The following code estimates welfare changes associated with varying conditions of nearshore environmental quality on the island of Hawai’i. The primary impact considered is the loss of coral reef biomass, which serves as an environmental and economic asset. Welfare effects are estimated through two distinct models that capture changes in household utility and behavioral responses to environmental degradation. These models use observed visitation and population data to approximate demand, and they translate changes in reef quality into utility impacts, following standard approaches in environmental and recreational economics. The two models include:
The graphical representation of this process is illustrated below
This section calls the necessary R packages and sources external scripts from the original study. These scripts provide the estimated coefficients from the Random Utility Model (RUM), which will be used to simulate individual preferences and estimate welfare changes under different environmental quality scenarios. These parameters reflect the marginal utility of various site attributes, including coral reef biomass, travel cost, and environmental conditions.
library(sf)
library(dplyr)
library(tidyr)
library(ggplot2)
library(sf)
library(gganimate)
### load functions
getwd()
source("codedata/dummier.R")
source("codedata/travel_cost.R")
source("codedata/gridplot.R")
### load model ###
load(file = "codedata/model3.mod")
### this are the travel cost model estimates ###
m1To assess environmental changes in the nearshore ecosystem, we incorporate biomass outputs from the Atlantis ecosystem model. These outputs represent projected ecological dynamics—specifically, coral reef and associated species biomass—under different management and environmental scenarios.
The biomass values will be:
Spatially aggregated by polygon to match the resolution used in the economic model.
Processed annually to calculate percentage changes in biomass relative to the baseline year.
These percentage changes provide the key environmental quality metric (e.g., reef health) that feeds into the Economic Model outlined in the previous section. Specifically, the estimated changes will modify site attributes within the Random Utility Model framework to simulate how changes in reef biomass affect recreational behavior and welfare.
The base level of coral is the average biomass of 2015-2020 for each near polygon.
Then for each time step out to 2100 we sum up the biomass of coral reef:
\(C_{pt} = \sum V_{pt} / \sum V_{pt}=0\)
The following code does this for scenario SSP1 1-3. This looks at the relative change of coral biomass within each polygon in the nearshore environment.
library(dplyr)
library(tidyr)
x1 <- readRDS("codedata/x1.rds")
keep=unique(x1$Box_ID_2)
###This grabs Atlantis and gets it into the write steps
SSP3_loPPkF <- readRDS("codedata/SSP3_control_last.rds")
biomass_spatial_stanza=SSP3_loPPkF$cover
ssp3_coral=subset(biomass_spatial_stanza,species=="Pocillopora" |species=="Porites branching" |species=="Porites massive"|species=="Montipora"
|species=="Leptoseris")
ssp3_coral$time=round(ssp3_coral$time, digits = 0)
ssp3_coral=ssp3_coral %>% distinct(time, polygon, species, .keep_all = TRUE)
ssp3_coral=ssp3_coral %>% group_by(polygon, time) %>% summarize(Coral_NearShore=sum(atoutput))
ssp3_coral=ssp3_coral[ssp3_coral$polygon %in% keep, ]
ssp3_coral=ssp3_coral %>%
filter(!(time %% 1))
ssp3_coral[order(ssp3_coral$time),]
ssp3_coral=ssp3_coral %>%
filter(!(time <= 9))
t=ssp3_coral %>% filter(time<=14)%>%
group_by(polygon) %>% summarise(mean=mean(Coral_NearShore))
ssp3_coral=ssp3_coral %>%
filter(!(time <= 13))
ssp3_coral=merge(t,ssp3_coral)
percentchangeSSP3=ssp3_coral %>%
group_by(polygon) %>%
mutate(Coral_perchange = (Coral_NearShore)/mean+1)
percentchangeSSP3=percentchangeSSP3[c(1,3,5)]
percentchangeSSP3$time<- sub("^", "time", percentchangeSSP3$time )
SSP3=spread(percentchangeSSP3, time, Coral_perchange)
SSP1_loPPkF <- readRDS("codedata/SSP1_control_last.rds")
biomass_spatial_stanza=SSP1_loPPkF$cover
ssp1_coral=subset(biomass_spatial_stanza,species=="Pocillopora" |species=="Porites branching" |species=="Porites massive"|species=="Montipora"
|species=="Leptoseris")
ssp1_coral=ssp1_coral %>% group_by(polygon, time) %>% summarize(Coral_NearShore=sum(atoutput))
ssp1_coral=ssp1_coral[ssp1_coral$polygon %in% keep, ]
ssp1_coral=ssp1_coral %>%
filter(!(time %% 1))
ssp1_coral[order(ssp1_coral$time),]
ssp1_coral=ssp1_coral %>%
filter(!(time <= 9))
t=ssp1_coral %>% filter(time<=14)%>%
group_by(polygon) %>% summarise(mean=mean(Coral_NearShore))
ssp1_coral=ssp1_coral %>%
filter(!(time <= 13))
ssp1_coral=merge(t,ssp1_coral)
percentchangeSSP1=ssp1_coral %>%
group_by(polygon) %>%
mutate(Coral_perchange = (Coral_NearShore)/mean)
percentchangeSSP1=percentchangeSSP1[c(1,3,5)]
percentchangeSSP1$time<- sub("^", "time", percentchangeSSP1$time )
SSP1=spread(percentchangeSSP1, time, Coral_perchange)
SSP2_loPPkF <- readRDS("codedata/SSP2_control_last.rds")
biomass_spatial_stanza=SSP2_loPPkF$cover
ssp2_coral=subset(biomass_spatial_stanza,species=="Pocillopora" |species=="Porites branching" |species=="Porites massive"|species=="Montipora"
|species=="Leptoseris"| species=="crustose coralline algae")
ssp2_coral=ssp2_coral %>% group_by(polygon, time) %>% summarize(Coral_NearShore=sum(atoutput))
ssp2_coral=ssp2_coral[ssp2_coral$polygon %in% keep, ]
ssp2_coral=ssp2_coral %>%
filter(!(time %% 1))
ssp2_coral[order(ssp2_coral$time),]
ssp2_coral=ssp2_coral %>%
filter(!(time <= 9))
t=ssp2_coral %>% filter(time<=14)%>%
group_by(polygon) %>% summarise(mean=mean(Coral_NearShore))
ssp2_coral=ssp2_coral %>%
filter(!(time <= 13))
ssp2_coral=merge(t,ssp2_coral)
percentchangeSSP2=ssp2_coral %>%
group_by(polygon) %>%
mutate(Coral_perchange = (Coral_NearShore)/mean)
percentchangeSSP2=percentchangeSSP2[c(1,3,5)]
percentchangeSSP2$time<- sub("^", "time", percentchangeSSP2$time )
SSP2=spread(percentchangeSSP2, time, Coral_perchange)
percentchangeSSP1 <- percentchangeSSP1 %>% mutate(SSP = "SSP1")
percentchangeSSP2 <- percentchangeSSP2 %>% mutate(SSP = "SSP2")
percentchangeSSP3 <- percentchangeSSP3 %>% mutate(SSP = "SSP3")
coral_graph=rbind(percentchangeSSP1,percentchangeSSP2,percentchangeSSP3)
rm(ssp1_coral,SSP1_loPPkF,ssp2_coral,SSP2_loPPkF, ssp3_coral,SSP3_loPPkF,percentchangeSSP1,percentchangeSSP2,percentchangeSSP3, biomass_spatial_stanza)This shows the overall predicted relative change in coral cover across each scenario.
coral_graph <- coral_graph %>%
mutate(
time_step = as.numeric(gsub("time", "", time)) + 2006
) %>%
filter(time_step <= 2100)
coral1=coral_graph %>% group_by(SSP, time_step) %>% summarize(mean_coral=mean(Coral_perchange))
group.colors <- c(
SSP1 = "royalblue",
SSP2 = "springgreen",
SSP3 = "deeppink"
)
coral1$percentage=coral1$mean_coral-1
ggplot(coral1, aes(x = time_step, y = percentage, color = SSP)) +
geom_point(alpha = 0.3) + # optional: show raw means
geom_smooth(method = "loess", se = TRUE, aes(fill = SSP), alpha = 0.2) +
scale_color_manual(values = group.colors) +
scale_fill_manual(values = group.colors) +
theme_minimal() +
theme(
text = element_text(family = "Times New Roman"),
plot.title = element_text(hjust = 0.5),
legend.position = "bottom",
legend.background = element_rect(fill = "white")
) +
labs(
title = "Projected Change in Coral Cover By SSPs",
x = "Year",
y = "Relative Change Coral Cover",
color = "SSP"
) +
theme_minimal(base_size = 14) +
theme(
text = element_text(family = "Times New Roman"), plot.title = element_text(hjust = 0.5),
legend.position = "bottom",
legend.title = element_text(family = "Times New Roman"),
legend.text = element_text(family = "Times New Roman")
)
ggsave('images/avercoral_scenario.png', dpi = 300, height = 5, width = 8, unit = 'in')The following code shows the predicted change in population under scenario SSP1 1-3 using the data available for state of Hawaii Zoraghein et al. (2020)
library(readxl)
pop<- read_excel("codedata/population_Zoragheinetal2020.xlsx")
pop$Population= as.numeric(pop$Population)
pop=pop %>%
group_by(Scenario) %>%
mutate(
BasePop = Population[Year == 2020],
Population_weight = (Population) / BasePop
) %>%
ungroup()
pop <- pop %>%
mutate(across(c( Scenario), toupper))
pop=subset(pop, Scenario!="SSP5" & Scenario!="SSP4")
group.colors <- c(
SSP1 = "royalblue",
SSP2 = "springgreen",
SSP3 = "deeppink"
)
ggplot(pop, aes(x = Year,
y = Population/1000000,
color = Scenario,
group = Scenario)) +
geom_line(size = 1.2) +
scale_color_manual(values = group.colors) +
labs(
title = "Population Growth By SSPs in Hawaii",
x = "Year",
y = "Population in Millions",
color = "Scenario"
) +
theme_minimal(base_size = 14) +
theme(
text = element_text(family = "Times New Roman"), plot.title = element_text(hjust = 0.5),
legend.position = "bottom",
legend.title = element_text(family = "Times New Roman"),
legend.text = element_text(family = "Times New Roman")
)
ggsave('images/population_scenario.png', dpi = 300, height = 5, width = 8, unit = 'in')The following code calculates the travel cost from each residential grid to each recreation site across the Hawaiian Islands. It uses economic and geographic data collected from each island, including household income, travel distance, and mode assumptions.
Travel costs are estimated following the methodology of Fezzi et al. (2023), who incorporate both time and monetary expenses into a unified measure of generalized cost. These estimates serve as one of the key explanatory variables in the Random Utility Model (RUM), which is used to simulate recreation site choice.
For each scenario predicted by the Atlantis ecosystem model (e.g., SSP1, SSP2, SSP3), we run a simulation loop across all time steps to estimate the welfare impacts of changing nearshore coral biomass.
Specifically, for each year in the scenario:
Biomass outputs are aggregated spatially (by polygon or site).
The percentage change in coral reef biomass relative to the baseline is calculated.
These changes are passed into the Random Utility Model, updating site attributes for each grid-cell household.
Compensating Variation (CV) is then computed to measure the change in household welfare due to environmental change.
This process is repeated for each:
Grid cell (representing origin locations of visitors)
Recreational site (destination choices)
Time step (e.g., yearly from 2025 to 2100)
Scenario (SSP1–3)
Below is a loop (conceptual or in code) that:
Iterates through each time step of the SSP1 scenario
Updates coral biomass conditions
Recalculates utilities and choice probabilities
Estimates welfare change (CV) for each grid-location pair
### parameters
x1 <- readRDS("codedata/x1.rds")
x1$coastal[x1$name == "Home"]=0
### Prepared dataset for maps plot:
out <- x1[order(x1$siteID),]
out <- out[c(1,3,8:12)]
out=out[!duplicated(out[2]),]
##using 85K because that was the survey amount
out$incometran_track=out$medincome_tract/85000
sites_join <- read.csv("codedata/sites_join.csv")
coasts <- sites_join[sites_join$coastal == 1 & sites_join$name!="Home",]
x1=merge(SSP1, x1, by.x = "polygon", by.y = "Box_ID_2", all.y = TRUE)
x1 <- x1[order(x1$km2_ID,x1$siteID),] ## btw it is correct that the dim(x) is lower than dim(dist) because
## I aggregated some sites
length(unique(x1$siteID))
### FINISHED LOADING AND MERGING DATASETS ###
#rm(dist, pop, sitesjoin)
### CALCULATE TRAVEL COST ####
x1$cost <- 0
x1$duration[ x1$siteID == 351] <- x1$duration[ x1$siteID == 351] + 40 ## this sites needs 40min walking to be reached
## Need to also do this for Hawaii
### Change here for different fuel cost
x1$cost <- x1$distance.meters*0.001 * (0.20 / 1.5 ) ## fuel cost
## assumption on fuel cost:
#
# 20 miles/gallon
# $4.3/gallon
# means $4.3 per 20 miles --> 0.215 $/mile
# 10 km --> 6.2 miles --> $1.33 travel cost per km
# 10km * 0.20 / 1.5 = $1.33 so the values coincide with the cost used above.
x1$cost <- x1$cost + (x1$duration / 60)* 3/4 * 24.4 ## add the value of time (assuming a 25$ per hour average wage rate)
x1$cost <- x1$cost * 2 ## the value for a return trip
#### put Molokini cost equal to the one of Makena Landing Beach Park + 100$ (cost of a snorkeling tour)
x1$cost[x1$siteID == 388] <- x1$cost[x1$siteID == 359] + 100
beta <- m1$results[,1]
X0 <- cbind( x1$cost,x1$home, x1$coastal, x1$citypark, x1$trailall,
x1$haleakala, x1$molokini, x1$parking, x1$showers,
x1$lifeguard, x1$rock, x1$manmade, x1$beachmedium, x1$beachlarge, x1$surf,
x1$swim, x1$snork, x1$res_fish_bio, x1$PNAS, x1$playground, x1$sports,
x1$snork*x1$PNAS, x1$snork*x1$res_fish_bio)
X0[is.na(X0)] <- 0
xsim <- x1
final <- data.frame(matrix(ncol =2, nrow = 0))
all <- data.frame(matrix(ncol =2, nrow = 0))
#colnames(final) <- c("time10", "time20" , "time30" , "time40")
#this needs to be ran as a loop
for (i in 2:82) {
xsim <- x1
xsim$PNAS <- xsim$PNAS *xsim[,i]
xsim$PNAS[xsim$PNAS <= 0] <- 0
X1 <- cbind( xsim$cost,xsim$home, xsim$coastal, xsim$citypark, xsim$trailall,
xsim$haleakala, xsim$molokini, xsim$parking, xsim$showers,
xsim$lifeguard, xsim$rock, xsim$manmade, xsim$beachmedium, xsim$beachlarge, xsim$surf,
xsim$swim, xsim$snork, xsim$res_fish_bio, xsim$PNAS, xsim$playground, xsim$sports,
xsim$snork*xsim$PNAS, xsim$snork*xsim$res_fish_bio)
X1[is.na(X1)] <- 0
### CALCULATION OF WTP FOR EACH GRID LOCATION
wtpw <- quality.change(beta = beta, ### parameters
X0=X0, ### X variables in the baseline
X1 = X1, #### X variables in the scenario
pid = x1$km2_ID, ### ID of the 1km grid
cost =1) ### column of the cost parameter in the X matrix
### TOTAL WTP
b=merge(wtpw,out, by.x=c("id"),by.y=c("km2_ID"))
b$time=i
a=b %>% group_by(island) %>% summarize(SSP1_block=sum(WTP *grid_km_censusblock_Pop18_densityKM * 365)
)
a$time=i
all=rbind(b,all)
final=rbind(a,final)
}
df=final
all_df=all
colnames(all_df)[2]="SSP1"x1 <- readRDS("codedata/x1.rds")
x1=merge(SSP2, x1, by.x = "polygon", by.y = "Box_ID_2", all.y = TRUE)
x1 <- x1[order(x1$km2_ID,x1$siteID),] ## btw it is correct that the dim(x) is lower than dim(dist) because
## I aggregated some sites
length(unique(x1$siteID))
### FINISHED LOADING AND MERGING DATASETS ###
#rm(dist, pop, sitesjoin)
### CALCULATE TRAVEL COST ####
x1$cost <- 0
x1$duration[ x1$siteID == 351] <- x1$duration[ x1$siteID == 351] + 40 ## this sites needs 40min walking to be reached
## Need to also do this for Hawaii
### Change here for different fuel cost
x1$cost <- x1$distance.meters*0.001 * (0.20 / 1.5 ) ## fuel cost
## assumption on fuel cost:
#
# 20 miles/gallon
# $4.3/gallon
# means $4.3 per 20 miles --> 0.215 $/mile
# 10 km --> 6.2 miles --> $1.33 travel cost per km
# 10km * 0.20 / 1.5 = $1.33 so the values coincide with the cost used above.
x1$cost <- x1$cost + (x1$duration / 60)* 3/4 * 24.4
x1$cost <- x1$cost * 2 ## the value for a return trip
#### put Molokini cost equal to the one of Makena Landing Beach Park + 100$ (cost of a snorkeling tour)
x1$cost[x1$siteID == 388] <- x1$cost[x1$siteID == 359] + 100
### PREPARE DATA FOR PREDICTIONS
### parameters
beta <- m1$results[,1]
X0 <- cbind( x1$cost,x1$home, x1$coastal, x1$citypark, x1$trailall,
x1$haleakala, x1$molokini, x1$parking, x1$showers,
x1$lifeguard, x1$rock, x1$manmade, x1$beachmedium, x1$beachlarge, x1$surf,
x1$swim, x1$snork, x1$res_fish_bio, x1$PNAS, x1$playground, x1$sports,
x1$snork*x1$PNAS, x1$snork*x1$res_fish_bio)
X0[is.na(X0)] <- 0
xsim <- x1
final <- data.frame(matrix(ncol =2, nrow = 0))
all <- data.frame(matrix(ncol =2, nrow = 0))
#colnames(final) <- c("time10", "time20" , "time30" , "time40")
#this needs to be ran as a loop
for (i in 2:83) {
xsim <- x1
xsim$PNAS <- xsim$PNAS *xsim[,i]
xsim$PNAS[xsim$PNAS <= 0] <- 0
X1 <- cbind( xsim$cost,xsim$home, xsim$coastal, xsim$citypark, xsim$trailall,
xsim$haleakala, xsim$molokini, xsim$parking, xsim$showers,
xsim$lifeguard, xsim$rock, xsim$manmade, xsim$beachmedium, xsim$beachlarge, xsim$surf,
xsim$swim, xsim$snork, xsim$res_fish_bio, xsim$PNAS, xsim$playground, xsim$sports,
xsim$snork*xsim$PNAS, xsim$snork*xsim$res_fish_bio)
X1[is.na(X1)] <- 0
### CALCULATION OF WTP FOR EACH GRID LOCATION
wtpw <- quality.change(beta = beta, ### parameters
X0=X0, ### X variables in the baseline
X1 = X1, #### X variables in the scenario
pid = x1$km2_ID, ### ID of the 1km grid
cost =1) ### column of the cost parameter in the X matrix
### TOTAL WTP
b=merge(wtpw,out, by.x=c("id"),by.y=c("km2_ID"))
b$time=i
a=b %>% group_by(island) %>% summarize(SSP2_block=sum(WTP *grid_km_censusblock_Pop18_densityKM * 365)
)
a$time=i
all=rbind(b,all)
final=rbind(a,final)
}
df=merge(df,final, all.x = TRUE)
colnames(all)[2]="SSP2"
all_df=merge(all_df, all, all.x = TRUE)####This is for SSP3
x1 <- readRDS("codedata/x1.rds")
x1=merge(SSP3, x1, by.x = "polygon", by.y = "Box_ID_2", all.y = TRUE)
x1 <- x1[order(x1$km2_ID,x1$siteID),] ## btw it is correct that the dim(x) is lower than dim(dist) because
## I aggregated some sites
length(unique(x1$siteID))
### FINISHED LOADING AND MERGING DATASETS ###
#rm(dist, pop, sitesjoin)
### CALCULATE TRAVEL COST ####
x1$cost <- 0
x1$duration[ x1$siteID == 351] <- x1$duration[ x1$siteID == 351] + 40 ## this sites needs 40min walking to be reached
## Need to also do this for Hawaii
### Change here for different fuel cost
x1$cost <- x1$distance.meters*0.001 * (0.20 / 1.5 ) ## fuel cost
## assumption on fuel cost:
#
# 20 miles/gallon
# $4.3/gallon
# means $4.3 per 20 miles --> 0.215 $/mile
# 10 km --> 6.2 miles --> $1.33 travel cost per km
# 10km * 0.20 / 1.5 = $1.33 so the values coincide with the cost used above.
x1$cost <- x1$cost + (x1$duration / 60)* 3/4 * 24.4
x1$cost <- x1$cost * 2 ## the value for a return trip
#### put Molokini cost equal to the one of Makena Landing Beach Park + 100$ (cost of a snorkeling tour)
x1$cost[x1$siteID == 388] <- x1$cost[x1$siteID == 359] + 100
### PREPARE DATA FOR PREDICTIONS
### parameters
beta <- m1$results[,1]
X0 <- cbind( x1$cost,x1$home, x1$coastal, x1$citypark, x1$trailall,
x1$haleakala, x1$molokini, x1$parking, x1$showers,
x1$lifeguard, x1$rock, x1$manmade, x1$beachmedium, x1$beachlarge, x1$surf,
x1$swim, x1$snork, x1$res_fish_bio, x1$PNAS, x1$playground, x1$sports,
x1$snork*x1$PNAS, x1$snork*x1$res_fish_bio)
X0[is.na(X0)] <- 0
xsim <- x1
final <- data.frame(matrix(ncol =2, nrow = 0))
all <- data.frame(matrix(ncol =2, nrow = 0))
#colnames(final) <- c("time10", "time20" , "time30" , "time40")
#this needs to be ran as a loop
for (i in 2:83) {
xsim <- x1
xsim$PNAS <- xsim$PNAS *xsim[,i]
xsim$PNAS[xsim$PNAS <= 0] <- 0
X1 <- cbind( xsim$cost,xsim$home, xsim$coastal, xsim$citypark, xsim$trailall,
xsim$haleakala, xsim$molokini, xsim$parking, xsim$showers,
xsim$lifeguard, xsim$rock, xsim$manmade, xsim$beachmedium, xsim$beachlarge, xsim$surf,
xsim$swim, xsim$snork, xsim$res_fish_bio, xsim$PNAS, xsim$playground, xsim$sports,
xsim$snork*xsim$PNAS, xsim$snork*xsim$res_fish_bio)
X1[is.na(X1)] <- 0
### CALCULATION OF WTP FOR EACH GRID LOCATION
wtpw <- quality.change(beta = beta, ### parameters
X0=X0, ### X variables in the baseline
X1 = X1, #### X variables in the scenario
pid = x1$km2_ID, ### ID of the 1km grid
cost =1) ### column of the cost parameter in the X matrix
### TOTAL WTP
b=merge(wtpw,out, by.x=c("id"),by.y=c("km2_ID"))
b$time=i
a=b %>% group_by(island) %>% summarize(SSP3_block=sum(WTP *grid_km_censusblock_Pop18_densityKM * 365)
)
a$time=i
all=rbind(b,all)
final=rbind(a,final)
}
df=merge(df,final, all.x = TRUE)
colnames(all)[2]="SSP3"
all_df=merge(all_df, all, all.x = TRUE)Two datasets were created in the prediction modeling.
all_df$SSP1_TotalWTP=(all_df$SSP1*all_df$grid_km_censusblock_Pop18_densityKM * 365)
all_df$SSP2_TotalWTP=(all_df$SSP2*all_df$grid_km_censusblock_Pop18_densityKM * 365)
all_df$SSP3_TotalWTP=(all_df$SSP3*all_df$grid_km_censusblock_Pop18_densityKM * 365)
all_df$Year=all_df$time+2018
saveRDS(all_df, file = "codedata/all_grids.rds")This code snippet produces a projected summary table of total welfare loss by island, under each scenario (SSP1–SSP3). The results are aggregated over time and space, incorporating:
Annual compensating variation (CV) estimates at the grid level
Population projections for the state of Hawai’i under SSP1–SSP3, used to scale welfare impacts
Two discounting frameworks:
A standard constant discount rate (e.g., circular A-4 rate, typically 3%)
A pluralistic (declining or hyperbolic) discount rate, reflecting alternative time preferences in environmental valuation
These discounted totals provide a long-term estimate of the economic cost of environmental degradation (via coral biomass loss) under different future pathways. The final output summarizes welfare changes per island, scenario, and discounting method, supporting policy evaluation and prioritization.
df_long <- all_df[c(1:11)] %>%
pivot_longer(cols = starts_with("SSP"),
names_to = "SSP",
values_to = "Value")
#points_sf_long$geometry=NULL
#points_sf_long$pct_change[is.nan(points_sf_long$pct_change)] <- 0
df_long$Year=df_long$time+2018
#df_long=merge(df_long, points_sf_long[c(5:7,10)] ,by.x=c("id","Year","SSP"), by.y=c("km2_ID","year","scenario"), all.x=TRUE)
library(dplyr)
library(zoo)
#df_long <- df_long %>% group_by(id, SSP) %>% arrange(year) %>% mutate( pct_change = if_else( SSP != "SSP1", zoo::na.approx(pct_change, x = Year, na.rm = FALSE), pct_change )) %>% ungroup()
#df_long$island=droplevels(df_long$island)
df_long$island <- as.character(df_long$island)
df_long$island[which(df_long$island=="Oahu")]="Oʻahu"
df_long$island[which(df_long$island=="Molokai")]="Molokaʻi"
df_long$island[which(df_long$island=="Lanai")]="Lānaʻi"
df_long$island[which(df_long$island=="Kauai")]="Kauaʻi"
df_long$island[which(df_long$island=="Hawaii")]="Hawaiʻi"
## Merge Population Of Hawaii Under SSP
df_long=merge(df_long,pop[c(7,8,11)], by.x=c("Year", "SSP"), by.y=c("Year","Scenario"), all.x=TRUE)
df_long <- df_long %>%
group_by(SSP) %>%
arrange(Year, .by_group = TRUE) %>%
mutate(Year_j = Year + seq_along(Year)*1e-9) %>%
mutate(pop = na.approx(Population_weight, Year_j, na.rm = FALSE)) %>%
select(-Year_j) %>%
ungroup()
# Start Discounting
df_long$NPV_Lane=0
df_long$NPV_Lane[which(df_long$Year>2020)]=.03
df_long$NPV=0
df_long$NPV[which(df_long$Year>2020)]=.02
df_long$NPV[which(df_long$Year>2079)]=.019
df_long$NPV[which(df_long$Year>2094)]=.018
df_long$time_step <- df_long$Year - 2020
df_long$time_step=df_long$time-2
df_long <- df_long %>%
mutate( time_step = case_when(
time_step %in% 1:5 ~ 0,
time_step >= 6 ~ time_step - 5,
TRUE ~ time_step) )
# value into annual terms by population km square
df_long$pop=as.numeric(df_long$pop)
df_long$wtp=df_long$Value*df_long$grid_km_censusblock_Pop18_densityKM*365
#df_long$wtp=df_long$Value*df_long$grid_km_censusblock_Pop18_densityKM*365
df_long$NPV_WTP=df_long$wtp/(1+df_long$NPV)^df_long$time_step
##OMB discounting long term decline
df_long$NPV_WTP_Pop=(df_long$wtp*df_long$pop)/(1+df_long$NPV)^df_long$time_step
df_long$NPV_WTP_Lane=(df_long$wtp*df_long$pop)/(1+df_long$NPV_Lane)^df_long$time_step
##pluristic discounting
df_long$NPV_WTP_pluristic=(df_long$wtp*df_long$pop*.4)/(1+0)^df_long$time_step+(df_long$wtp*df_long$pop*.1)/(1)^df_long$time_step+(df_long$wtp*df_long$pop*.3)/(1+.03)^df_long$time_step+(df_long$wtp*df_long$pop*.2)/(1+.1)^df_long$time_step
df_long$NPV_WTP_pluristic_nonpop=(df_long$wtp*.4)/(1+0)^df_long$time_step+(df_long$wtp*.1)/(1)^df_long$time_step+(df_long$wtp*.3)/(1+.03)^df_long$time_step+(df_long$wtp*.2)/(1+.1)^df_long$time_step
#df_long$NPV_WTP_kmgrid=(df_long$wtp*df_long$pct_change)/(1+df_long$NPV)^df_long$time_step# Adjust for inflation from original data 2017 to 2014
library(knitr)
t=df_long %>% group_by(SSP) %>%summarise(NPV_WTP=sum(NPV_WTP)/1000000000*1.249,NPV_WTP_Pop=sum(NPV_WTP_Pop)/1000000000*1.249, NPV_WTP_pluristic_nonpop=sum(NPV_WTP_pluristic_nonpop)/1000000000*1.249,NPV_WTP_pluristic=sum(NPV_WTP_pluristic)/1000000000*1.249)
kable(t, caption = "Summary of NPV_WTP_pluristic_nonpop", digits = 2)The raw welfare estimates generated in previous steps are not yet adjusted for differences in income levels or population density. The next code snippet computes total willingness to pay (WTP) by:
Aggregating individual welfare estimates by income tract and grid cell
Weighting by local population to reflect the total affected individuals
This adjusted dataset is especially useful for understanding distributional impacts of environmental change and for producing spatially explicit visualizations of welfare loss.
In the paper we present 2030, 2050, 2100 but here we have also 2075
# Define the cutpoints for 7 intervals
hawaii_counties=st_read("codedata/Coastline.shp")
hawaii_counties <- sf::st_transform(hawaii_counties, 4326)
all_df <- readRDS("codedata/all_grids.rds")
locations=x1[c(87,90,91)]
locations=unique(locations)
all_df=merge(all_df, locations, by.x=c("id"),by.y=c("km2_ID"))
all_df <- all_df %>%
mutate(SSP1_WTP_bin = case_when(
SSP1_TotalWTP <= -1e6 ~ "≤ -1M",
SSP1_TotalWTP <= -1e5 ~ "-1M to -100K",
SSP1_TotalWTP <= -1e3 ~ "-100K to -1K",
SSP1_TotalWTP <= -100 ~ "-1K to -100",
SSP1_TotalWTP < 0 ~ "-100 to 0",
SSP1_TotalWTP == 0 ~ "0",
SSP1_TotalWTP <= 100 ~ "0 to 100",
SSP1_TotalWTP <= 1e3 ~ "100 to 1K",
SSP1_TotalWTP <= 1e4 ~ "1K to 10K",
TRUE ~ "> 10K"
))
all_df <- all_df %>%
mutate(SSP2_WTP_bin = case_when(
SSP2_TotalWTP <= -1e6 ~ "≤ -1M",
SSP2_TotalWTP <= -1e5 ~ "-1M to -100K",
SSP2_TotalWTP <= -1e3 ~ "-100K to -1K",
SSP2_TotalWTP <= -100 ~ "-1K to -100",
SSP2_TotalWTP < 0 ~ "-100 to 0",
SSP2_TotalWTP == 0 ~ "0",
SSP2_TotalWTP <= 100 ~ "0 to 100",
SSP2_TotalWTP <= 1e3 ~ "100 to 1K",
SSP2_TotalWTP <= 1e4 ~ "1K to 10K",
TRUE ~ "> 10K"
))
all_df <- all_df %>%
mutate(SSP3_WTP_bin = case_when(
SSP3_TotalWTP <= -1e6 ~ "≤ -1M",
SSP3_TotalWTP <= -1e5 ~ "-1M to -100K",
SSP3_TotalWTP <= -1e3 ~ "-100K to -1K",
SSP3_TotalWTP <= -100 ~ "-1K to -100",
SSP3_TotalWTP < 0 ~ "-100 to 0",
SSP3_TotalWTP == 0 ~ "0",
SSP3_TotalWTP <= 100 ~ "0 to 100",
SSP3_TotalWTP <= 1e3 ~ "100 to 1K",
SSP3_TotalWTP <= 1e4 ~ "1K to 10K",
TRUE ~ "> 10K"
))
color_palette <- c(
"≤ -1M" = "darkred",
"-1M to -100K" = "brown",
"-100K to -1K" = "lightcoral",
"-1K to -100" = "pink",
"-100 to 0" = "mistyrose",
"0" = "white",
"0 to 100" = "greenyellow",
"100 to 1K" = "limegreen",
"> 10K" = "darkgreen"
)
all_df$SSP1_WTP_bin <- factor(
all_df$SSP1_WTP_bin,
levels = names(color_palette)
)
all_df$SSP2_WTP_bin <- factor(
all_df$SSP2_WTP_bin,
levels = names(color_palette)
)
all_df$SSP2_WTP_bin <- factor(
all_df$SSP2_WTP_bin,
levels = names(color_palette)
)
# Then plot:
ggplot(all_df %>% filter(Year %in% c(2020, 2050, 2075, 2100)),
aes(x = longitude, y = latitude, color = SSP1_WTP_bin)
) +
geom_point(size = 0.5, alpha = 1) +
geom_sf(
data = hawaii_counties,
fill = NA,
color = "black",
size = 0.4,
inherit.aes = FALSE
) +
coord_sf(xlim = c(-161, -154), ylim = c(18.5, 22.5)) +
scale_color_manual(
values = color_palette,
name = "Total WTP",
drop = FALSE,
limits = names(color_palette)
) +
facet_wrap(vars(Year), ncol = 2) +
theme_light()These produce animations of changes in annual welfare to the year 2100
Legend for the following animations
color_palette <- c(
"≤ -1M" = "darkred",
"-1M to -100K" = "brown",
"-100K to -1K" = "lightcoral",
"-1K to -100" = "pink",
"-100 to 0" = "mistyrose",
"0" = "white",
"0 to 100" = "greenyellow",
"100 to 1K" = "limegreen",
"> 10K" = "darkgreen"
)
# Create a blank plot canvas
plot(
NULL, xaxt = "n", yaxt = "n", bty = "n",
xlab = "", ylab = "", xlim = c(0, 1), ylim = c(0, 1)
)
# Draw the standalone legend
legend(
"center",
legend = names(color_palette),
fill = unname(color_palette),
bty = "o", # border around legend
cex = 1.0, # text size
title = "Total WTP",
title.cex = 1.2, # title size
ncol = 1,
xpd = TRUE
)
2030>>>>2100
library(ggplot2)
library(gganimate)
base_map <- ggplot(all_df %>% filter(Year >= 2020 & Year <= 2100)) +
geom_point(aes(x = longitude, y = latitude, color = SSP1_WTP_bin),
size = 0.5, alpha = 1) + geom_sf(data = hawaii_counties, fill = NA, color = "black", size = 0.5) +
coord_sf(xlim = c(-161, -154), ylim = c(18.5, 22.5)) +
scale_color_manual(
values = color_palette,
name = "Total WTP",
drop = FALSE,
limits = names(color_palette)
) +
theme_light(base_size = 14) +
labs(title = 'SSP1 Year: {closest_state}')
anim <- base_map +
transition_states(Year, transition_length = 1, state_length = 1) +
ease_aes('cubic-in-out') +
shadow_wake(alpha = 0.3, wake_length = 0.2)
gganimate::animate(
anim,
nframes = 81,
fps = 10,
width = 800,
height = 600,
renderer = gifski_renderer()
)
gganimate::anim_save(
"hawaii_wtpssp1_animation.gif",
animation = gganimate::last_animation()
)2030>>>>2100
base_map <- ggplot(all_df %>% filter(Year >= 2020 & Year <= 2100)) +
geom_point(aes(x = longitude, y = latitude, color = SSP2_WTP_bin),
size = 0.5, alpha = 1) + geom_sf(data = hawaii_counties, fill = NA, color = "black", size = 0.5) +
coord_sf(xlim = c(-161, -154), ylim = c(18.5, 22.5)) +
scale_color_manual(
values = color_palette,
name = "Total WTP",
drop = FALSE,
limits = names(color_palette)
) +
theme_light(base_size = 14) +
labs(title = 'SSP2 Year: {closest_state}')
anim <- base_map +
transition_states(Year, transition_length = 1, state_length = 1) +
ease_aes('cubic-in-out') +
shadow_wake(alpha = 0.3, wake_length = 0.2)
gganimate::animate(
anim,
nframes = 81,
fps = 10,
width = 800,
height = 600,
renderer = gifski_renderer()
)
gganimate::anim_save(
"hawaii_wtpssp2_animation.gif",
animation = gganimate::last_animation()
)2030>>>>2100
base_map <- ggplot(all_df %>% filter(Year >= 2020 & Year <= 2100)) +
geom_point(aes(x = longitude, y = latitude, color = SSP3_WTP_bin),
size = 0.5, alpha = 1) + geom_sf(data = hawaii_counties, fill = NA, color = "black", size = 0.5) +
coord_sf(xlim = c(-161, -154), ylim = c(18.5, 22.5)) +
scale_color_manual(
values = color_palette,
name = "Total WTP",
drop = FALSE,
limits = names(color_palette)
) +
theme_light(base_size = 14) +
labs(title = 'SSP3 Year: {closest_state}')
anim <- base_map +
transition_states(Year, transition_length = 1, state_length = 1) +
ease_aes('cubic-in-out') +
shadow_wake(alpha = 0.3, wake_length = 0.2)
gganimate::animate(
anim,
nframes = 81,
fps = 10,
width = 800,
height = 600,
renderer = gifski_renderer()
)
gganimate::anim_save(
"hawaii_wtpssp3_animation.gif",
animation = gganimate::last_animation()
)dishawaii=read_sf("/codedata/hawaii_statepercentiles.shp")
ejscreen_data <- dishawaii %>%
rowwise() %>%
mutate(
env_burden = any(c_across(c(P_PM25,P_OZONE,P_DSLPM,P_RSEI_AIR,P_PTRAF,
P_LDPNT,P_PNPL,P_PRMP,P_PTSDF,P_UST,P_PWDIS,P_NO2,P_DWATER)) >= 90, na.rm = TRUE),
soc_vulnerable = any(c_across(c(P_PEOPCOLO,P_LOWINCPC,P_UNEMPPCT,P_DISABILI,P_LINGISOP,P_LESSHSPC,P_UNDER5PC,P_OVER64PC,P_LIFEEXPP)) >= 90, na.rm = TRUE),
Disadvantage_block_either = as.integer(env_burden | soc_vulnerable, 1, 0),
Disadvantage_block = as.integer(env_burden & soc_vulnerable, 1, 0)
) %>%
ungroup()
ejscreen_data=ejscreen_data[c(1:4,232:235)]
ejscreen_data <- st_transform(ejscreen_data, 4326)
latlon <- x1[order(x1$siteID),]
latlon=latlon[c(85,87,90:91)]
library(sf)
library(dplyr)
latlon=distinct(latlon)
latlon <- st_as_sf(latlon, coords = c("longitude", "latitude"), crs = 4326)
nearest_idx <- st_nearest_feature(latlon, ejscreen_data)
# Join attributes from the nearest polygon
latlon_joined <- cbind(
latlon,
ejscreen_data[nearest_idx, ]
)
latlon_joined=data.frame(latlon_joined)
dishawaii=read_sf("/Users/ashleylowe/EJ/EPA_IRA_Disadvantage_Hawaii.shp")
dishawaii <- st_transform(dishawaii, 4326)
nearest_idx <- st_nearest_feature(latlon, dishawaii)
# Join attributes from the nearest polygon
latlon_joined <- cbind(
latlon_joined,
dishawaii[nearest_idx, ]
)
all_df_sf=merge(all_df_time,latlon_joined, by.x = c("id","island"),by.y = c("km2_ID", "island"))
all_df_sf$Year=all_df$time+2018
#all_df_sf=merge(latlon, all_df, by.y = c("id","island"),by.x = c("km2_ID", "island"))
summary_table <- all_df %>%
st_drop_geometry() %>%
group_by(env_burden, island) %>%
summarise(
variance = var(avg_welfare_loss_zip),
median = median(avg_welfare_loss_zip),
mean = mean(avg_welfare_loss_zip),
.groups = "drop"
) %>%
mutate(
env_burden = ifelse(env_burden, "Disadvantaged", "Not Disadvantaged")
)
all_df_sf %>%
filter(Year == 2030) %>%
group_by(Disadvanta, island) %>%
summarise(
total_pop = sum(grid_km_censusblock_Pop18_densityKM, na.rm = TRUE),
total_welfare = sum(
SSP1 * 365 * 1.249 *
grid_km_censusblock_Pop18_densityKM,
na.rm = TRUE
),
welfare_per_person = total_welfare / total_pop,
average_welfare = mean(
SSP1 * 365 * 1.24,
na.rm = TRUE
),
.groups = "drop"
)