library(tidyverse)
library(tmap)
library(units)
library(sf)
library(leaflet)
library(dbscan)
library(sfnetworks)
library(tigris)
library(tidygraph)
library(plotly)
library(osmdata)
library(here)
library(tidytransit)
library(tidycensus)
library(leafsync)

epsg <- 4326

Step 1. Download Required data from GTFS.

# TASK ////////////////////////////////////////////////////////////////////////
# Download MARTA (Metropolitan Atlanta Rapid Transit Authority) GTFS data using `read_gtfs()` function and assign it to `gtfs` object
gtfs <- read_gtfs("https://www.itsmarta.com/google_transit_feed/google_transit.zip")
# //TASK //////////////////////////////////////////////////////////////////////


# =========== NO MODIFICATION ZONE STARTS HERE ===============================
# Edit stop_name to append serial numbers (1, 2, etc.) to remove duplicate names
stop_dist <- stop_group_distances(gtfs$stops, by='stop_name') %>%
  filter(dist_max > 200)

gtfs$stops <- gtfs$stops %>% 
  group_by(stop_name) %>% 
  mutate(stop_name = case_when(stop_name %in% stop_dist$stop_name ~ paste0(stop_name, " (", seq(1,n()), ")"),
                               TRUE ~ stop_name))

# Create a transfer table
gtfs <- gtfsrouter::gtfs_transfer_table(gtfs, 
                                        d_limit = 200, 
                                        min_transfer_time = 120)

# NOTE: Converting to sf format uses stop_lat and stop_lon columns contained in gtfs$stops.
#       In the conversion process, stop_lat and stop_lon are converted into a geometry column, and
#       the output sf object do not have the lat lon column anymore.
#       But many other functions in tidytransit look for stop_lat and stop_lon.
#       So I re-create them using mutate().
gtfs <- gtfs %>% gtfs_as_sf(crs = epsg)

gtfs$stops <- gtfs$stops %>% 
  ungroup() %>% 
  mutate(stop_lat = st_coordinates(.)[,2],
         stop_lon = st_coordinates(.)[,1]) 

# Get stop_id for rails and buses
rail_stops <- gtfs$routes %>% 
  filter(route_type %in% c(1)) %>% 
  inner_join(gtfs$trips, by = "route_id") %>% 
  inner_join(gtfs$stop_times, by = "trip_id") %>% 
  inner_join(gtfs$stops, by = "stop_id") %>% 
  group_by(stop_id) %>% 
  slice(1) %>% 
  pull(stop_id)

# Extract MARTA rail stations
station <- gtfs$stops %>% filter(stop_id %in% rail_stops)

# Extract Midtown Station
midtown <- gtfs$stops %>% filter(stop_id == "134")

# Create a bounding box to which we limit our analysis
bbox <- st_bbox(c(xmin = -84.45241, ymin = 33.72109, xmax = -84.35009, ymax = 33.80101), 
                 crs = st_crs(4326)) %>% 
  st_as_sfc()


# =========== NO MODIFY ZONE ENDS HERE ========================================

Step 2. Download Required data from Census

# TASK ////////////////////////////////////////////////////////////////////////
# Specify Census API key whichever you prefer using census_api_key() function
tidycensus::census_api_key(Sys.getenv("census_api"))
# census_api_key(" **YOUR CODE HERE..** ")
# //TASK //////////////////////////////////////////////////////////////////////



# TASK ////////////////////////////////////////////////////////////////////////
# Using get_acs() function, download Census Tract level data for 2022 for Fulton, DeKalb, and Clayton in GA.
# and assign it to `census` object.
# Make sure you set geometry = TRUE.

# Required data from the Census ACS:
#  1) Median Household Income (name the column `hhinc`)
#  2) Minority Population (%) (name the column `pct_minority`)
# Note: You may need to download two or more Census ACS variables to calculate minority population (%). "Minority" here can refer to either racial minorities or racial+ethnic minorities -- it's your choice.

census <- suppressMessages(
  get_acs(geography = "tract",
          state = "GA",
          county = c("Dekalb", "Fulton", "Clayton"), 
          variables = c(hhinc = 'B19019_001E',
                        race_all = "B02001_001",
                        white = "B02001_002"),
          year = 2021,
          survey = "acs5",
          geometry = TRUE,
          output = "wide")
) %>% mutate(pct_minority = (race_allE - whiteE) / race_allE)

# //TASK //////////////////////////////////////////////////////////////////////



# =========== NO MODIFICATION ZONE STARTS HERE ===============================
census <- census %>% 
  st_transform(crs = 4326) %>% 
  separate(col = NAME, into = c("tract", "county", "state"), sep = ", ")

# Convert it to POINT at polygon centroids and extract those that fall into bbox
# and assign it into `home` object
home <- census %>% st_centroid() %>% .[bbox,]
# =========== NO MODIFY ZONE ENDS HERE ========================================

Step 3. Download Required data from OSM.

# TASK ////////////////////////////////////////////////////////////////////////
# 1. Get OSM data using opq() function and bbox object defined in the previous code chunk.
# 2. Specify arguments for add_osm_feature() function using 
#    key = 'highway' and 
#    value = c("motorway", "trunk", "primary", "secondary", "tertiary", "residential", 
#              "motorway_link", "trunk_link", "primary_link", "secondary_link", 
#              "tertiary_link", "residential_link", "unclassified")
# 3. Convert the OSM data into an sf object using osmdata_sf() function
# 4. Convert osmdata polygons into lines using osm_poly2line() function

osm_road <- opq(bbox = bbox) %>%
  add_osm_feature(key = "highway",
                  value = c("motorway", "trunk", "primary", 
                            "secondary", "tertiary", "unclassified",
                            "residential")) %>%
  osmdata_sf() %>% 
  osm_poly2line()
# //TASK //////////////////////////////////////////////////////////////////////


# TASK ////////////////////////////////////////////////////////////////////////
# 1. Convert osm_road$osm_lines into sfnetwork using as_sfnetwork() function
# 2. Activate edges
# 3. Clean the network using edge_is_multiple(), edge_is_loop(), to_spatial_subdivision(), to_spatial_smooth()
# 4. Assign the cleaned network to an object named 'osm'
osm <- osm_road$osm_line %>%
  select(osm_id, highway) %>% 
  sfnetworks::as_sfnetwork(directed = FALSE) %>% 
  activate('edges') %>%
  filter(!edge_is_multiple()) %>%
  filter(!edge_is_loop()) %>% 
  convert(., sfnetworks::to_spatial_subdivision) %>% # subdivide edges
  convert(., sfnetworks::to_spatial_smooth) # delete pseudo nodes  
# //TASK //////////////////////////////////////////////////////////////////////


# TASK ////////////////////////////////////////////////////////////////////////
# Add a new column named 'length' to the edges part of the object `osm`.
osm <- osm %>% mutate(
  length = edge_length()
)
# //TASK //////////////////////////////////////////////////////////////////////

Step 4. Simulate a park-and-ride trip (home -> closest station -> Midtown station).

# =========== NO MODIFICATION ZONE STARTS HERE ===============================
# Extract the first row from `home` object and store it `home_1`
home_1 <- home[1,]
# =========== NO MODIFY ZONE ENDS HERE ========================================


# TASK ////////////////////////////////////////////////////////////////////////
# Find the shortest path from `home_1` to all other stations
# using st_network_paths() function.
paths <- st_network_paths(osm, from = home_1, to = station, type = "shortest")
## Warning in shortest_paths(x, from, to, weights = weights, output = "both", : At
## vendor/cigraph/src/paths/dijkstra.c:534 : Couldn't reach some vertices.
# //TASK //////////////////////////////////////////////////////////////////////

  
# =========== NO MODIFICATION ZONE STARTS HERE ===============================
# Using the `paths` object, get network distances from `home_1` to all other stations.
dist_all <- map_dbl(1:nrow(paths), function(x){
  osm %>%
    activate("nodes") %>% 
    slice(paths$node_paths[[x]]) %>% 
    st_as_sf("edges") %>% 
    pull(length) %>% 
    sum()
}) %>% unlist() 

# Replace zeros with a large value.
if (any(dist_all == 0)){
  dist_all[dist_all == 0] <- max(dist_all)
}

# Find the closest station.
closest_index <- which.min(dist_all)
closest_station <- station[closest_index,]

# Find the distance to the closest station.
closest_dist <- min(dist_all)

# Calculate how long it takes to traverse `closest_dist` 
# assuming we drive at 30 miles/hour speed.
# Store the output in trvt_osm_m.
car_speed <- set_units(30, mile/h)
trvt_osm_m <- closest_dist/set_units(car_speed, m/min) %>%  # Distance divided by 30 mile/h
  as.vector(.)
# =========== NO MODIFY ZONE ENDS HERE ========================================


# TASK ////////////////////////////////////////////////////////////////////////
# 1. From `osm` object, activate nodes part and
# 2. use `closest_index` to extract the selected path
paths_closest <- osm %>%
  activate("nodes") %>%
  slice(paths$node_paths[[closest_index]])
# //TASK //////////////////////////////////////////////////////////////////////


# TASK ////////////////////////////////////////////////////////////////////////
# Use filter_stop_times() function to create a subset of stop_times data table
# for date = 2024-11-14, minimum departure time of 7AM, maximum departure time of 10AM.
# Assign the output to `am_stop_time` object
am_stop_time <- filter_stop_times(gtfs_obj = gtfs, 
                                extract_date = "2024-11-14",
                                min_departure_time = 3600*7,
                                max_arrival_time = 3600*10) 
# //TASK //////////////////////////////////////////////////////////////////////



# TASK ////////////////////////////////////////////////////////////////////////
# 1. Use travel_times() function to calculate travel times from the `closest_station` 
#    to all other stations during time specified in am_stop_time. Allow ONE transfer.
# 2. Filter the row for which the value of 'to_stop_name' column 
#    equals midtown$stop_name. Assign it into `trvt` object.
trvt <- travel_times(
  filtered_stop_times = am_stop_time,
  stop_name = closest_station$stop_name,
  time_range = 3600,        
  arrival = FALSE,          
  max_transfers = 1,        
  return_coords = TRUE
) %>%
  filter(to_stop_name == midtown$stop_name)
  
# //TASK //////////////////////////////////////////////////////////////////////

# =========== NO MODIFICATION ZONE STARTS HERE ===============================
# Divide the calculated travel time by 60 to convert the unit from seconds to minutes.
trvt_gtfs_m <- trvt$travel_time/60

# Add the travel time from home to the nearest station and
# the travel time from the nearest station to Midtown station
total_trvt <- trvt_osm_m + trvt_gtfs_m
# =========== NO MODIFY ZONE ENDS HERE ========================================

Step 5. Convert Step 4 into a function

# Function definition (do not modify other parts of the code in this code chunk except for those inside the TASK section)

get_trvt <- function(home, osm, station, midtown){
  
  # TASK ////////////////////////////////////////
  # If the code in Step 4 runs fine,
  # Replace where it says **YOUR CODE HERE..** below with 
  # the ENTIRETY of the code in the previous code chunk (i.e., Step 4)
  home_1 <- home[1,]

  paths <- st_network_paths(osm, from = home_1, to = station, type = "shortest")
  
  dist_all <- map_dbl(1:nrow(paths), function(x){
    osm %>%
      activate("nodes") %>% 
      slice(paths$node_paths[[x]]) %>% 
      st_as_sf("edges") %>% 
      pull(length) %>% 
      sum()
  }) %>% unlist()
  
  if (any(dist_all == 0)){
    dist_all[dist_all == 0] <- max(dist_all)
  }
  
  closest_index <- which.min(dist_all)
  closest_station <- station[closest_index,]
  
  closest_dist <- min(dist_all)
  
  car_speed <- set_units(30, mile/h)
  trvt_osm_m <- closest_dist / set_units(car_speed, m/min) %>% as.vector(.)
  
  paths_closest <- osm %>%
    activate("nodes") %>%
    slice(paths$node_paths[[closest_index]])
  
  am_stop_time <- filter_stop_times(
    gtfs_obj = gtfs,
    extract_date = "2024-11-14",
    min_departure_time = 3600 * 7,  # 7AM
    max_arrival_time = 3600 * 10    # 10AM
  )
  
  # Use travel_times() function to calculate travel times from the `closest_station` 
  # to all other stations during time specified in am_stop_time. Allow ONE transfer.
  trvt <- travel_times(
    filtered_stop_times = am_stop_time,
    stop_name = closest_station$stop_name,
    time_range = 3600,        # 1 hour window
    arrival = FALSE,          # Start journey from the closest station
    max_transfers = 1,        # Allow 1 transfer
    return_coords = TRUE
  ) %>%
    filter(to_stop_name == midtown$stop_name)

  trvt_gtfs_m <- trvt$travel_time / 60
  
  total_trvt <- trvt_osm_m + trvt_gtfs_m
  
  # //TASK //////////////////////////////////////

  # =========== NO MODIFICATION ZONE STARTS HERE ===============================
  if (length(total_trvt) == 0) {total_trvt = 0}

  return(total_trvt)
  # =========== NO MODIFY ZONE ENDS HERE ========================================
}

Step 6. Apply the function for the whole study area

# Prepare an empty vector
total_trvt <- vector("numeric", nrow(home))

# Apply the function for all Census Tracts
# Fill `total_trvt` object with the calculated time
for (i in 1:nrow(home)){
  total_trvt[i] <- get_trvt(home[i,], osm, station, midtown)
}

# Cbind the calculated travel time back to `home`
home_done <- home %>% 
  cbind(trvt = total_trvt)

Step 7. Create maps and plots

Run the code below to generate thematic maps and plots

Write a short description of what you observe from the maps and plots

# Map
tmap_mode('view')
## tmap mode set to interactive viewing
tm_shape(census[census$GEOID %in% home$GEOID,]) + 
  tm_polygons(col = "hhinc", palette = 'GnBu') + 
  tm_shape(home_done) + 
  tm_dots(col = "trvt", palette = 'Reds', size = 0.1)
tm_shape(census[census$GEOID %in% home$GEOID,]) + 
  tm_polygons(col = "pct_minority", palette = 'GnBu') + 
  tm_shape(home_done) + 
  tm_dots(col = "trvt", palette = 'Reds', size = 0.1)
# ggplot
inc <- ggplot(data = home_done,
              aes(x = hhinc, y = trvt)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  labs(x = "Median Annual Household Income",
       y = "Park-and-ride Travel Time from Home to Midtown Station") +
  theme_bw()

minority <- ggplot(data = home_done,
                   aes(x = pct_minority, y = trvt)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  labs(x = "Minority Population (%)",
       y = "Park-and-ride Travel Time from Home to Midtown Station") +
  theme_bw()

ggpubr::ggarrange(inc, minority)
## `geom_smooth()` using formula = 'y ~ x'
## Warning: Removed 2 rows containing non-finite outside the scale range
## (`stat_smooth()`).
## Warning: Removed 2 rows containing missing values or values outside the scale range
## (`geom_point()`).
## `geom_smooth()` using formula = 'y ~ x'

From the maps, the correlation between household income and travel time, as well as minority population and travel time, is difficult to observe across census tracts. For instance, in the map showing household income and travel time, some tracts with lower household income exhibit shorter travel times, while others show longer travel times.

In the final plot comparing travel times across census tracts by median annual household income and minority population, we see that the first plot, with median household income on the x-axis, has a positive slope. This suggests that as household income increases, so does travel time. This outcome aligns with equitable transit service for Park-and-Ride users, as it implies that lower-income groups benefit from shorter travel times. In contrast, the second plot, which shows travel time relative to minority population, has an almost flat slope, indicating that travel times are similar across tracts regardless of minority population. This points to a model of equality rather than equity, as it implies similar travel times for everyone without considering minority population differences.