This study addresses the following central question:
“Has the spatial diversity of FEMA disaster declarations increased from 1953 to 2017?”
To answer this, we examine three sub-questions:
Previous studies primarily focus on how many disasters occur, while
the spatial expansion and diversification of disasters remain
understudied.
This research is important for several reasons:
Spatial diversity cannot be captured by simple frequency
counts.
This study fills a meaningful gap in existing literature by addressing
long-term spatial patterns in disaster declarations.
A long-term administrative dataset that enables comparison of disaster patterns across the entire U.S. since 1953
Appropriate for analyzing spatial and temporal patterns
Only variables necessary for analysis were selected
Original structure kept; no additional manipulation
State centroid coordinates taken from R’s built-in state.center (approximate locations)
# Function Lord
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(ggplot2)
library(lubridate)##
## Attaching package: 'lubridate'## The following objects are masked from 'package:base':
##
## date, intersect, setdiff, unionlibrary(maps)
# Read CSV File
disaster <- read.csv("C:/Users/user/Desktop/data/database.csv",
stringsAsFactors = FALSE)
# Checking the name of Variables
str(disaster)## 'data.frame': 46185 obs. of 14 variables:
## $ Declaration.Number : chr "DR-1" "DR-2" "DR-3" "DR-4" ...
## $ Declaration.Type : chr "Disaster" "Disaster" "Disaster" "Disaster" ...
## $ Declaration.Date : chr "05/02/1953" "05/15/1953" "05/29/1953" "06/02/1953" ...
## $ State : chr "GA" "TX" "LA" "MI" ...
## $ County : chr "" "" "" "" ...
## $ Disaster.Type : chr "Tornado" "Tornado" "Flood" "Tornado" ...
## $ Disaster.Title : chr "Tornado" "Tornado and Heavy Rainfall" "Flood" "Tornado" ...
## $ Start.Date : chr "05/02/1953" "05/15/1953" "05/29/1953" "06/02/1953" ...
## $ End.Date : chr "05/02/1953" "05/15/1953" "05/29/1953" "06/02/1953" ...
## $ Close.Date : chr "06/01/1954" "01/01/1958" "02/01/1960" "02/01/1956" ...
## $ Individual.Assistance.Program : chr "Yes" "Yes" "Yes" "Yes" ...
## $ Individuals...Households.Program: chr "No" "No" "No" "No" ...
## $ Public.Assistance.Program : chr "Yes" "Yes" "Yes" "Yes" ...
## $ Hazard.Mitigation.Program : chr "Yes" "Yes" "Yes" "Yes" ...# Extract date -> year
disaster$Declaration.Date <- as.Date(disaster$Declaration.Date, format = "%m/%d/%Y")
disaster$year <- year(disaster$Declaration.Date)
# Create a subset with only the essential variables for analysis (for readability)
disaster_clean <- disaster %>%
select(year,
State,
Disaster.Type,
Declaration.Date) %>%
filter(!is.na(year),
!is.na(State),
State != "")# Descriptive statistics of key variables
summary(disaster_clean)## year State Disaster.Type Declaration.Date
## Min. :1953 Length:46185 Length:46185 Min. :1953-05-02
## 1st Qu.:1993 Class :character Class :character 1st Qu.:1993-03-17
## Median :2003 Mode :character Mode :character Median :2003-03-20
## Mean :1998 Mean :1998-10-26
## 3rd Qu.:2008 3rd Qu.:2008-07-09
## Max. :2017 Max. :2017-02-14State: Two-letter state code (e.g., TX, CA, NY)
Disaster.Type: Disaster categories such as Tornado, Flood, Hurricane
year: Year of the disaster declaration (1953–2017)
These three variables each play a key role:
State → “How many regions are affected?” (spatial extent)
Disaster.Type × State → “Which disasters occur in which regions?” (pattern diversity)
year → “How do these patterns change over time?” (long-term trends)
We examine the research questions step-by-step using three primary plots.
Plot 1 calculates the number of states with at least one disaster declaration per year, visualized with a bar chart, with an overlaid smooth line to show general trends.
# Number of States Affected Per Year
year_state_count <- disaster_clean %>%
group_by(year) %>%
summarise(
n_states = n_distinct(State),
.groups = "drop") %>%
filter(!is.na(year))
head(year_state_count, 100)## # A tibble: 65 × 2
## year n_states
## <dbl> <int>
## 1 1953 11
## 2 1954 17
## 3 1955 17
## 4 1956 14
## 5 1957 16
## 6 1958 7
## 7 1959 6
## 8 1960 10
## 9 1961 12
## 10 1962 20
## # ℹ 55 more rows# Plot 1: Bar + Smooth line
ggplot(year_state_count, aes(x = year, y = n_states)) +
geom_col(fill = "steelblue", alpha = 0.7) +
geom_smooth(method = "loess", se = FALSE, color = "darkred", linewidth = 1) +
labs(
title = "Number of States Affected by Disasters per Year",
x = "Year",
y = "Number of States with Disaster Declarations") +
theme_minimal(base_size = 12)## `geom_smooth()` using formula = 'y ~ x'
Key Observations - Continuous increase from the late 1950s to early 2000s - Peak in early 2000s - Slight decline afterward
Core Interpretation - Disaster impact has expanded spatially over the long term - Despite yearly fluctuations, the pattern is clear: 1950s < 1980s < 2000s - Some years show more than 40 states affected → Indicates U.S. disasters increasingly impact many regions simultaneously
Plot 2 counts unique disaster-type × state combinations for each year. It can measure complexity and diversity of the national disaster “portfolio”.
We use the simple measure: unique combinations per year.
If in one year the combinations were (Flood, TX), (Flood, CA), (Hurricane, FL) → then the count = 3
# Unique Type × State Combinations Over Time
combo_diversity <- disaster_clean %>%
mutate(
combo = paste(Disaster.Type, State, sep = "_")) %>%
group_by(year) %>%
summarise(
n_unique_combos = n_distinct(combo),
.groups = "drop") %>%
filter(!is.na(year))
head(combo_diversity)## # A tibble: 6 × 2
## year n_unique_combos
## <dbl> <int>
## 1 1953 12
## 2 1954 17
## 3 1955 17
## 4 1956 15
## 5 1957 16
## 6 1958 7ggplot(combo_diversity, aes(x = year, y = n_unique_combos)) +
geom_line(color = "darkorange", linewidth = 1) +
geom_point(color = "darkorange", size = 1) +
labs(
title = "Yearly Diversity of (Disaster Type × State) Combinations",
x = "Year",
y = "Number of Unique Combinations") +
theme_minimal(base_size = 12)
Key Interpretation - Large increase from the 1950s (10–20 combos) to the 2000s (60–90 combos) - Disaster types spread across more regions - Trend shows long-term growth despite fluctuations - Recent decreases may be temporary or reflect external factors
Conclusion Not only has the range expanded, but the U.S. hazard environment has become far more complex and multi-layered.
Summary of how the spatial centers of each disaster type have moved over time:
Annual Centroids by Disaster Type (Top 5) 1. Fire, 2. Flood, 3. Snow, 4. storm, 5. Hurricane
Use the state center coordinates (state.center) for the 50 U.S. states to obtain approximate latitude and longitude for each state.
Map each FEMA disaster declaration to the corresponding state’s centroid coordinates.
For each year, compute the average latitude (lat) and longitude (lon) of all disaster declarations to obtain the yearly centroid.
state_coords <- data.frame(
State = state.abb,
lon = state.center$x,
lat = state.center$y)
# 3. Match FEMA data with state centroid coordinates (latitude/longitude)
dis_geo <- disaster %>%
left_join(state_coords, by = "State") %>%
filter(!is.na(lon), !is.na(lat), !is.na(Disaster.Type), !is.na(year))
# 4. Extract the top 5 most frequent disaster types
top5_types <- dis_geo %>%
count(Disaster.Type, sort = TRUE) %>%
slice_head(n = 5) %>%
pull(Disaster.Type)
# 5. Compute yearly centroids (mean latitude/longitude) for top 5 disaster types
centroids_top5 <- dis_geo %>%
filter(Disaster.Type %in% top5_types) %>%
group_by(Disaster.Type, year) %>%
summarise(
centroid_lon = mean(lon, na.rm = TRUE),
centroid_lat = mean(lat, na.rm = TRUE),
.groups = "drop") %>%
arrange(Disaster.Type, year)# 1. U.S. map data
us_map <- map_data("state")
# 2. Define a function to plot centroid movement for each disaster type
plot_centroid_type <- function(type_label) {
# Filter for the selected disaster type
df_type <- centroids_top5 %>%
filter(Disaster.Type == type_label)
ggplot() +
# Background U.S. map
geom_polygon(
data = us_map,
aes(x = long, y = lat, group = group),
fill = "white",
color = "gray70",
inherit.aes = FALSE) +
geom_path(
data = df_type,
aes(x = centroid_lon, y = centroid_lat, group = 1),
color = "gray",
linewidth = 0.5) +
# Centroid points with color mapped to year
geom_point(
data = df_type,
aes(x = centroid_lon, y = centroid_lat, color = year),
size = 2) +
scale_color_gradient(
low = "blue",
high = "orange",
name = "Year") +
labs(
title = paste0("Top 5 Disaster: ", type_label),
subtitle = "FEMA Disaster Declarations, 1953–2017",
x = "Longitude",
y = "Latitude") +
coord_fixed(1.3) +
theme_minimal(base_size = 11) +
theme(panel.grid = element_line(linewidth = 0.2, color = "gray85"))}# 3. Generate plots for the top 5 disaster types
plot_centroid_type("Fire")
plot_centroid_type("Flood")
plot_centroid_type("Snow")
plot_centroid_type("Storm")
plot_centroid_type("Hurricane")
Snow Spread across a wide East–Central range Early (blue) and recent (orange) points intermixed → No consistent directional shift → High yearly variability
Storm
Combining all analyses, long-term structural changes in U.S. disaster patterns are as follows:
Expansion of Spatial Extent (Plot 1) → More states are being affected over time
Increase in Structural Complexity (Plot 2) → More diverse types of disasters occurring across more regions
Geographic Shifts in Disaster Centers (Plot 3) → Some types (Fire, Hurricane) show clear directional movement → Others (Flood, Storm) remain stable → Snow is highly variable
→ Together, these indicators show a structural reconfiguration of national disaster patterns compared to the past.
Limitations - Dataset includes only declared FEMA disasters → Non-declared events excluded - state.center provides approximate—not precise—geographic locations
Further Work - Weighted centroid analysis incorporating cost or damage severity - Clustering analysis based on disaster occurrence patterns