Social inequalities in climate change-attributed impacts of Hurricane Harvey | Panda Anku

Climate change-induced impacts of Hurricane Harvey

To determine the relative share of flood impacts during Hurricane Harvey attributable to climate change, we calculated climate change-attributed depths and damages using scenarios that compare the flooding that actually occurred to scenarios of flooding with less precipitation (i.e., flooding without climate change). Damages were calculated using depth-damage relationships specific to the building type, as defined in the National Structural Inventory (NSI) data. The mean damage was calculated for each structure in our dataset using the nonlinear damage functions (damage as a function of flood depths) constructed from National Flood Insurance claims data and the NSI types, as described in ref. 20.

Previous research examined seven possible scenarios of 7, 8, 13, 19, 20, 24, and 38% of precipitation during the storm that could be attributed to climate change; see ref. 16 and Methods for details. We calculate the climate change-attributed portion of depths and damages by subtracting flooding data from the scenarios with less precipitation from the baseline flood that occurred. Here, we present results for the two “best estimates”; a lower scenario of 20% less precipitation without climate change (the “best estimate” from ref. 13 and similar to the multigroup best estimate average of 19% from ref. 9), and a higher one of 38% less precipitation (the “best small-region estimate” from ref. 10). Results for the other five scenarios are presented in the Supplementary Information in Tables S1–S13. Results are shown for residential parcels.

Our analysis shows that 9.7 percent of residential parcels (~106,000 parcels) had buildings that flooded during Hurricane Harvey. For all seven climate change-attribution scenarios we consider, almost every flooded building (>99%) experienced at least some flooding attributed to climate change. These depths varied: the median increased flood depths attributed to climate change was 22 cm in the (20%) lower climate change-attribution scenario, and 27 cm in the (38%) higher scenario.

These climate change-attributed flood depths often made the difference between flooding a building and not flooding the same building at all. In the higher scenario (38% of precipitation is attributable to climate change), 49.4% of the buildings that were flooded would have been flooded anyway, but 50.6% flooded only because of climate change; i.e., they would not have been flooded during the hurricane had there been no anthropogenic climate change to generate increased rainfall. Since Harris County is large, this corresponds to an estimated 53,616 parcels that would not have been flooded without climate change. Figure 1 shows a map of areas that experienced flood impacts only because of climate change in the 38% scenario. For the lower “best estimate” (20%), the comparable figure is almost a third—i.e., 31.9% of the flooded houses would not have flooded without climate change. Even in the most conservative scenario we test—only 7% of the precipitation is associated with climate change—12.8% of the flooded residential buildings would not have flooded at all without climate change.

Fig. 1: Map of climate change-attributed flooding (38% scenario).
figure 1

Each hexagonal bin symbolizes the number of residential buildings that would not have flooded without the added impact of climate change in Harris County, Texas during Hurricane Harvey.

We also calculate the property damages wrought by these climate change-attributed flood depths using the information on buildings from the National Structure Inventory and depth-damage functions outlined in ref. 20 (see Methods). Our modeled estimate of the baseline flood damage to residential properties, including flooding both attributed and not attributed to climate change, is US$ 6.41 billion in Harris County. We estimate the climate change-attributed portion of these damages to be approximately $2.39 billion (37.2%) of total damages in the lower scenario or $3.7 billion (57.8%) in the higher climate change scenario.

Analysis of climate change-attributed impacts

Given the sizeable impacts of climate change on residential flooding from Hurricane Harvey, we next conduct regression analyses assessing what social and demographic characteristics of neighborhoods and land parcels are associated with these climate change-attributed impacts. We analyze neighborhood-level variables including the racial composition and median income of the census tracts, and including potentially noteworthy moderating (interacting) relationships between racial composition and income. We also examine parcel-level variables including the parcel’s appraised value, whether it is a single-family residential home, a mobile home, or a multifamily home, the year the residential structure was built, and whether the parcel has a building located in FEMA’s 100-year floodplain. In these regressions, statistically significant relationships and effect size calculations can be interpreted to identify disproportionate impacts for a social or demographic group.

Our first set of regressions assesses how these characteristics relate to two dependent variables attributed to climate change1: flood depths (in cm for buildings where flooding was >20 cm); and ref. 2 flood damages (the estimated amount of damage to residential buildings in U.S. dollars). The multivariable regressions use a Tobit specification, as a Tobit regression is appropriate for a left-censored variable where there are a large number of 0 cases (because many parcels did not have climate change-attributed flood depths). Table 1 shows these results for the 20 and 38% scenarios.

Table 1 Tobit regression of climate change-attributed depths for Hurricane Harvey in Harris County, Texas.

We identify six primary findings from these analyses that hold across the different scenarios for both depths and damages. First, parcels in neighborhoods with more Latina/x/o residents had higher climate change-attributed impacts. Figure 2 uses descriptive statistics where we multiply the number of parcels in three categories (i.e., not flooded, flooded because of climate change, would have flooded even without climate change) with the proportion of different racial groups in each neighborhood to provide a schematic to illustrate the racial disparities in flood depths for the 38% scenario. Figure 3 uses descriptive statistics (in the same manner as Fig. 2) to show these disparities for damages, with the per capita damages for a Latina/x/o person from climate change-attributed flooding estimated at ~$1,035. Although this estimate is only narrowly higher than that for whites ($828), it should be noted that home values are higher in white neighborhoods, and therefore this disparity per unit of home value is greater21,22,23.

Fig. 2: Percent of properties associated with each racial and ethnic group (38% scenario).
figure 2

Estimated percentages for residential properties in Harris County, Texas during Hurricane Harvey. Note: Group A included 1,002,026 parcels, group B 53,616 parcels, and group C 52,439 parcels.

Fig. 3: Estimated per capita property damage from flooding by racial composition (38% scenario).
figure 3

Estimated per capita damages for residential properties in Harris County, TX during Hurricane Harvey.

Second, parcels in neighborhoods with higher incomes had higher climate change-attributed impacts. Third, in neighborhoods with more Latina/x/o residents, the impact of income is reversed. In these neighborhoods, a greater impact was observed in the lower-income neighborhoods. This finding clarifies the previous two: While greater neighborhood incomes are linked to more climate change-induced impacts, the opposite is the case in Latina/x/a neighborhoods. Fourth, multifamily residential parcels (compared to single-family parcels) experienced less flood impacts associated with climate change. Fifth, location in FEMA’s 100-year floodplain was linked to greater climate change-attributed impacts. Sixth, older residential structures tended to have greater flood impacts.

In addition to these primary findings, other results were less consistent. For climate change-attributed damages but not climate change-attributed depths, we found evidence of a curvilinear (convex) effect for the appraised value of the parcel, but effect sizes were relatively small. We also found that mobile homes experienced less flood depths but this was not statistically significant for flood damages. Finally, we did not find statistically significant relationships for census tracts with a high proportion of non-Latina/x/o blacks or non-Latina/x/o of other races, including for moderating relationships with income.

Analysis of flooding only because of climate change impacts

In the second set of regression analyses, we ask what social and demographic characteristics are linked to parcels that would not have flooded without climate change by transforming our flood depths and flood damages variables into binary outcomes that denote whether a parcel’s buildings would not have flooded or did not flood at all. Parcels that would have flooded even without climate change-attributed precipitation are excluded, meaning that we conceptualize the sample as all parcels that would not have had flooded buildings if not for climate change, and then distinguish between those that did or did not flood in the climate change scenarios we examine. Table 2 shows the results of these binary logistic regression analyses.

Table 2 Logistic regression of climate change-attributed depths for Hurricane Harvey in Harris County, Texas.

Findings from these logistic analyses largely mirror those from the Tobit models on climate change-attributed flood depths and damages, thereby providing robust support for the overall findings. Most central to this study’s focus on climate justice, we find that Latina/x/o neighborhoods, especially low-income Latina/x/o neighborhoods, had greater odds of flooding (compared to other types of neighborhoods) only because of the added climate change-induced precipitation. This finding held for both depths and damages across each of the climate change-attribution scenarios, although the interaction effect for Latina/x/o and median income is slightly smaller in models for the 38% scenario (where p values are 0.064 for Model 2 and 0.07 for Model 4). Figure 4 graphs these findings with predicted probabilities by estimating the percentage of climate change-only flooded properties at different population shares of Latina/x/o residents and median income. As an example, the estimates show that for a high Latina/x/o population share, low-income (90%, $25,000 median income) neighborhood, we would estimate that ~9% of parcels in the 38% scenario (and 6% in the 20% scenario) would not have flooded if not for climate change.

Fig. 4: Predicted probabilities of parcel flooding only because of climate change.
figure 4

Predicted probabilities calculated for binary logistic regression results in Table 2 at levels of percent Latina/x/o, for Hurricane Harvey in Harris County, TX.

Four additional findings are also similar to the Tobit regression findings. First, a parcel’s location in a higher-income neighborhood is associated with higher odds of that parcel flooding only because of climate change. While this finding suggests greater hazard exposure for residents living in neighborhoods that are more economically well-off, it is also in juxtaposition to the opposite effect for income found in Latina/x/o neighborhoods. Second, multifamily residential parcels (compared to single-family parcels) had lower odds of crossing the flooding threshold of 20 cm because of climate change-induced impacts. Third, parcels located inside the FEMA 100-year floodplain had greater odds of climate change-attributed flooding. Fourth, older residences had higher odds of flooding only because of climate change.

Analysis inside and outside of floodplains

In the third set of analyses, we ask how social inequalities in climate change-attributed impacts are linked to the location in FEMA-delineated 100-year floodplains. Location in this Special Hazard Flood Area (SFHA) is the primary indicator of flood risk in the United States. For instance, any property within the 100-year flood zone is required to purchase flood insurance through the National Flood Insurance Program (NFIP) in order to be eligible for a mortgage from a federal agency24. Properties outside the SFHA, in contrast, are not required to purchase the NFIP coverage. Nevertheless, many homeowners within the SFHA still do not have flood insurance25,26,27. These uninsured may be undertaking other strategies to mitigate damage from flooding. More broadly, at the very least these within the SFHA are made aware that their residence is significantly exposed to flood risk. By contrast, residents outside of the SFHA are not similarly warned, and may therefore perceive a lower (or even nonexistent) risk in their outside-the-SFHA locations, even if the risk they face may also be significant27,28.

Our descriptive analyses show large impacts outside of the 100-year floodplains: 76.1% of flooded parcels are located outside of the SFHA floodplains, an impact totaling $4.9 billion in damages. The climate change-attributed portion of damages is higher outside of the floodplains (38.5% of damages in the lower scenario and 59.5% in the higher scenario) than inside of the floodplains (33.2% of damages in the lower scenario and 52.2% in the higher scenario). Coupling these climate change-attributed impacts with SFHA floodplain location, we estimate that between 29 and 45% of all damages from Harvey (totaling $1.9 to $2.9 billion in our model) occurred because of climate change and outside of the floodplain.

We find evidence for social inequalities in climate change-attributed impacts outside of the floodplain, but less so inside the floodplain. We re-analyzed the Tobit and binary logistic regression models in the two previous sections to account for a moderating relationship between floodplain location and census tract-level racial composition and median income variables; results are found in the Supplementary Information in Tables S9–S13. Among parcels outside of the floodplains, the econometric models show that climate change-attributed flooding is more likely in census tracts with more Latina/x/o residents. Previous findings relating to income, proportion Latina/x/o, and the moderating effect between these two variables hold in these models. In Fig. 5, we estimate (using descriptive statistics in a similar approach to that of Fig. 2) that ~52% of all parcels outside of the floodplain flooded because of climate change are estimated to be Latina/x/o households compared to 38% inside of the floodplain.

Fig. 5: Percent of Latina/x/o parcels flooded because of climate change inside and outside of floodplains (38% scenario).
figure 5

Estimated percentages for residential properties in Harris County, Texas during Hurricane Harvey.

Taken together, these findings suggest that there are more pronounced inequalities in climate change-attributed impacts in flooding outside of FEMA’s 100-year SFHA floodplains. Floodplain location is a key policy tool used to attempt to compel the uptake of flood insurance and other flood mitigation measures. This is important, as the insured homeowners, or those that were forewarned, are more likely to have the resources to pay for reconstruction and recovery. As such, a house located outside the floodplain is less likely to have access to recovery funding and is less likely to recover well29,30,31,32. Thus, the racial inequalities we find in the damage can be further exacerbated during the disaster recovery process.

Leave a Comment