Inequities in the distribution of flood risk under floodplain restoration and climate change scenarios
Handling Editor: Shuai Wang
Abstract
- The combined impacts of climate change and ecological degradation are expected to worsen inequality within society. These dynamics are exemplified by increases in flood risk globally. In general, low-income and socially vulnerable populations disproportionately bear the cost of flood damages. Climate change is expected to increase the number of people exposed to fluvial flood risk and cause greater property damages. Floodplain restoration has the potential to mitigate these impacts, but the distribution of future risks among different types of property owners under these altered conditions is often unknown.
- Here, we develop a simple probabilistic approach for estimating flood risk to property owners under floodplain restoration and climate change scenarios for a range of flood recurrence intervals. We apply this approach in the Vermont, USA portion of the Lake Champlain Basin.
- Over a 100-year time horizon, we estimate that the value of property damages caused by flood inundation is approximately $2.13 billion under the baseline scenario. Climate change is expected to increase damages to $5.29 billion, a 148% increase; however, floodplain restoration has the potential to reduce these impacts by approximately 20%.
- For all scenarios, a larger proportion of lower-value properties, specifically mobile homes, face greater flood risk compared to higher-value properties. Climate change is expected to cost higher-value properties and commercial properties more than other types of properties, but these same groups are also expected to benefit most from floodplain restoration.
- In general, these results raise concern that those least able to prepare for and recover from flood damages are also the people who face the greatest threats. In response, public policy interventions must consider not only where flood risk is most severe, but also the vulnerability of people exposed to such risk.
Read the free Plain Language Summary for this article on the Journal blog.
1 INTRODUCTION
Interactions between ecological degradation and climate change threaten to exacerbate social and economic inequality globally (IPBES, 2019; UNEP, 2021). Increases in flood risk epitomize the impacts of these interacting global changes on human well-being. Flooding is the most widely experienced, deadliest and costliest natural hazard in the United States (US) and globally (Miller et al., 2008; Perry, 2000). Flood risk is not equally distributed across socioeconomic and demographic groups. In general, those who are most vulnerable and least resilient to natural hazards are disproportionately exposed to flood hazards (Chakraborty et al., 2014; Collins et al., 2019). In the US, these groups include racial and ethnic minorities, low-income households and mobile home owners (Tate et al., 2021).
Households considered to be socially vulnerable are less likely to be prepared for disasters (Phillips et al., 2005), are less capable of responding to imminent flood events and have less capacity to recover after damages, and displacement occurs (Cutter et al., 2003). In particular, income and wealth are major determinants of households’ ability to respond to and recover from flooding. These households typically have limited access to transportation, lower savings rates, less insurance coverage and are often further disadvantaged by deficiencies in materials used to build their homes (Baker et al., 2011).
Due to climate change, heavy rainfall events are expected to occur with greater severity and frequency in coming years (IPCC, 2021), increasing fluvial flood risk globally (Field, 2012; Hirabayashi et al., 2013). Already climate-change-induced increases in extreme precipitation events have contributed one-third of flood damages incurred in the US between 1988 and 2017, with a cumulative impact of $73 billion (Davenport et al., 2021). Under future climate change in the US, the area of inundation from 100-year flood events is expected to increase by up to 16% in the northeast, southeast and western regions of the country (Bates et al., 2020), and the population exposed to inundation is predicted to increase by 30%–127% (Swain et al., 2020). These changes in flood hazard and exposure translate to a roughly 30%, or $747 million, increase in annual flood-related damages nationally (Wobus et al., 2014).
The loss and degradation of forest and wetland ecosystems has only compounded climate-induced increases in flood risk (Wheater & Evans, 2009). As of 2009, 7.2 million km2 of wetlands had been lost globally (Hu et al., 2017), and between 2000 and 2012, 2.3 million km2 of forest were lost (Hansen et al., 2013). These declines have been primarily driven by increased development, agricultural expansion and wildfires (Asselen et al., 2013; Curtis et al., 2018). The loss of wetlands and other riparian ecosystems has considerably increased flood damages, through diminished flood attenuation and storage (Gulbin et al., 2019; Li et al., 2018; Taylor & Druckenmiller, 2021). Similarly, Bradshaw et al. (2007) find that a 10% loss in forest cover is associated with a 4%–28% increase in flood frequency and a 4%–8% increase in total flood duration. Together, these trends highlight that land-use change will play a major role in mediating future flood risk.
In response to growing concerns about the interacting effects of climate change and ecological degradation on flood hazards, public and private institutions are directing greater investment towards interventions designed to mitigate flood damages. Efforts to protect properties from flooding have historically focused on structural interventions (i.e. grey infrastructure), such as dams and levees. However, protection and restoration of natural floodplain ecosystems is increasingly being considered as a complementary strategy to reduce negative impacts of flooding (Hudson & Middelkoop, 2015; Pudar et al., 2020). Compared to traditional forms of grey infrastructure, floodplain ecosystems can be less expensive to maintain and more resilient to catastrophic flood events (Daigneault et al., 2016). Recent studies show that natural ecosystems’ capacity to mitigate flood damages, through the attenuation and storage of floodwaters, provides significant economic value to property owners (Narayan et al., 2017; Taylor & Druckenmiller, 2021; Watson et al., 2016). Likewise, the reduction in flood damages by avoiding development in floodplains far exceeds the costs of acquiring and conserving natural land in these areas (Johnson et al., 2020).
Floodplain restoration interventions may include revegetation, bank reshaping and wetland construction, among other strategies (Bernhardt et al., 2005). By increasing the landscape roughness, revegetation attenuates floods, decreasing inundation downstream (Hupp & Osterkamp, 1996). Natural floodplain ecosystems may also provide a suite of valuable co-benefits, such as carbon sequestration, water purification, riparian habitat and recreational opportunities (Kousky & Walls, 2014; Perry et al., 2015). The effects of these interventions, however, are spatially heterogeneous and depend on the local topography and geomorphic characteristics within watersheds (Singh et al., 2018; Ward et al., 2001).
Understanding how interactions between climate change and floodplain restoration may change the distribution of flood risk to different types of property owners remains a critical gap. While impacts of climate change on flood risk will likely be unevenly distributed among property owners (Batibeniz et al., 2020; Mills et al., 2018), floodplain restoration has the potential to either mitigate or exacerbate such inequities (Gourevitch et al., 2020). As such, disaggregating the distribution of flood risk, and evaluating how that distribution changes under alternative scenarios, is critical to making natural hazard mitigation and climate change adaption more equitable.
Previous studies have modelled the potential impacts of floodplain restoration (e.g. Dixon et al., 2016; Singh et al., 2018; Wobus et al., 2019) and climate change scenarios (e.g. Bates et al., 2020; Swain et al., 2020) on flood inundation, but rarely do they evaluate the distribution of flood damages to different groups under these alternative scenarios. Other recent studies have evaluated the distribution of flood risk among demographic and socioeconomic groups based on historical flood events, FEMA flood maps or other static inundation models, but do not consider how alternative scenarios alter risk profiles (e.g. Montgomery & Chakraborty, 2015; Tate et al., 2021). These studies also typically evaluate risk for aggregate spatial units, such as census tracts, and thereby mask the distribution of flood damages among groups of people within these spatial units (Maantay & Maroko, 2009).
We address these gaps by assessing the distribution of flood risk to individual properties under several floodplain restoration and climate change scenarios at the basin scale. Using a low complexity, probabilistic approach, we map flood inundation depths and extent under a range of flood events (Diehl et al., 2021). For each scenario, we evaluate the exposure of properties, the damage to each property for each event as a percentage of its appraised value and the net present value of damages to each property over a 100-year time horizon (Figure 1). We apply this approach in the Vermont portion of the Lake Champlain Basin (LCB) for all rivers with catchments >10 mi2 (Figure 2), a scale relevant to typical policy decisions in the region. This study is among the first to integrate a flood model and property depth-damage functions to evaluate the distribution of flood risk among different types of property owners under floodplain restoration and climate change scenarios. This advancement is critical for policymakers and floodplain managers who require information on the equity of changes in flood risk. Although the scope of our study is limited to Vermont, the framework we develop is generalizable and applicable to a wide range of other regions.
2 METHODS
2.1 Study context
The LCB encompasses parts of Vermont, New York and Quebec. The focus of this study is the Vermont portion of the basin. The area of this part of the basin is approximately 18,300 km2 and the population is 422,000. Within the basin, five main rivers drain into Lake Champlin: the Lamoille River, Mettawee River, Missisquoi River, Otter Creek and Winooski River (Figure 2). Among these rivers, the Winooski River watershed is the largest in both area (2,750 km2) and population (143,000 people). The upland areas of the basin are predominantly forested, while the lowland areas are mostly agricultural.
Under climate change, flooding in the northeastern US is expected to become increasingly frequent and more severe (Marsooli et al., 2019; Siddique & Palmer, 2020). Population growth, increased impervious surface area, floodplain development, stream channel incision and installation of undersized culverts over the past several decades have also contributed to increased flood hazards and exposure across the region (Mears & McKearnan, 2012). Most recently, in the state of Vermont, Tropical Storm Irene caused over $1 billion in flood-related damages in 2011 (Galford et al., 2014). The storm displaced over 1,500 families and damaged more than 3,500 residential properties, including 500 mobile homes (Baker et al., 2014).
During Tropical Storm Irene and other flood events of the 20th century in Vermont, existing natural ecosystems and green infrastructure significantly reduced property damages from inundation (Watson et al., 2016). In recognition of this, conservation organizations and state government agencies are increasingly considering the role of nature-based solutions in flood mitigation. To meet this need, these organizations require more detailed information on (a) the location and magnitude of damages caused by flood inundation, (b) the extent to which floodplain restoration and climate change may impact these damages and (c) the distribution of damages among property owners.
2.2 Mapping flood inundation
To map flood inundation depth and extent under a range of recurrence intervals (i.e. annual exceedance probabilities), we used the probHAND model, developed by Diehl et al. (2021). This model is based on the height above nearest drainage (HAND)-synthetic rating curve (SRC) approach, developed and adopted by others for rapid identification of flood risk hazards over large scales (Zheng et al., 2018). In recent years, the HAND-SRC approach has been widely applied in the US, Canada, Brazil and India (Bhatt & Rao, 2018; Chaudhuri et al., 2021; Johnson et al., 2019; Nobre et al., 2011; Speckhann et al., 2018). Building on the HAND-SRC approach, the probHAND model adds an uncertainty analysis to the identification of flood depths for specified flood frequencies at the reach scale (10–100 km). HAND maps, indicative of the relative elevation of each point on the landscape above the river channel (e.g. Figure 3a), are developed from high-resolution digital elevation models (DEMs), using a suite of D-infinity routing algorithms (Rennó et al., 2008). Empirically derived stage-discharge rating curves are then calculated on a reach-by-reach basis, using HAND-based elevations and the Manning's equation, relating channel geometry and roughness to discharge (Zheng et al., 2018). From corresponding flood frequency curves, and distributions of uncertainties in model parameters, the probHAND model then identifies the HAND elevation associated with a given recurrence interval flood for each reach. By intersecting the HAND elevation with the original DEM, the probHAND model produces raster maps of inundation depth and extent.
Spatial data inputs to the probHAND model for this study included 1 m resolution LiDAR-derived DEMs, NHDplus-defined reaches where the average drainage area within the reach was >10 mi2, and a Vermont-specific 2016 land cover dataset (O’Neil-Dunne, 2018). All raster-based spatial inputs were resampled to a 10 m resolution to enhance computational efficiency. We associated Manning's n coefficients with each land cover class, based on the calibration results from Trueheart et al. (2020). We also obtained peak discharge data for each recurrence interval for each reach from the USGS StreamStats application (Ries III et al., 2017). For Vermont, this information is derived from regression equations relating drainage area, number of lakes and ponds in the upstream catchment, proportion of basin higher than 365 m in elevation, and the geographical coordinates of the basin centroid (Olson & Veilleux, 2014).
The probHAND model was evaluated using 42 spatially robust field documented high-water marks from a large flood event in the Winooski River watershed. In comparing the observed data with the modelled outputs for flood events with comparable peak discharge values, Diehl et al. (2021) found approximately 90% of the observed high-water marks were captured by predicted flood extent when accounting for the uncertainty in these predictions. Additionally, the probHAND model was also evaluated using a more temporally rich dataset of inundation extents predicted by a 1D hydrodynamic model (HEC-RAS), which represents a more physically based representation of inundation patterns (Diehl et al., 2021). The probHAND predictions most closely matched other modelled flood extents for larger floods; 72%–82% overlap for floods with a recurrence interval between 10 and 500 year floods.
2.3 Alternative scenarios
This study uses the probHAND model to evaluate the impact of climate change and floodplain restoration on flood inundation, relative to baseline conditions. To model the impacts of climate change on flood inundation, we increased the discharge associated with each flood recurrence interval, for a given reach, by 80% (e.g. Figure 3d; see Q vs. Q′), as suggested by the Vermont Climate Assessment (Galford et al., 2014). For example, the peak discharge historically associated with a 100-year flood for Reach A is 5,000 m3/s; under climate change, we assume that the discharge associated with a 100-year flood is 9,000 m3/s.
To simulate floodplain restoration, we converted cropland, pasture, hay and barren land cover classes to deciduous forest (Figure 3b). We implemented this scenario by modifying the Manning's n coefficients associated with each land cover class, similar to Singh et al. (2018). For example, we converted any pasture cells, represented by n = 0.06, to deciduous forest, represented by n = 0.13. When Manning's n is increased, the rating curves for that reach gets pushed outward, thereby increasing the stage associated with a given discharge (Figure 3d, see ‘baseline rating curve’ vs. ‘restored rating curve’). The model then propagates these restoration-induced changes in stage downstream via dynamic reductions in discharge, corresponding to the percent increase in upstream stage. This approach was designed to estimate the upper bound potential of floodplain restoration through reforestation and to highlight the potential distributional impacts of these types of interventions. While spatial optimization of restoration interventions is also important, it is not computationally feasible with this framework due to the high spatial resolution and large spatial extent of the model domain.
To capture the downstream impacts of floodplain restoration, we added a dynamic module to the probHAND model, allowing downstream propagation of changes in streamflow. Because the original probHAND model was developed to identify flood extents for a discrete point in time (i.e. at the flood peak), calculations are performed for each reach independent of neighbouring reaches. The impacts of altering land cover on streamflow, however, often have downstream consequences (Thomas & Nisbet, 2007). The approach we develop here, based on the logic of other hydrologic inundation models (Dixon et al., 2016; Sholtes & Doyle, 2011), propagates the scenario-driven changes in stage, from one reach to the next, starting at the top of the watershed. Change in predicted flood stage within each reach is quantified as the difference between the baseline stage and the average stage at the entrance and exit of the reach, under an alternative scenario. When the average Manning's coefficient is increased, flood stage within a reach increases due to slower travel times in flood wave propagation (Figure 3c, see t0 vs. t1). By assuming that there are no changes in flood volume as a result of restoration, longer travel times result in decreased discharge (Figure 3c, see Q0 vs. Q1). Changes in discharge are subsequently associated with new stage values (Figure 3d, see S vs. S′).
2.4 Uncertainty analysis
As part of the probHAND approach, we used a Monte Carlo simulation to estimate uncertainty in inundation depth and extent across all recurrence intervals (Diehl et al., 2021). For each iteration of the Monte Carlo simulation (N = 1,000), we randomly sampled normal and truncated normal probability distribution functions (PDFs) fitted to each of the input parameters in the Manning's equation (i.e. hydraulic geometry, roughness coefficient, energy grade slope and peak discharge). To define PDFs for the parameter uncertainty, we compared the calculated baseline values to values derived from other hydraulic or regression models or from measured field data. See Diehl et al. (2021) for more information on the process for fitting each of the probability distribution functions. Using the outputs from the Monte Carlo simulation, we calculated the uncertainty in inundation depths and associated damages to each property in the basin. In our results, we typically report the mean of the outputs from the Monte Carlo simulation.
2.5 Estimating damages
We estimated inundation-related damages to residential and commercial properties using depth-damage functions specified by the US Federal Emergency Management Agency (FEMA) guidelines. These functions were developed using past flood insurance claims collected by the National Flood Insurance Program (NFIP) and are used to estimate the damage to inundated properties as a percentage of their value (FEMA, 2003). While depth-damage functions are widely used in flood risk assessments nationally, including FEMA's HAZUS-MH software (Scawthorn et al., 2006), these functional relationships are highly uncertain (Wing et al., 2020).
For each scenario, we identified the flood depth for all properties within the basin by overlaying the flood inundation map output from the probHAND model with spatial locations of all properties (Figure 4a), derived from Vermont's E911 site location database (VCGI, 2013). We then calculated the percent damage to each inundated property using depth-damage functions specific to the type of property (i.e. single-family home, multi-family home, commercial property, mobile home; Figure 4b). The estimated percent property damage was converted to a monetary estimate of damages using the Vermont Grand List, a publicly available database of property tax records that includes recent appraisals of property values.
(1)
(2)
is the discount rate (Equation 3). We used a discount rate of 3%, which is consistent with the current recommendations made by the US Office of Management and Budget for cost–benefit analysis of public investments (U.S. Office of Management & Budget, 1992).
(3)2.6 Distributional impacts
We assessed the distribution of flood risk to properties under each scenario using multiple categorical groupings. Based on spatial location and property characteristics included in the E911 dataset, we binned properties based on their HUC8 watershed location, property value and property type. Property values were grouped by quintile distribution (Table 1). Property types include single-family homes, multi-family homes, mobile homes and commercial properties. We also disaggregated flood damages by flood recurrence interval.
| Quintile | Property value range |
|---|---|
| Lowest quintile | ≤$101,000 |
| 2nd quintile | $101,000–$163,000 |
| 3rd quintile | $163,000–$226,000 |
| 4th quintile | $226,000–$328,000 |
| Highest quintile | ≥$328,000 |
3 RESULTS
3.1 Damages across scenarios
We estimate that the NPV of damages under the baseline scenario (i.e. ‘business-as-usual’) is approximately $2.13 billion over a 100-year time horizon. Floodplain restoration, under historical climatic conditions, has the potential to reduce damages to $1.77 billion, a 17% decrease from the baseline. Climate change is expected to increase damages to $5.29 billion, a 148% increase over the baseline; however, floodplain restoration has the potential to reduce damages under climate change to $4.28 billion, a 19% decrease from the climate change scenario.
3.2 Damages across watersheds
Damages from flooding are not borne equally across watersheds. Under the baseline scenario, damages in the Winooski River watershed are approximately $1.26 billion, which is greater than damages in all other watersheds combined (Figure 5; top panel). This trend holds for the other scenarios as well. Compared to the other watersheds, the Winooski River watershed has the largest population, the highest median property values and is the largest watershed by area.
3.3 Damages across flood recurrence intervals
Flood damages are also primarily attributable to higher frequency, lower severity flood events (Figure 5; bottom panel). Under the baseline scenario, 2-year floods (i.e. 50% annual probability) are expected to cause $1.01 billion in property damage, 47% of total damages. As annual exceedance probabilities decrease, the NPV of damages gradually lessens. For example, 100-year floods are expected to cause $0.14 billion in damages, while 500-year floods are expected to cause $0.04 billion in damages.
3.4 Distribution of flood risk among property owners
Across all scenarios, mobile homes and lower-value properties are disproportionately exposed to flooding. Of all properties in the lowest quintile, 3.8% are located in the 500-year floodplain under the baseline scenario; by contrast, only 0.7% of all properties in the highest quintile are expected to experience any flood damage (Figure 6a). Similarly, under the baseline, 5.8% of all mobile homes are exposed to inundation during a 500-year flood event (Figure 6b). By comparison, approximately 1.9% of all multi-family and commercial properties are exposed to flooding. Although the percentage of exposed properties changes under the alternative scenarios, the ranking of which groups are most/least exposed does not.
When the monetary value of flood damages is disaggregated across the same property value quintiles and property types, we found the opposite trend. In aggregate, flood damages cost higher-value properties and commercial properties more than other property groups. Although properties may experience similar inundation depths, the value of damages highly depends on the value of the property. As such, the monetary value of flood damages is greatest for properties in the highest value quintile and commercial properties (Figure 6c,d).
Climate change is expected to most severely affect higher-value properties and commercial properties, but these same groups of properties are also expected to benefit most from floodplain restoration. Under the climate change scenario, 1.2% of properties in the highest value quintile and 5.7% of commercial properties are exposed to flood inundation, representing 69% and 51% increases in exposure from the baseline scenario, respectively (Figure 7). By contrast, 3.8% of properties in the lowest value quintile and 2.4% of mobile homes face exposure under climate change, representing 23% and 21% increases in exposure, respectively (Figure 7). Under the combined climate change and floodplain restoration scenario, 1.8% of properties in the highest value quintile and 5.4% of commercial properties are exposed to flood inundation, constituting 5.6% and 4.8% decreases in exposure, as compared to the climate change scenario (Figure 7). In comparison, properties in the lowest quintile and mobile homes are expected to experience 2.6% and 2.9% reductions in exposure (Figure 7).
4 DISCUSSION
In this study, we find that flood damages to residential and commercial properties in the Vermont portion of the LCB may exceed $2 billion over a 100-year time horizon under the baseline scenario. Climate change is expected to more than double damages, yet floodplain restoration has the potential to mitigate these impacts by roughly 20%. Across all scenarios, lower-value property and mobile homes owners are disproportionately exposed to flood inundation. Although climate change is expected to increase exposure most severely for higher-value properties and commercial properties, these properties are also expected to benefit most from floodplain restoration. Our findings complement widespread evidence that lower-income households, both in the US and globally, are disproportionately exposed to environmental hazards and will likely bear the greatest burden of damages from climate change (Hsiang et al., 2017; Ringquist, 2005; Srinivasan et al., 2008). If these disparities are ignored, the uneven distribution of flood risk may continue to reinforce inequality within society.
This study is among the first to directly link climate change and floodplain restoration scenarios with flood damages to individual properties, particularly at the basin scale. This linkage is critical to informing the consequences of alternative actions and evaluating trade-offs between policies. Previous studies have examined the impacts of alternative scenarios on flood inundation, but do not connect these outcomes with impacts to people (e.g. Dixon et al., 2016; Singh et al., 2018). Other studies have assessed the distribution of flood damages among socioeconomic and demographic groups, but do not consider how climate change and restoration scenarios might alter these distributions (e.g. Montgomery & Chakraborty, 2015; Tate et al., 2021). The novelty of our approach is facilitated by our use of a relatively simple GIS-based flood inundation model (Diehl et al., 2021), which was intentionally designed to evaluate alternative scenarios at large spatial extents. Despite some loss in complexity in our modelling of flood dynamics, we gain a flexibility and scalability not possible with other approaches.
Similar to Tariq (2013) and Wobus et al. (2019), we show that a large portion of flood damages are attributable to higher frequency, lower severity flood events (Figure 5b). This result is a function of two interacting factors: (a) a large proportion of properties exposed to any flood risk are located within the 2-year floodplain and are repeatedly inundated and (b) for many properties, the difference in expected damages between higher- and lower-frequency events is relatively small, due to the saturation of damages beyond 2 m of inundation (see Figure 4b). In the US, flood insurance pricing, zoning decisions and hazard mitigation benefit–cost analyses are often based on FEMA’s 100- and 500-year flood maps. Our finding, however, raises concern that the exclusive use of lower-frequency flood maps may lead to underestimation of expected annual damages and miscalculation of risk. Ongoing updates to FEMA flood maps present a critical opportunity to reconsider the likely significant role of higher-frequency events in flood risk assessment.
Our results also show that the choice of metric matters when evaluating the distribution of flood risk. When we consider the proportion of properties exposed to inundation, we found that mobile home owners and lower-value properties are disproportionately exposed to flood risk (Figure 6a,b). By contrast, when the total economic value of damages is used, commercial and higher-value properties are shown to face the greatest risks (Figure 6c,d). This disparity is largely a function of the implicit bias of monetary valuation when evaluating flood damages. For example, property A, a $100,000 home, and property B, a $1 million home, experience the same depth of inundation during a flood and the damages to the two properties as a proportion of their value is the same. However, while property owner A is likely to have less capacity to respond to and recover from the damages, the value of damages to property B is 10× greater than the value of damages to property A. This bias in monetary valuation is codified by FEMA’s hazard mitigation assessment methodologies, through their use of benefit–cost analysis (Rose et al., 2007). In the absence of equity weighting, these methodologies create perverse incentives in prioritizing flood mitigation interventions, whereby wealthier property owners often receive greatest protection (Frontuto et al., 2020; Kind et al., 2017).
While there are several other socioeconomic and demographic characteristics that influence social vulnerability to flooding, the scale and availability of these data limited the scope of our analysis. In Vermont, these socioeconomic and demographic variables, such as race/ethnicity, household composition and income/wealth, are not associated with property datasets to protect the privacy of property owners. These data are only available at the census tract level. Because population density in Vermont is relatively low, census tracts are relatively large. As such, floodplains comprise a small proportion of the area of census tracts. Projecting tract-level census data onto properties located within the floodplain would assume that owners of those properties are representative of all residents within the tract. This is not only a poor assumption, but would also mask the distribution of flood risk in Vermont.
Spatial targeting of floodplain restoration is an important area of future research. In our analysis, we restore all cropland, pasture, hay and barren land cover classes to deciduous forest. This approach was designed to estimate the upper bound potential of floodplain restoration through reforestation and to highlight the potential distributional impacts of these types of interventions. Due to our large spatial extent, high spatial resolution and use of a Monte Carlo simulation, consideration of additional more targeted restoration scenarios was not computationally feasible. However, given resource and budgetary constraints, government agencies and non-profit organizations across the state will be required to spatially prioritize their investments in restoration. In addition, floodplain restoration has the potential to provide a suite of other ecosystem service benefits, as well as habitat for biodiversity. In many restoration contexts, these co-benefits need to be balanced in accordance with the preferences of stakeholder groups. For this, multi-objective optimization tools can be useful for efficiently targeting the highest value restoration sites at the lowest cost, while illuminating trade-offs between competing objectives (Gourevitch et al., 2020).
In addition to considering the locations where the benefit–cost ratio will be greatest, prioritization schemes will need to incorporate metrics of social vulnerability. Harris County, Texas, where the City of Houston is located, offers a promising example of how to implement such prioritization. In the wake of Hurricane Harvey in 2017, Harris County passed a $2.5 billion bond to fund flood control projects. Instead of prioritizing projects based on flood hazard alone, that is, using the ‘worst-first’ principle, the commission responsible for allocating the funds decided to rank projects with consideration to the social vulnerability of the populations exposed to flood risk, using the Social Vulnerability Index developed by Centers for Disease Control and Prevention (Harris County Flood Control District, 2019). As compared to historical flood protection interventions, this alternative prioritization scheme effectively redistributes tax revenues from wealthier neighbourhoods to poorer neighbourhoods, thereby potentially ameliorating existing disparities in flood resilience.
As previously mentioned though, allocating resources based on social vulnerability in Vermont, and other rural areas, would require additional socioeconomic and demographic data at much finer spatial resolutions than what is currently available. Despite these limitations, other allocation methods, such as those based on means testing or equity-weighted utility functions, may have similar impacts as prioritization based on social vulnerability (Frontuto et al., 2020; Kind et al., 2017). Each of these methods has different advantages and limitations, and may benefit certain groups of people more than others. Better understanding the differences between these methods is key area of future research; however, implementing any of these is a likely improvement over the status quo.
5 CONCLUSIONS
This study demonstrates that although climate change is likely to result in substantial increases in flood damages, floodplain restoration has the potential to mitigate those impacts. Already, flood events are becoming more frequent and severe, and in response, policymakers are reconceptualizing how investments in hazard mitigation are prioritized. We argue here that floodplain restoration, as well as other nature-based solutions, has a critical role to play in any policy or management solution designed to build resilience to climate change.
Our findings also highlight the importance of considering the distribution of flood risk among different types of stakeholders. These results correspond with widespread evidence from other studies indicating that the impacts of climate change will disproportionately affect those who are socially vulnerable, thereby reinforcing inequality (Hsiang et al., 2017; Ringquist, 2005). By better understanding the distributional impacts of flooding, and allocating resources away from those who are more resilient to environmental hazards and towards those who need them most, flood mitigation interventions have the potential to reduce social and economic inequality.
ACKNOWLEDGEMENTS
We thank Stephanie Drago and Kristen Underwood for assistance with data acquisition and their contributions to the development of the probHAND model. We also thank the Gund Institute for Environment at the University of Vermont for providing institutional support. J.D.G was supported by the National Science Foundation under the Vermont EPSCoR Program [grant numbers EPS-1101317 and NSF OIA 1556770] and USDA McIntire-Stennis funding [grant number 2014-32100-06050] awarded to the University of Vermont. R.M.D and B.C.W. were supported by the Lake Champlain Basin Program. Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation, the Vermont EPSCoR Program, USDA McIntire-Stennis program or the Lake Champlain Basin Program.
CONFLICT OF INTEREST
The authors declare no conflicts of interests.
AUTHORS' CONTRIBUTIONS
J.D.G. and R.M.D. designed the study and acquired the data; J.D.G. conducted the analysis and drafted the manuscript; all authors assisted with revising and editing the manuscript; B.C.W. and T.H.R. supervised the project and provided administrative support.
Open Research
DATA AVAILABILITY STATEMENT
All input data for this project are publicly available from Vermont State and US Federal government agencies. Code for this project is made available via a FigShare repository https://doi.org/10.6084/m9.figshare.17056346 (Gourevitch et al., 2021).
