Residential Buildings At Risk of Flooding - Brooklyn

This project identifies and visualizes residential buildings located within the 100-year flood plain in Brooklyn, New York. I explored the spatial distribution of populations at risk of flooding using dasymetric mapping, which redistributes demographic data from large census datasets to more practical spatial units, such as tax lots or land-use parcels. The purpose of this analysis was to generate more accurate estimates of the number of people residing within the 100-year flood plain, as designated by the Federal Emergency Management Agency (FEMA). This project also aims to demonstrate the implications of the modifiable areal unit problem (MAUP), the issue where analytical results can vary depending on the spatial units used in the calculation.

Create map layer data

# Read PLUTO (Brooklyn only BK) and compute ARA

pluto <- st_read(here::here("nyc_mappluto_24v3_1_fgdb/MapPLUTO24v3_1.gdb"),

query = "select * from MapPLUTO_24v3_1_clipped where Borough = 'BK'", quiet = TRUE)

pluto <- pluto %>%

mutate(ARA = case_when(ResArea == 0 & UnitsRes != 0 ~ 1 * (BldgArea * (UnitsRes/UnitsTotal)) + ResArea,

ResArea > 0 ~ ResArea,

TRUE ~ 0))

# ACS: block groups and tracts for Kings County

bk_bg <- get_acs(geography = "block group",

variables = c(bg_tot_pop = "B03002_001",

bg_white_nh = "B03002_003" ,

bg_black_nh = "B03002_004",

bg_asian_nh = "B03002_006" ,

bg_native_hawaiian = "B03002_007",

bg_other_nh = "B03002_008",

bg_two_more_nh = "B03002_009",

bg_hispanic = "B03002_012"),

year = 2021, output = "wide",

geometry = TRUE, state = "NY",

county = "Kings", survey = "acs5",

progress = FALSE)

bk_tracts <- get_acs(geography = "tract",

variables = c(tract_tot_pop = "B03002_001",

tract_white_nh = "B03002_003",

tract_black_nh = "B03002_004",

tract_asian_nh = "B03002_006",

tract_native_hawaiian = "B03002_007",

tract_other_nh = "B03002_008" ,

tract_two_more_nh = "B03002_009",

tract_hispanic = "B03002_012" ),

year = 2021, output = "wide",

geometry = TRUE, state = "NY",

county = "Kings", survey = "acs5",

progress = FALSE)

# Join census IDs to PLUTO and select needed fields

pluto <- st_transform(pluto, crs = st_crs(bk_bg))

pluto <- pluto %>%

st_join(bk_bg %>% select(block_group_2021 = GEOID, 3, 5, 7, 9, 11, 13, 15, 17))

pluto <- pluto %>%

st_join(bk_tracts %>% select(tract_2021 = GEOID, 3, 5, 7, 9, 11, 13, 15, 17))

pluto <- pluto %>%

select(Borough, Address, OwnerName, LotArea, BldgArea,

ResArea, UnitsRes, ARA, block_group_2021,

starts_with("bg_"), tract_2021, starts_with("tract_"))

# Unit/area totals (for dasymetric weighting)

bk_bg_units <- pluto %>%

as_tibble() %>%

group_by(block_group_2021) %>%

summarise(bg_res_units = sum(UnitsRes, na.rm = TRUE),

bg_ARA = sum(ARA, na.rm = TRUE))

bk_tract_units <- pluto %>%

as_tibble() %>%

group_by(tract_2021) %>%

summarise(tract_res_units = sum(UnitsRes, na.rm = TRUE),

tract_ARA = sum(ARA, na.rm = TRUE))

pluto <- pluto %>% left_join(bk_bg_units)

pluto <- pluto %>% left_join(bk_tract_units)

# Dasymetric population estimates on lots

pluto <- pluto %>%

ungroup() %>%

mutate(bg_tot_dasym = case_when(UnitsRes == 0 & ResArea != 0 ~

bg_tot_popE * (ARA/bg_ARA),

UnitsRes > 0 ~ bg_tot_popE * (UnitsRes/bg_res_units),

TRUE ~ 0),

tract_tot_dasym = case_when(UnitsRes == 0 & ResArea != 0 ~

tract_tot_popE * (ARA/tract_ARA),

UnitsRes > 0 ~ tract_tot_popE * (UnitsRes/tract_res_units),

TRUE ~ 0))

# Floodplain polygons (100-year), validate and union

floods <- st_read("https://data.cityofnewyork.us/api/geospatial/ezfn-5dsb?method=export&format=GeoJSON")

floods <- st_make_valid(floods)

floods <- floods %>%

mutate(flood_plain = "100-Year") %>%

rename(geometry = x)

floods <- floods %>%

st_transform(crs = st_crs(pluto)) %>%

st_union(by_feature = FALSE) %>%

st_as_sf()

Figure 1: This map overlays the Brooklyn tax lots (NYC Map PLUTO) with the 100-Year FEMA flood plain. Most of Brooklyn's flood-exposed residential buildings are in low-lying areas in the South and Southwest, including neighborhoods such as Coney Island, Red Hook, and Canarise.

Table showing the MAUP

# Block-group and tract totals intersecting floods (ACS totals)

bk_bg_floods <- bk_bg %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(flood_pop_bg = sum(bg_tot_popE, na.rm = TRUE)) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

bk_tract_flood <- bk_tracts %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(flood_pop_tract = sum(tract_tot_popE, na.rm = TRUE)) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

# Dasymetric totals intersecting floods (from PLUTO)

pluto_floods <- pluto_res %>% st_join(floods)

pluto_floods <- pluto_floods %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(flood_pop_bg_dasym = sum(bg_tot_dasym, na.rm = TRUE),

flood_pop_tract_dasym = sum(tract_tot_dasym, na.rm = TRUE)) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

pluto_floods <- pluto_floods %>%

left_join(bk_tract_flood) %>%

left_join(bk_bg_floods)

# Race-specific summaries (BG + Tract)

bg_white_flood <- bk_bg %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(

flood_white_dasym = sum(bg_white_nhE, na.rm = TRUE),

flood_white_acs = sum(bg_white_nhE, na.rm = TRUE)

) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

bg_black_flood <- bk_bg %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(

flood_black_dasym = sum(bg_black_nhE, na.rm = TRUE),

flood_black_acs = sum(bg_black_nhE, na.rm = TRUE)

) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

bg_hispanic_flood <- bk_bg %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(

flood_hispanic_dasym = sum(bg_hispanicE, na.rm = TRUE),

flood_hispanic_acs = sum(bg_hispanicE, na.rm = TRUE)

) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

# Helper to build combined BG/Tract race tables

create_race_summary <- function(data_bg, data_tract, race_var_bg, race_var_tract) {

bg_summary <- data_bg %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(

Dasymetric_BG = sum(!!sym(race_var_bg), na.rm = TRUE),

ACS_BG = sum(!!sym(race_var_bg), na.rm = TRUE)

) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

tract_summary <- data_tract %>%

st_join(floods) %>%

as_tibble() %>%

group_by(flood_plain) %>%

summarise(

Dasymetric_Tract = sum(!!sym(race_var_tract), na.rm = TRUE),

ACS_Tract = sum(!!sym(race_var_tract), na.rm = TRUE)

) %>%

mutate(flood_plain = if_else(is.na(flood_plain), "Not in plain", flood_plain))

combined_summary <- bg_summary %>%

full_join(tract_summary, by = "flood_plain")

return(combined_summary)

}

white_summary <- create_race_summary(bk_bg, bk_tracts, "bg_white_nhE", "tract_white_nhE")

black_summary <- create_race_summary(bk_bg, bk_tracts, "bg_black_nhE", "tract_black_nhE")

hispanic_summary<- create_race_summary(bk_bg, bk_tracts, "bg_hispanicE", "tract_hispanicE")

(Figure 2) The table highlights how different spatial aggregation units—dasymetric block vs ACS block and dasymetric census vs ACS census—influence the estimation of populations at risk of flooding in Brooklyn.

The number of residents located within the 100-year floodplain varies depending on whether population data is allocated using block-level ACS counts, tract-level ACS counts, or tax-lot-based dasymetric mapping. For all three racial and ethnic groups shown, the dasymetric block-level estimates are consistently lower than the ACS block totals, while tract-based ACS estimates are higher. This table demonstrates the MAUP: population estimates change when census geographies are aggregated differently, even though the people do not move. MAUP-related uncertainty should be explicitly acknowledged when estimating who is at risk from flooding, because funding and the allocation of resources often rely on population counts.

Data Used:

NYC MapPLUTO 24v3.1 (shoreline-clipped) – parcel geometry and attributes (Borough, UnitsRes, ResArea, BldgArea, UnitsTotal).

ACS 2017–2021 (5-year) via tidycensus – total population and race/ethnicity at block group and tract levels (Kings County).

NYC 100-year Floodplain polygons (NYC Open Data). CRS aligned to the PLUTO layer before spatial joins.