City of Boston · MIT
Demographic projection · 2023
Integrating high-resolution flood risk into U.S. population projections flips the blocks most exposed to frequent flooding from a +5.76% projected growth to a −1.42% loss over 30 years. The standard SSP2 baseline holds those blocks on a continued-growth trajectory. The 2000–2020 empirical relationship between flood exposure and population change does not support that trajectory.
Each line connects the same set of relatively flood-exposed U.S. Census blocks under two scenarios. The SSP2 baseline is on the left (the official forecast). The climate-adjusted forecast is on the right, integrating empirically estimated flood-population responses. The 5-year return period (the most frequent flooding) produces the largest demographic swing.
Source: Shu, Porter, Hauer, Sandoval-Olascoaga et al., Nature Communications 14:7870 (2023) · First Street Foundation Flood Model (3-meter resolution) × SSP2 baseline projections (Hauer 2019) × propensity-matched OLS estimation, state-by-state · Treatment threshold: blocks at ~3× the within-state mean proportion of inundated properties · Counterintuitive finding: more frequent (5-yr RP), less severe flooding produces the largest demographic response
3-meter
Resolution of First Street's fluvial, pluvial, and coastal flood model
−7.19 pp
Cumulative growth swing for blocks most exposed to 5-year-RP flooding
7 to 10%
Miami-Dade tipping point. Growth rises with exposure up to here, then turns negative
Why it matters
Almost every demographic projection in active use treats flood exposure as exogenous to population. County-level Census Bureau forecasts, SSP scenarios feeding IPCC integrated assessment models, and state-level planning forecasts all hold population spatially fixed and treat the flood line as the only variable.
From 2000 to 2023, the most flood-exposed blocks lost population. Where flood exposure is sufficiently concentrated, people leave, or never arrive. This paper asks whether that relationship can be measured precisely enough to incorporate into the projections, producing a forecast that accounts for climate exposure at high spatial resolution.
How we did it
The flood layer is the First Street Foundation Flood Model. 3-meter resolution, covering fluvial, pluvial, and coastal sources at 5-, 20-, 100-, and 500-year return periods, with future projections built under RCP 4.5 from a 21-GCM ensemble. The demographic layer is the Hauer (2019) SSP2 county-level baseline projection, proportionally downscaled to U.S. Census blocks.
We then run a three-step empirical strategy. First, propensity-score matching within each state pairs synthetic treatment and control blocks on observable covariates (job density, job growth, median income, median home value, multi-deprivation, employment rate, 2000 population, density). Second, state-by-state OLS estimates block-level population change rates (2000 to 2020) as a function of flood exposure (linear plus squared, across return periods) with stepwise AIC selection and 80/20 train/test splits. Third, we apply the estimated coefficients to FSF's 30-year flood projections to produce a climate-adjusted forecast to circa 2053.
What we found
The headline. Census blocks with relatively high exposure to 5-year-RP flooding swing from +5.76% baseline cumulative growth to −1.42% loss when the climate-population relationship is integrated. That is a −7.19 percentage-point swing, the largest in the paper. For 20-year-RP exposure, the swing is −1.40pp. For the composite of any high exposure (5, 20, or 100-year), −1.54pp. The most dramatic adjustment shows up where flooding is most frequent, not where it is most severe.
The Miami-Dade case. Within Miami-Dade County the relationship between exposure and growth is non-linear. Population continues to rise with exposure up to roughly 7 to 10% of properties affected, then turns negative beyond that threshold. The most exposed blocks (Atlantic shoreline, Miami Beach, and blocks closest to the low-lying Everglades) are where the climate-adjusted forecast shows the largest projected population declines.
The Cincinnati counter-example. The flood-population relationship is not a national constant. In Cincinnati, highly inundated blocks are actually more likely to grow than less-exposed ones. A coastal and inland asymmetry that breaks any simple narrative about exposure meaning decline. The state-by-state estimation strategy is what makes this visible.
Within-county redistribution. Within a county that is still projected to grow on net, the spatial distribution of growth shifts. High-exposure blocks lose population. Lower-exposure blocks gain population. The county-level forecast becomes more spatially uneven, with growth concentrating in less-exposed blocks.
In the press
Bloomberg
"Over the last two decades, as San Antonio and surrounding Bexar County, Texas, grew by more than 600,000 people, some 17% of the city's blocks experienced a decrease in population."
More Than 3 Million Americans Are Already Climate Migrants, Researchers Say·December 2023
Associated Press: Flooding drives millions to move as climate migration patterns emerge
December 2023
CBS News: About 3 million Americans are already "climate migrants," analysis finds. Here's where they left.
December 2023
Axios: Climate migration may become "abandonment" as people flee flooding
December 2023
Exposure to high frequency flooding (5 and 20-year return periods) results in 2-7% lower growth rates than baseline projections. This is exacerbated in areas with relatively high exposure to frequent flooding where growth is expected to decline over the next 30 years. From the abstract.
Cite this paper
Shu, E. G., Porter, J. R., Hauer, M. E., Sandoval-Olascoaga, S., Gourevitch, J., Wilson, B., Pope, M., Melecio-Vazquez, D., & Kearns, E. (2023). Integrating climate change induced flood risk into future population projections. Nature Communications, 14, 7870. https://doi.org/10.1038/s41467-023-43493-8