance Program (FMA) to remove or mitigate repetitively flooded structures from the floodplain based on a 75% cost share to the Federal Government (FEMA, 1999; FEMA, 2006). In reality, the implementation of these reforms have had little impact on floodplain management (Pinter, 2005), but few studies have quantified changes in floodplain development following the 1994 national policy revisions.
This paper focuses on national flood policy in the U.S., which is the national plan of action that guides decisions to achieve the outcome of decreasing flood losses by reducing risk, exposure, and/or vulnerability. National flood policy provides a minimum standard for floodplain management regulations while allowing states and localities the flexibility to implement additional policies focused on reducing flood losses. Floodplain management regulations include zoning ordinances, subdivision regulations, building codes, etc. to control future development in floodplains and to mitigate preexisting development (IFMRC, 1994). The floodplain management strategy of interest for monitoring in this paper is the limiting or removal of development in the floodplain area regulated by national policy.
Establishing the 100-Year Floodplain in U.S. Flood Policy
The U.S. flood policy is based on the 100-year floodplain boundary, which is the horizontal extent of land that would be inundated by a flood with a 1% chance of occurring in any given year (i.e., 100-year recurrence interval). The 100-year floodplain serves as a uniform boundary to administer and enforce flood policy (Reuss, 2004; Robinson, 2004). In 1968, as part of the NFIP, Flood Insurance Rate Maps (FIRMs) were created to depict the 100-year and 500-year floodplain boundary. Within the 100-year floodplain boundary, the 1994 reformed NFIP discourages new development (but see Pinter, 2005) and requires existing buildings to purchase flood insurance and mitigate their property by adopting more stringent building requirements. Immediately outside the 100-year boundary there are no official requirements for flood mitigation or insurance.
The effect of FIRMs on reducing flood losses has not been consistently monitored due to the lack of national guidelines and data collection standards for flood loss events (Pielke et al., 2002; Cartwright, 2005). The effectiveness of floodplain management strategies, like the HMGP and FMA, depend on the transfer of policy goals into specific measurable terms that can be monitored (NCDEM, 2007). One goal for national floodplain management is to reduce flood exposure, which can be measured and quantified in terms of population and property value located inside the 100-year floodplain. Assessing changes in these measurable indicators can inform floodplain managers on where and how effective floodplain management strategies to limit and reduce development have been.
Edge Effects of the 100-Year Floodplain Boundary
FIRMs depict the 100-year boundary as being well-defined, yet the process of delineating the 100-year floodplain boundary is irreducibly complex and has uncertainty due to limited and low resolution data, the stochastic nature of floods, and epistemic errors in model assumption (Table 1). The zone classification of the 100-year floodplain relates to two broad distinctions depending on the accuracy of the floodplain delineation. Zones AE, AH, and VE, comprising 73% of FIRMs in North Carolina, are the most accurately delineated zones as they were derived from detailed Flood Insurance Studies and engineering models. The remaining 27% of FIRMs in North Carolina are classified as Zone A, which means they were derived without the benefit of detailed studies. In addition, FIRMs are updated once every 10 to 20 years at the county level (FEMA, 2002) and display the 100-year boundary as stationary through time. However, climate change studies indicate an increased likelihood of more extreme precipitation events, which will produce more frequent and larger floods (Pielke and Downton, 2000; Hirsh et al., 2004; O’Brien et al., 2006). Moreover, land development in watersheds can double the 100-year flood height (Bana e Costa et al., 2004; Hirsh et al., 2004). Thus, the extent of the 100-year floodplain can change faster than FIRMs are updated. This generally results in an underestimated prediction of the 100-year floodplain boundary, which can lead to the unintentional exclusion of properties from flood insurance and mitigation policy.
|Sources of Uncertainty||Vertical Error (m)||References|
|Temporal and spatial coverage||Unknown||Apel et al. (2004); IFMRC (1994); Vaill (2000)|
|Elevation data quality||0.49-0.91||IFMRC (1994); Smemoe et al. (2007)|
|Model assumptions||0.15-0.61||Thomas and Baker (2004); IFMRC (1994)|
|Changes due to climate/land use||1.2-2.7||Hirsh et al. (2004); Tobin (2004); Lulloff (2004); Burby (2002)|
|Known potential error range||1.8-4.2|
There are no established flood policies to protect properties immediately outside the 100-year floodplain boundary, although the probability of being flooded is only slightly less than just inside the boundary. No consistent monitoring has been undertaken to examine the unintentional effect of national flood policy on development adjacent to the 100-year floodplain, which could be located inside the 100-year floodplain as land use and climate change alters the flood regime. It is important to monitor this development, since it has nearly the same probability of flooding and none of the protection afforded by the NFIP.
Hurricane Floyd and the Map Modernization Program
The 1999 flooding in North Carolina during Hurricane Floyd provides an example of how FIRM uncertainty can contribute to significant flood losses. Hurricane Floyd caused 56 deaths and tangible economic losses ranging from US$3 to 6 billion (Dorman and Bakolia, 2002; Pielke et al., 2002). Studies examining this event found that FIRMs underestimated the extent of the 100-year floodplain and unregulated development was allowed to occur in areas of high risk for flood losses (Dorman and Bakolia, 2002).
Following Hurricane Floyd, FEMA initiated a map modernization program to update and produce Digital Flood Insurance Rate Maps (DFIRMs), with North Carolina serving as its pilot state. The underlying assumption is that DFIRMs will decrease future flood losses because of their increased accessibility, accuracy, and up-to-date status. Prior to the map modernization program over 75% of FIRMs were older than 10 years and could no longer be used with confidence (FEMA, 2002).
The shift from hardcopies to digital data has allowed map revisions and distributions to be handled digitally, reducing costs with an estimated 2.8:1 benefit-cost ratio (Dorman and Bakolia, 2002; Raber, 2003). The use of digital data promotes spatial data sharing between government officials and increases accessibility to both policy makers and the general public. In North Carolina, DFIRMs are created using light detection and ranging (LiDAR) digital elevation models (DEM) which are more accurate and at a finer resolution than the DEM used to create FIRMs. DFIRMs are produced using the same methods and assumptions as FIRMs, with the key difference being the quality of the elevation data. However, it should be noted that in other states, there are DFIRMs created without LiDAR, some of which were digitally converted from an existing paper map. The benefits of creating DFIRMs have been quantified; however, an important but overlooked issue that needs to be assessed is the impact of changes in the 100-year floodplain boundary from FIRM to DFIRM on development and whether these changes can reduce future flood losses.
Research Objectives and Outline of Paper
The overarching goal of this study is to develop a standardized method for quantitatively assessing the effectiveness of the 1994 revised NFIP policy goal to limit and reduce development in floodplains. We do this by developing spatiotemporal indicators of population and economic exposure that leverage freely available national databases, thereby enabling monitoring for any county in the U.S. A Geographic Information Systems (GIS) based model was developed that quantifies the spatial distribution of people and property in relation to the 100-year floodplain both before and after the 1994 revised NFIP and the associated HMGP and FMA were enacted. The Temporal Flood Loss Exposure Model (T-FLEM) is developed as a tool that uses population and building tax value as socioeconomic indicators to examine development trends within floodplain boundaries. T-FLEM is applied in five North Carolina counties to quantitatively monitor whether flood policy (1) reduced exposure within the 100-year floodplain and (2) inadvertently created the potential for greater losses immediately outside the 100-year boundary.
In this paper, we first explore the theoretical framework of the model in terms of hazard exposure theory. We then develop the methodological approach of T-FLEM to refine the distribution of the socioeconomic indicators of exposure, and quantify the spatial and temporal changes in these indicators. The results from the application of T-FLEM are then used to assess flood policy effectiveness.
- Top of page
- Study Area
- Literature Cited
The study area consists of five North Carolina counties, each representing a distinct physiographic area: the Mountain Region, Piedmont, and Coastal Plain (Figure 1). The county scale was selected because FIRMs are produced at that scale and flood policy is enforced by county and municipal agencies (FEMA, 2002). The counties examined are Buncombe (Mountain Region); Orange, Durham, and Wake (Piedmont); and Craven (Coastal Plain). Three contiguous Piedmont counties were analyzed to examine how floodplain extents progress from headwaters (Orange) to the larger rivers of Durham and Wake. Furthermore, the Piedmont counties, particularly Wake County, are rapidly urbanizing with pressure to develop floodplains. General physical, social, and economic characteristics of each county are provided in Table 2. The percent area contained inside the 100-year floodplain for each county is reflective of its topography, with the Mountain Region having the smallest (3.6%) and the Coastal Plain having the largest percentage (22.9%) of area located within the floodplain.
|County||Major Watershed||1990 Population||2000 Population||Percent Change||2006 Building Tax Value (US$ Billion)||Average Annual Precipitation (mm)||Percent County in Floodplain|
- Top of page
- Study Area
- Literature Cited
Quantification of Risk
Natural hazards are at the juxtaposition of physical, social, and economic factors in space and time (Cutter et al., 2000). The concept of risk applied to a natural hazard has traditionally involved empirical research that quantitatively describes the frequency of natural events and their impacts on society by
where Hazard refers to the probability of an event and Exposure relates to potential losses, usually expressed in terms of mortality or economic value (Merz et al., 2004; Vatsa, 2004). In this study, the hazard component is the probability of experiencing a flood defined by the FIRMs 100-year and 500-year (1 and 0.2% occurrence probability, respectively) floodplain boundaries. Exposure is defined in two ways, social and economic. Human population counts within the specified floodplain boundary are used to indicate social exposure and building tax values are used to indicate economic exposure, together referred to as socioeconomic exposure. This definition of exposure does not account for elevation of properties and assumes all buildings are located at or below base flood elevation. Changes through time in socioeconomic exposure to the flood hazard (i.e., the quantity of people or property within the floodplain) reflect the enforcement of floodplain management strategies to limit or reduce development in the floodplain over that period of time. It should be qualified that other forms of floodplain management and mitigation strategies, such as building codes, free board, elevated structures, and flood proofing, are not examined in this study, and have the potential to reduce flood damages. Since flood depth is not addressed in this study, the total value of the property is assumed to be at risk for damage, and we are monitoring the maximum flood risk to population and property value. Thus, a foremost priority in the development of T-FLEM was to provide spatial and temporal quantification of the exposure of people and property within floodplain boundaries in order to quantitatively assess the effectiveness of the 1994 revised national flood policy in the study area.
Data Inputs for T-FLEM
A foremost priority in our model development was to enable for a consistent methodology to be developed, thus the use of existing national databases that were freely available. The FIRMs operating in the study area between 1990 and 2000 were obtained in spatial format from FEMA. The county FIRMs were updated between 1992 and 1996 in the study area, with individual FIRM panels inside the county being modified on occasion throughout the decade. Environmental Systems Research Institute's (ESRI's) Topological Integrated Geographic Encoding and Referencing (TIGER) data provided census population counts in spatial block boundaries for 1990 and 2000 (TIGER, 2006), providing a snapshot of the population inside the floodplain before and after the 1994 FMA. Freely available county parcel data were obtained, and while not a national database, it provides an example of how local data can be incorporated into T-FLEM. Lastly, the National Land Cover Dataset (NLCD) was freely available from the U.S. Geological Survey seamless website (USGS, 2006) for 1992 and 2001.
Automated Methodology of T-FLEM
T-FLEM was designed to automate data preparation, create socioeconomic distributions, and calculate floodplain exposure. Floodplain shapefiles contain the spatial location of the 100-year and 500-year floodplain boundary. The area between the 100-year and 500-year floodplain boundary is hereafter refered to as the “marginal 500-year floodplain” (Figure 2). The marginal 500-year floodplain was used to assess development immediately adjacent to the 100-year floodplain (see section Edge Effects of the 100-Year Floodplain Boundary above). This area has a known hazard (0.2% to 1%), but any development in this area lies outside of national floodplain policy jurisdiction. We use the marginal 500-year floodplain as a proxy for the uncertainty in the delineation of the 100-year boundary because the difference in elevation usually varies on the order of decimeters, which is less than the uncertainty in floodplain delineation (Table 1).
Population and building tax values, our indicators of social and economic exposure, were originally represented in a GIS as discrete vector polygons. Such polygons contain large uncertainty in how the population or building tax value is spatially distributed inside each individual census block (population) or parcel (building tax). Vector representation is problematic when only a portion of a shapefile intersects the floodplain boundary, because it is not possible to discern how much of the population or building tax value is located inside the floodplain (Figure 3).
There are two common approaches to address this problem in exposure analysis (Bhaduri et al., 2007). First, the area weighted approach assumes that population and building tax values are uniformly distributed throughout the block or parcel. Thus, the portion of the block or parcel inside the floodplain boundary is the portion of the population or building tax value in the floodplain. For example, if 25% of the area of the census block is located inside the floodplain, then 25% of the population in that census block is assumed to be inside the floodplain (Figures 3A and 3B). In this figure, 11% of census block 1135 was located in the 100-year floodplain; therefore, 11% of the 52 people in that block, six people, are estimated to be in the 100-year floodplain. Second, the centroid approach aggregates the entire socioeconomic value (i.e., population and building tax) onto the center of the polygon. If the centroid is located inside the floodplain, then the entire socioeconomic value of the block is counted as being inside the floodplain, but if the centroid is outside the floodplain, no socioeconomic value is counted. In the above example, the centroid of census block 1135 is located outside the 100-year floodplain, so it is estimated that none of the population is located in the 100-year floodplain (Figures 3A and 3B). If the centroid of that block had been located inside the floodplain, then all 52 people in the census block would be counted as located inside the floodplain.
Neither method is based on valid assumptions, as the centroid method places the entire population at a single point and the area weighted method assumes the population is uniformly distributed over the landscape. T-FLEM addressed this dilemma by using NLCD as an ancillary variable to distribute socioeconomic value at a finer resolution than the national level data. The error for population and building tax redistribution was constrained by the spatial boundaries of census blocks and parcels, respectively. The goal was not to create perfect socioeconomic surfaces, but to create a better distribution than national level vector data provided.
Creating Building Tax and Population Surfaces. T-FLEM is based on using NLCD to place more building tax value in developed pixels. Three key assumptions were that the NLCD (1) was reasonably accurate (2001 has 74% and 1992 has 64% classification accuracy; Khorram et al., 2000; Homer et al., 2004); (2) can be used to estimate land use; and (3) was temporally close to parcel and population data to be relevant. The 2001 NLCD was produced independently from the 1992 NLCD using newly available methods; however, the two datasets are considered to be reasonably compatible (Homer et al., 2004). The difference between the NLCD datasets is responsible for some of the changes in floodplain population values. Using the centroid and area weighted methods, which are independent of NLCD, supported T-FLEM trends of population change from 1990 to 2000 in the 100-year floodplain. The NLCD was reclassified by giving each land cover type a numeric value proportional to the likelihood of development (Table 3). The numeric value assumed that more buildings are located in developed areas (e.g., commercial and residential) than undeveloped areas (e.g., agriculture and wetlands). We were not aware of any previous research associating a weight between NLCD and building tax value, so a logarithmic weight scheme was used to place more building tax value in developed than undeveloped pixels.
|1992 NLCD||2001 NLCD||Building Tax||Population|
|11 – Water||11 – Water||0||0|
|21 – Low intensity residential||21 – Developed, open space||1,000||100|
|22 – High intensity residential||22 – Developed, low intensity||10,000||120|
|23 – Developed, medium intensity|
|23 – Commercial, industrial||24 – Developed, high intensity||10,000||10|
|31 – Bare rock/sand||31 – Barren land||1||1|
|32 – Quarries|
|33 – Transitional|
|41 – Deciduous forest||41 – Deciduous forest||100||3|
|42 – Evergreen forest||42 – Evergreen forest|
|43 – Mixed forest||43 – Mixed forest|
|51 – Shrub||51 – Shrub||1||2|
|61 – Orchards||71 – Grasslands||1||2|
|71 – Grasslands|
|81 – Pasture/hay||81 – Pasture/hay||100||1|
|82 – Row crops||82 – Cultivated crops|
|83 – Small grains|
|91 – Woody wetlands||90 – Woody wetlands|
|92 – Herbaceous wetlands||95 – Herbaceous wetlands||1||0|
The parcel shapefile was converted into a raster surface using the 2006 building tax values, and not the land tax value, since the focus on this paper was on the exposure of structures located in the 100-year floodplain. The reclassified NLCD (NLCDReclass Build) was divided by the sum of NLCDReclass Build pixels inside each respective parcel to get the percentage of building tax value attributed to each pixel (weighted coefficient file). The weighted coefficient file was multiplied by the value of the building tax in each parcel (Building Parcel Value) to create the Building Tax Surface.
The population surface was created following the same steps as for building tax, but by using census data and by developing alternative numeric values for the NLCD. The 1990 population surface was created using 1990 census blocks and 1992 NLCD. The 2000 population surface was created using 2000 census blocks and 2001 NLCD. The values used to reclassify the NLCD for the population placed the majority of people in residential land cover types (weights of 10, 100, and 120 in Table 3). Agricultural and forested areas were given some weight since these are rural areas with smaller populations. Lastly, population was not permitted to be located in water or wetlands, and these land cover types were given a weight of zero (Figures 3C and 3B).
While there were no studies that linked building tax to land cover type, Oak Ridge National Laboratory had developed a residential population model known as LandScan USA that used the NLCD to partially inform population spatial distributions. We used the residential population reclassified NLCD values in LandScan USA for our NLCD population values in T-FLEM (Table 3). LandScan USA is a 90 m resolution, dasymetric national population distribution model for both residential and daytime populations (Bhaduri et al., 2007). The spatiotemporal distribution of the population in LandScan USA is based on national level data such as NLCD, roads, topography, schools, and businesses. LandScan USA is one of the few population distribution models with quantified uncertainty (Qiang et al., 2006;Sabesan et al., 2007). The model is not available for public use, and so was not used as the population surface in this project because floodplain managers would be unable to duplicate this study in their own area.
Intersecting Flood Hazards and Socioeconomic Surfaces. The socioeconomic surfaces were created at the same resolution as the NLCD (30 m). This resolution was too coarse for parcel data, resulting in some parcels being smaller than a single pixel. To account for the finer resolution vector data, the socioeconomic surfaces were disaggregated into 6 m pixels, which was equivalent to the smallest parcels. This finer resolution also provided greater detail regarding whether the socioeconomic value located along floodplain boundaries were inside or outside the floodplain. Zonal statistics were used to calculate the socioeconomic value inside the 100-year and marginal 500-year floodplain.
The 1990 and 2000 population surfaces were used as a before and after indicator of how social exposure changed since the 1994 mitigation policy was implemented. A paired t-test was used to compare the before (1990) to the after (2000) population exposure inside the 100-year floodplain. Similarly, we compared the population density (population per km2) and building tax density (dollars per km2) within the 100-year floodplain to the marginal 500-year floodplain using a paired t-test to examine changes in development densities adjacent to the 100-year floodplain.
Comparing T-FLEM to Other Methods
The value of T-FLEM to floodplain managers is dependent on the level of confidence that can be placed on the distribution model. T-FLEM building tax and population values were verified by calculating the difference between T-FLEM and the traditional vector methods (centroid and weighted area). Additionally, T-FLEMs population surface was compared with the residential LandScan USA population surface, which is considered to be at the forefront of population distribution models (Bhaduri et al., 2007). All comparisons were conducted as densities to enable statistical analyses across counties and floodplains.
- Top of page
- Study Area
- Literature Cited
Comparing T-FLEM to Other Methods
T-FLEM population estimates varied between 10% to 34% of LandScan USA estimates of the total population within the 100-year floodplain (Figure 4), and the difference between the methods was not statistically significant (paired t-test, p = 0.98). T-FLEM was also more closely correlated to the LandScan USA population estimates compared to the centroid and weighted-area methods (Table 4). The difference between T-FLEM and the other models and methods for estimating the population density inside the marginal 500-year floodplain was greater than the differences between methods within the 100-year floodplain; however, none of the differences were statistically significant (Table 4).
|(Pop/km2)||Difference From T-FLEM (Pop/km2)|
|T-FLEM||LandScan USA||Centroid||Percent Area|
|(US$ Millions/km2)||Difference From T-FLEM (US$ Millions/km2)|
|T-FLEM||LandScan USA||Centroid||Percent Area|
The building tax density estimates from T-FLEM were only compared to the centroid and weighted-area methods, as LandScan USA is population only. T-FLEM estimates for tax value were not significantly different than using centroid or weighted-area methods, and this applied to the 100-year and marginal 500-year floodplains (Table 4).
In all, T-FLEM produced population distributions that more closely resembled the distributions of higher quality data (LandScan USA) than the raw national vector data. Further, T-FLEM produced spatial distributions of building tax comparable with preexisting vector-based methods, and is likely a better estimate based on its finer resolution and the results shown from the population comparisons with LandScan USA. Hereafter we only present results of T-FLEM analysis.
Temporal Changes in Exposure Between 1990 and 2000
From 1990 to 2000, county population increased throughout the study area from 10.5% in Craven to 32.6% in Wake (Table 5, Figure 5). In all five counties, the population inside the 100-year floodplain increased less than the county-wide population increase. In Craven County, population decreased inside the 100-year floodplain by 6.5% between 1990 and 2000, despite the county population increasing by 10.5%. In Durham and Wake Counties, the 100-year floodplain population increase was 9.6 and 13.3%, respectively, or about half of the counties’ population increase. In Buncombe and Orange, the increase in 100-year floodplain population was comparable to the county population increase. To supplement these findings, the 30 m resolution population surface (T-FLEM prior to disaggregating to the 6 m resolution) for all of North Carolina found a significant difference in the 100-year floodplain population between 1990 and 2000 (p = 0.02, Table 6). The overall population in the 100-year floodplain increased in the Mountain Region and Piedmont, but decreased in the Coastal Plain.
|Total County||100-Year Floodplain||Marginal 500-Year Floodplain|
|FLEM||100 Floodplain Mean Population Density|
The percent increase in population inside the marginal 500-year floodplain between 1990 and 2000 was equal to the county population increase (Figure 5) in Durham, Wake, and Craven. In Buncombe and Orange, the percent increase in population in the marginal 500-year floodplain was less than half of the percent increase in population that occurred in each respective county. The population change in the marginal 500-year floodplain was not statistically significant (p = 0.09, Table 5). Using the 30 m resolution to examine population change in the marginal 500-year floodplain for all of North Carolina showed little change in population from 1990 to 2000. Only the Piedmont had significant population increases in the marginal 500-year floodplain, which is the most rapidly growing region in North Carolina (Table 6).
Spatial Changes in Exposure Adjacent to the 100-Year Floodplain
Population and building tax density for each county increased in the marginal 500-year floodplain, i.e., immediately adjacent to the 100-year floodplain (Figure 6, Table 7). Orange County had the largest increase in population density (486%), while the remaining counties’ population density more than doubled from the 100-year to marginal 500-year floodplain. The large increase in Orange County is the result of the marginal 500-year FIRM being only delineated in Chapel Hill, a relatively small area with high socioeconomic value.
|Population Density (People/km2)||Building Tax Density (US$M/km2)|
|Total County||100-Year||Marginal 500-Year||p-Value||Total County||100-Year||Marginal 500-Year||p-Value|
The building tax density at least tripled in every county between the 100-year and marginal 500-year floodplain (p = 0.04). Both the population and building tax density in the 100-year floodplain were less than the county densities throughout the study area. However, population and building tax density in the marginal 500-year floodplain was greater than the county densities in Buncombe, Orange, and Craven Counties. With regards to the 30 m resolution population surface for the entire state, all regions (mountain, piedmont, and coastal) had a significant increase in population density adjacent to the 100-year floodplain, especially in the Coastal Plains (Table 8). Moreover, it is evident that the increase in density directly adjacent to the 100-year floodplain was established prior to the 1994 flood policy revisions, since the difference between the 100-year and marginal 500-year floodplain population densities were consistent between 1990 and 2000 (data not shown).
|2000||Mean Population Density (Population/km2)|
|Sample Size||100-Year||Marginal 500-Year||p-Value|
- Top of page
- Study Area
- Literature Cited
Temporal Monitoring to Assess Flood Policy Effectiveness
The key objective of a monitoring program is to determine the effectiveness and efficiency of a policy in meeting stated goals. Floodplain development monitoring programs need to be more frequent than they have been in the past, and based on a standardized methodology rather than case studies. For example, Burby (2002) found a 53% increase in the number of structures located on floodplains throughout the U.S. since 1968, but because of the lack of consistent monitoring, it is impossible to know how much of this change occurred after the 1994 flood policy focus on reducing floodplain development. While floodplain development has continued to occur after the 1994 policy was enacted (Pinter, 2005), without consistent monitoring of population and property, it will be impossible for policy makers to determine (1) how effective flood policy has been at reducing exposure (2) and where flood policy has been enforced, (3) or more importantly where it has not been enforced.
For evaluating flood policies, changes within a floodplain can be compared with changes in the surrounding area, e.g., the county within which flood policies are implemented. There are three alternatives for floodplain population changes through time, which indicate the relative success of flood policy implementation. First, if population growth in the floodplain was greater than in the county, then floodplains were more attractive to development, and thus flood policy implementation was not successful. Second, if the population growth in the floodplain was less than the county, then flood policies were effective at limiting floodplain development. Lastly, if floodplain population decreased despite an increase in county population, then flood policies were effective. These choices simplify development patterns and assume that floodplain management is the key driver for population change in the floodplain; however, population change is influenced by multiple factors such as: existing zoning, current land use, topography, preexisting infrastructure, housing markets, land availability in the floodplain, and the quality of land available for development.
Because T-FLEM was both accurate (Table 4) and based on widely available data, it enabled us to examine flood policy implementation across the five study counties using this simple categorization. Craven County had a decrease in population between 1990 and 2000 within the 100-year floodplain, despite overall county population growth, indicating that Craven County is enforcing flood policies. The population change in the remaining counties showed 100-year floodplain population growth that was less than the overall county growth, although Orange and Buncombe appeared to be less effective than Durham and Wake Counties at limiting development in the floodplain (Table 5 and Figure 5).
Based on these county-specific trends, we were also interested in how effective flood policies had been at a broader geographic scale. We applied T-FLEM using 30 m resolution population surfaces, which we were able to create for all of North Carolina. These data showed that flood policy effectiveness generally followed topographic trends, with the Coastal Plain Counties enforcing flood policies more effectively than the Mountain and Piedmont Counties (Table 6). For the Coastal Plain, complete destruction of over 4,000 homes following Hurricane Floyd in 1999 reduced the population in the floodplain and increased the likelihood of strictly enforcing floodplain management strategies such as limiting development in the floodplain. For example, after Floyd 1,888 housing units were approved for buyout and 2,534 additional homes were under review (ncfloodmaps, 2006), which would reduce the number of structures in the 100-year floodplain through buyout programs (Fraser et al., 2006). In addition, Coastal Plain streams on average overflow their banks five times a year, while streams in the Piedmont and Mountain Region typically only reach bankfull conditions once every 1.5 to 2 years (Sweet and Geratz, 2003). The additional exposure to flooding and hurricanes might provide the incentives necessary to enforce flood policy to the point of reducing exposure inside the 100-year floodplain.
Spatial Monitoring of Development to Assess Flood Policy Effectiveness
An important unintended consequence of the distinct 100-year floodplain boundary is the potential for development directly outside this boundary. The people and properties immediately outside the 100-year floodplain are not affected by flood policy, even though the probability of being flooded is only slightly less than those just inside the 100-year boundary. James (2004) suggested that residential development in the U.S. is shifting to locations adjacent to the 100-year floodplain, and that the 1% flood occurs more frequently than expected. Thus, people are moving into hazardous areas without the requirement or benefits of falling under flood policies.
Our study showed that this pattern is occurring in North Carolina as well. For the five study counties, socioeconomic density immediately adjacent to the 100-year floodplain boundary, increased by more than 180% (Figure 6, Table 7). This indicates that flood exposure has increased adjacent to the 100-year floodplain. Moreover, this pattern was consistent throughout all topographic regions of North Carolina (Table 8).
Increasing socioeconomic density adjacent to the 100-year floodplain is likely a product of the clear-cut representation of the 100-year boundary in FIRMs and the lack of flood policy for the marginal 500-year floodplain. In reality, the 100-year floodplain is both dynamic and fraught with uncertainty (see Edge Effects of the 100-Year Floodplain Boundary above). Thus, a more precautionary approach would be to depict the 100-year floodplain boundary as a continuum of hazardous condition rather than a distinct, binary condition. The distinctly depicted and enforced boundary has had the unintended consequence of encouraging development immediately adjacent to the 100-year floodplain, and thus when flooding has occurred outside the 100-year floodplain, the impacts have been catastrophic (Dorman and Bakolia, 2002). Moreover, the increase in development density within the marginal 500-year floodplain could potentially equate to significant future flood losses as the 100-year floodplain shifts in response to climate change, urbanization, and other factors (IFMRC, 1994; Pielke and Downton, 2000). Several state floodplain association managers have suggested using the 500-year boundary as the extent to which critical infrastructure cannot be developed (Bourget and Bailey, 2004; Robinson, 2004), although this has been met with resistance.
One alternative approach for adapting current flood policies would be to use the marginal 500-year floodplain as a proxy for the uncertainty in extent of the 100-year floodplain. Another solution with less monetary changes would be to display uncertainty on the newly developed DFIRMs. While uncertainty does not negate a model, it should be acknowledged and displayed, especially when the developed DFIRMs have the capability of being cheaply generated and illustrated as a range of flood probability from the 100 to 500-year floodplain (Smemoe et al., 2007). The institutionalization of flood probabilities displayed on DFIRMs can provide managers with an increased confidence in flood maps and a sense of the social and economic exposure at different flood magnitudes.
- Top of page
- Study Area
- Literature Cited
The focus of monitoring is to measure changes of quantifiable indicators that relate to the goals of flood policy, and to use this feedback to modify or create more efficient and cost-effective policies (Allan et al., 2008). This study created a model, T-FLEM, which was automated to allow consistent, rapid calculation of socioeconomic exposure based on their spatial relationship with the floodplain. The data and methodology enable local managers to assess the effectiveness of floodplain management at the county scale to advise policy makers to implement successful policy changes (Figure 7). While county level exploratory analysis does not allow for a conclusive explanation of why floodplain exposure changed as it did through time, T-FLEM was designed to be run by local floodplain managers with the assumption that county officials have the contextual background to understand the “why” behind model output.
This study applied the T-FLEM methodology to five North Carolina counties and showed that current flood policy is limiting development on the floodplains, but its implementation is only reducing floodplain development in a single county (Craven). While current flood policy is limiting potential flood losses in most counties, it has inadvertently increased the potential for future flood losses immediately adjacent to the 100-year floodplain. The unintended consequence was driven by the lack of regulations outside the 100-year floodplain, and the lack of consideration of uncertainty in the 100-year boundary.
Based on our results, we recommend that (1) floodplain management agencies implement consistent monitoring programs to quantitatively assess development in and adjacent to floodplains, (2) flood insurance and mitigation regulations and cost shares be encouraged and made available for properties within the 500-year floodplain, (3) new DFIRMs illustrate not only the 1% probability, but at least the 0.2%, and ideally all flood probabilities, and (4) new DFIRMs display some estimate of uncertainty by including potential future conditions.
- Top of page
- Study Area
- Literature Cited
This study was made possible by University of North Carolina’s Graduate School University Merit Assistantship, a National Science Foundation Graduate Research Fellowship, and the UNC Institute for the Environment. We appreciate the comments and guidance provided by Larry Band and Jim Fraser. We would like to thank Oak Ridge National Laboratory, particularly Dr. Budhendra Bhaduri, for allowing the on-site use of LandScan USA for model assessment. Three anonymous reviewers and the editors provided insightful comments and suggestions that improved this paper. This material is based upon work supported under a National Science Foundation Graduate Research Fellowship. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
- Top of page
- Study Area
- Literature Cited
- 2006. The Impacts of Climate Change on the Risk of Natural Disasters Disasters 30(1):5-18. ,
- 2008. Adaptive Management and Watersheds: A Social Science Perspective. Journal of the American Water Resources Association 44(1):166-174. , , , and ,
- 2004. Flood Risk Assessment and Associated Uncertainty. Natural Hazards and Earth System Sciences 4:295-308. , , and ,
- 2004. Multicriteria Evaluation of Flood Control Measures: The Case of Ribeira de Livramento. Water Resources Management 18:263-283. , , and ,
- 2006. Flood Disasters: Lessons From the Past – Worries for the Future, Geoscience Research Group. M.R. Company, Munich, pp. 1-6. ,
- 2007. LandScan USA: A High Resolution Geospatial and Temporal Modeling Approach for Population Distribution and Dynamics. Geo Journal 69:103-117. , , , and ,
- 2004. Binational Perspectives on Flood Risk. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 35-36. and ,
- 2002. Flood Insurance and Floodplain Management: The U.S. Experience. Journal of Environmental Hazards 3(3): 1-30. ,
- 2005. An Examination of Flood Damage Data Trends in the United States. Journal of Contemporary Water Research and Education 130:20-25. ,
- 2004. Towards an Understanding of the Root Causes of Forced Migration: The Political Economy of “Natural” Disasters. http://www.web.mit.edu/cis/www/migration/pubs/rrwp/25_towards.doc , accessed September 2006, pp. 1-26. and ,
- CRED (Centre for Research on the Epidemiology), 2008. Disaster Data: A Balanced Perspective. CRED Crunch 11:1-2.
- 2005. Are Natural Hazards and Disaster Losses in the US Increasing? EOS 86(41) 381, 388-389. and ,
- 2000. Revealing the Vulnerability of People and Places: A Case Study of Georgetown County, South Carolina. Annals of the Association of American Geographers 90(4):713-737. , , and ,
- 2002. North Carolina Floodplain Mapping Program. Center for Geographic Information Analysis, Raleigh, North Carolina, 7 p. and ,
- FEMA (Federal Emergency Management Agency), 1999. Hazard Mitigation Grand Program Desk Reference. Washington, D.C. http://www.fema.gov/library/viewRecord.do?id=1472 , accessed June 2008.
- FEMA (Federal Emergency Management Agency), 2002. National Flood Insurance Program: Program Description. Washington, D.C. http://www.fema.gov/library/viewRecord.do?id=1480 , accessed April 2007.
- FEMA (Federal Emergency Management Agency), 2006. A Chronology of Major Events Affecting the National Flood Insurance Program. Completed for FEMA under contract by the American Institute for Research, December 2005 under Contract No. 282-98-0029.
- 2006. Mitigating Repetitive Loss Properties. Final Report Prepared for the URS and FEMA. The Center for Urban and Regional Studies, Chapel Hill, North Carolina, 43 p. , , and ,
- GAO, 2004. National Flood Insurance Program: Actions to Address Repetitive Loss Properties. GAO Report GAO-04-401T, Washington, D.C.
- GAO, 2005. National Flood Insurance Program: Oversight of Policy Issuance and Claim. GAO Report GAO-05-532T, Washington, D.C.
- 1999. Natural Hazard Mitigation: Recasting Disaster Policy and Planning. Island Press, Washington, D.C., 591 p. , , , , and ,
- 2004. What Does the 1% Flood Standard Mean? Revisiting the 100-Year Flood. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 117-119. , , and ,
- 2004. Development of a 2001 National Land-Cover Database for the United States. Photogrammetry Engineering Remote Sensing 70(7):829-840. , , , , and ,
- IFMRC (Interagency Floodplain Management Review Committee), 1994. Sharing the Challenge: Floodplain Management in to the 21st Century. U.S. Government Printing Office, Washington, D.C., 189 p.
- 2004. Objective Guidance of Floodplain Use. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 43-45. ,
- 2000. Issues Involved in the Accuracy Assessment of Large Scale Land Use/Land Cover Mapping and Monitoring From Remotely Sensed Data. Geoscience and Remote Sensing Symposium, IGARSS 2000 Proceedings; IEEE 2000 International, Center for Earth Observations, Box 7106, North Carolina State University, Raleigh, NC 27695. , , , , and ,
- 2001. Tsunami Project: The Uninsured Elements of Natural Catastrophic Losses: Seven Case Studies of Earthquake and Flood Disasters Summary Report. 62 p. http://www.nottingham.ac.uk/business/cris/ukec/2003%20joanne%20bayer%20paper.pdf , accessed November, 2006. , , , and ,
- 2004. Are We Really Mapping/Managing the 1% Chance Floodplain? In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 77-79.
- 2004. Estimation Uncertainty of Direct Monetary Flood Damage to Buildings. Natural Hazards and Earth System Sciences 4:153-163. , , , and ,
- 1999. Disasters by Design. John Henry Press, Washington, D.C. ,
- 2006. Urban Disasters as Indicators of Global Environmental Change: Assessing Functional Varieties of Vulnerability. In : Earth System Science in the Anthropocene, E.Ehlers, and T.Kraft (Editors). Springer-Verlag, Berlin, Germany, pp. 135-152. ,
- NCDEM (North Carolina Department of Emergency Management), 2007. Hazard Mitigation in North Carolina: Measuring Success. http://www.p2pays.org/ref/14/13619.htm , accessed June 2007, 93 p.
- NC Floodmaps (North Carolina Floodmaps), 2006. North Carolina Floodplain Hurricane Floyd and 10-Year Disaster Assistance Report. http://www.ncfloodmaps.com/pubdocs/historicadata.htm#Housing, accessed November 2006.
- 2006. Climate Change and Disaster Management. Disasters 30(1):64-80. , , , and ,
- 2000. Precipitation and Damaging Floods: Trends in the United States, 1932-1997. Journal of Climate 13:3625-3637. and ,
- 2002. Flood Damage in the United States, 1926-2000: A Reanalysis of National Weather Service Estimates. National Center for Atmospheric Research, Boulder, Colorado, 96 p. , , and ,
- 2005. One Step Forward, Two Steps Back on US Floodplains. Science 308:207-208. ,
- 2006. Estimating Small-Area Populations by Age and Sex Using Spatial Interpolation and Statistical Inference Methods. Transactions in GIS 10(4):577-598. , , , , and ,
- 2003. The Effect of Lidar Posting Density on DEM Accuracy and Flood Extent Delineation. NASA Affiliated Research Center; University of South Carolina, Columbia, South Carolina, 33 p.
- 2004. Notes on Probability Analysis and Flood Frequency Standards. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 20-25.
- 2004. Issues and Perspectives on Floodplain Management. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 124-127. ,
- 2004. History of the 1% Chance Flood Standard. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 2-8. ,
- 2007. Metrics for the Comparative Analysis of Geospatial Datasets With Applications to High-Resolution Grid-Based Population Data. Geo Journal 69: 81-91. , , , , , and ,
- 2007. Demonstrating Floodplain Uncertainty Using Flood Probability Maps. Journal of the American Water Resources Association 43(2):359-371. , , , and ,
- 2003. Bankfull Hydraulic Geometry Relationships and Recurrence Intervals for North Carolina’s Coastal Plain. Journal of the American Water Resources Association 39(4):861-871. and ,
- 2004. Will the Data Support Modeling for a New Standard? In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 54-56. and ,
- TIGER (Topologically Integrated Geographic Encoding and Referencing), 2006. U.S. Census Bureau, Geography Division. http://www.census.gov/geo/www/tiger/tiger2k/tgr2000.html, accessed October 2006.
- 2004. Evaluation of the National Flood Insurance Program. In : Reducing Flood Losses: Is the 1% Chance (100-Year) Flood Standard Sufficient? 2004 Assembly of the Gilbert F. White National Flood Policy Forum. National Academies Keck Center, Washington, D.C., pp. 46-48. ,
- USGS (United States Geological Survey), 2006. Seamless Data Distribution System, Earth Resources Observation and Science. http://www.seamless.usgs.gov/ , accessed October 2006.
- 2000. Analysis of the Magnitude and Frequency of Floods in Colorado, Water Resources Investigations Report 99-4190. U.S. Geological Survey, Denver, Colorado. ,
- 2004. Risk, Vulnerability, and Asset-Based Approach to Disaster Risk Management. The International Journal of Sociology and Social Policy 24(10/11): 1-48. ,