Ulaanbaatar District-Level Analysis
Source:vignettes/ub-health-environment.Rmd
ub-health-environment.Rmd
library(mongolstats)
library(mongolmaps)
library(dplyr)
library(ggplot2)
library(sf)
library(tidyr)
# Checked teaching helpers are shipped with the package and can be inspected.
source(system.file("examples", "analysis-helpers.R", package = "mongolstats"), local = TRUE)
source(system.file("examples", "plot-style.R", package = "mongolstats"), local = TRUE)Introduction
Ulaanbaatar, the capital of Mongolia, is home to about half of the country’s population. This vignette looks at its nine districts (duureg) using NSO tables published “by soum and district”: population and growth, school capacity, the air quality monitoring network, and the limits of district-level health statistics.
Districts and Their Codes
mongolmaps provides the district boundaries with their NSO codes and areas, so NSO tables join without matching names.
ub_districts <- mn_ub_districts()
ub_districts |>
st_drop_geometry() |>
select(name, nso_code, area_km2)
#> # A tibble: 9 × 3
#> name nso_code area_km2
#> <chr> <chr> <dbl>
#> 1 Baganuur 51101 625.
#> 2 Bagakhangai 51104 156.
#> 3 Bayangol 51107 24.8
#> 4 Bayanzurkh 51110 1237.
#> 5 Nalaikh 51113 689.
#> 6 Songinokhairkhan 51116 1204.
#> 7 Sukhbaatar 51119 211.
#> 8 Khan-Uul 51122 498.
#> 9 Chingeltei 51125 91.1
# Fetch an NSO table for the nine districts only
get_ub_data <- function(tbl_id, selections = list()) {
data <- nso_data(
tbl_id,
selections = c(list(Region = ub_districts$nso_code), selections),
labels = "en"
) |>
mutate(District = trimws(Region_en))
expected <- tidyr::expand_grid(
Region = ub_districts$nso_code,
Year_en = as.character(selections$Year)
)
analysis_keys(data, c("Region", "Year_en"), expected)
data
}
# Latest year a table covers
latest_year <- function(tbl_id) {
analysis_latest_common_year(tbl_id)
}1. Demographic Profile
Population Density
The population table DT_NSO_0300_002V4 gives the
mid-year resident population down to khoroo level. (This table id names
two tables in the catalogue; mongolstats warns and uses the first match,
which is the khoroo-level table and also includes the districts.)
pop_tbl <- "DT_NSO_0300_002V4"
pop_year <- latest_year(pop_tbl)
pop_data <- get_ub_data(pop_tbl, list(Year = pop_year))
ub_pop_map <- pop_data |>
mn_join(by = "Region", level = "soum", within = "Ulaanbaatar") |>
mutate(density = if_else(is.finite(value) & value > 0 &
is.finite(area_km2) & area_km2 > 0,
value / area_km2, NA_real_
))
stopifnot(nrow(ub_pop_map) == 9L, !anyDuplicated(ub_pop_map$nso_code))
mn_map(
ub_pop_map,
fill = density, trans = "log10", label = TRUE,
title = paste0("Population Density by District, ", pop_year)
) +
labs(
subtitle = "Residents per km² (log scale)",
fill = "Residents\nper km²",
caption = "Source: NSO Mongolia, DT_NSO_0300_002V4"
) +
guides(fill = guide_colourbar(barwidth = grid::unit(85, "mm"), barheight = grid::unit(3, "mm"))) +
analysis_map_theme()Bayangol is the densest district, with about 10,400 residents per km². Bagakhangai is the least dense, at about 28 per km². Districts such as Bayanzurkh and Songinokhairkhan include large areas of mountain and pasture, so a district average hides very dense neighbourhoods. Khoroo-level data show more detail:
khoroo_codes <- nso_dim_values(pop_tbl, "Region")$code |>
stringr::str_subset("^511\\d{4}$") # seven-digit codes are khoroos
khoroo_raw <- nso_data(
pop_tbl, list(Region = khoroo_codes, Year = pop_year),
labels = "en"
)
analysis_keys(khoroo_raw, "Region", tibble(Region = khoroo_codes))
khoroo_status <- nso_value_status(khoroo_raw)
khoroo_check <- khoroo_raw |>
mutate(Region = substr(Region, 1, 5)) |>
summarise(khoroo_sum = analysis_sum(value), .by = Region) |>
full_join(select(pop_data, Region, district_total = value),
by = join_by(Region), relationship = "one-to-one"
) |>
mutate(agrees = is.finite(khoroo_sum) & is.finite(district_total) & khoroo_sum == district_total)
khoroo_check |>
mutate(agrees = if_else(agrees, "Yes", "No")) |>
knitr::kable(
col.names = c("District code", "Khoroo total", "District total", "Totals agree"),
digits = 0, format.args = list(big.mark = ","),
caption = "Population reconciliation before using the illustrative khoroo map"
)| District code | Khoroo total | District total | Totals agree |
|---|---|---|---|
| 51101 | 29,076 | 29,076 | Yes |
| 51104 | 4,401 | 4,401 | Yes |
| 51107 | 257,391 | 257,391 | Yes |
| 51110 | 452,084 | 452,084 | Yes |
| 51113 | 40,781 | 40,781 | Yes |
| 51116 | 341,746 | 341,746 | Yes |
| 51119 | 142,356 | 142,356 | Yes |
| 51122 | 278,532 | 278,532 | Yes |
| 51125 | 131,278 | 131,278 | Yes |
khoroo_pop <- mn_join(khoroo_raw, by = "Region", level = "bag", within = "Ulaanbaatar") |>
mutate(density = if_else(is.finite(value) & value > 0 &
is.finite(area_km2) & area_km2 > 0,
value / area_km2, NA_real_
))
stopifnot(!anyDuplicated(khoroo_pop$nso_code))
unmapped_khoroos <- setdiff(khoroo_raw$Region, khoroo_pop$nso_code)
central <- c("Bayangol", "Bayanzurkh", "Chingeltei", "Khan-Uul", "Songinokhairkhan", "Sukhbaatar")
mn_map(
filter(khoroo_pop, parent_pcode %in% ub_districts$pcode[ub_districts$name %in% central]),
fill = density, trans = "log10",
title = paste0("Population Density by Khoroo, Central Ulaanbaatar, ", pop_year)
) +
labs(subtitle = "Residents per km² (log scale), six central districts", fill = "Residents\nper km²") +
guides(fill = guide_colourbar(barwidth = grid::unit(85, "mm"), barheight = grid::unit(3, "mm"))) +
analysis_map_theme()The khoroo boundaries are unofficial. 0 source codes lack a matched boundary. The reconciliation table does not establish a matching boundary vintage: splits, mergers and changed codes still need review. This is an illustrative density map, not validated fine-scale risk or exposure data. Missing or nonpositive values are not plotted on the log scale.
Population Growth Trends
Rapid growth can strain local health and school services.
trend_years <- as.character(as.integer(pop_year) - 9:0)
pop_trend <- get_ub_data(pop_tbl, list(Year = trend_years)) |>
mutate(Year = as.integer(Year_en))
pop_trend |>
ggplot(aes(x = Year, y = value)) +
geom_line(linewidth = 1, color = "#167D8D") +
geom_point(size = 1.5, color = "#167D8D") +
facet_wrap(~District, ncol = 3) +
scale_y_continuous(labels = scales::label_comma()) +
scale_x_continuous(breaks = as.integer(trend_years)[c(1, 5, 10)]) +
labs(
title = paste0("Resident Population by District, ", min(trend_years), "–", max(trend_years)),
subtitle = "Mid-year resident population",
y = "Population", x = NULL,
caption = "Source: NSO Mongolia, DT_NSO_0300_002V4"
) +
theme(legend.position = "bottom", panel.grid.minor = element_blank())Across the ten annual observations from 2016 to 2025, there are nine elapsed years. The table below shows signed total and annualised changes; negative values are declines. Baganuur had the smallest absolute percentage change. Changes in these totals do not identify migration or service capacity by themselves.
knitr::kable(growth,
col.names = c("District", "Elapsed years", "Total change (%)", "Annualised change (%)"),
digits = c(0, 0, 1, 1), caption = "Population change over the observed interval"
)| District | Elapsed years | Total change (%) | Annualised change (%) |
|---|---|---|---|
| Khan-Uul | 9 | 79.5 | 6.7 |
| Bayanzurkh | 9 | 39.5 | 3.8 |
| Bayangol | 9 | 22.6 | 2.3 |
| Nalaikh | 9 | 12.9 | 1.4 |
| Songinokhairkhan | 9 | 9.7 | 1.0 |
| Bagakhangai | 9 | 6.4 | 0.7 |
| Sukhbaatar | 9 | 3.8 | 0.4 |
| Baganuur | 9 | 1.3 | 0.1 |
| Chingeltei | 9 | -16.1 | -1.9 |
2. School Enrolment
NSO publishes few socioeconomic indicators by district. The number of students per general education school gives a rough measure of pressure on public services.
edu_year <- analysis_latest_common_year(c("DT_NSO_2001_002V2", "DT_NSO_2002_057V2"))
schools <- get_ub_data("DT_NSO_2001_002V2", list(Year = edu_year)) |>
select(Region, District, Schools = value)
students <- get_ub_data("DT_NSO_2002_057V2", list(Year = edu_year)) |>
select(Region, Students = value)
edu_load <- analysis_pair(schools, students, "Region")
analysis_count_values(edu_load$Schools)
analysis_count_values(edu_load$Students)
stopifnot(
all(is.finite(edu_load$Schools)), all(edu_load$Schools > 0),
all(is.finite(edu_load$Students))
)
edu_load <- mutate(edu_load, Students_Per_School = Students / Schools)
ggplot(edu_load, aes(x = Students_Per_School, y = reorder(District, Students_Per_School))) +
geom_col(fill = "#167D8D", width = 0.7) +
geom_text(aes(label = round(Students_Per_School)), hjust = -0.1, size = 3) +
scale_x_continuous(labels = scales::label_comma(), expand = expansion(mult = c(0, 0.15))) +
labs(
title = paste0("Students per General Education School, ", edu_year),
subtitle = "Full-time students divided by number of schools",
x = "Students per school", y = NULL,
caption = "Source: NSO Mongolia, DT_NSO_2001_002V2 and DT_NSO_2002_057V2"
) +
theme(panel.grid.major.y = element_blank(), panel.grid.minor = element_blank())This ratio is mean enrolment per institution, not classroom size, spare capacity, crowding or deprivation. Schools differ in size, sector and shift system. Confirm that the school-year definition and institutional coverage match between the two sources.
3. Health Statistics by District: A Caution
Rates need a numerator and denominator for the same population. Very uneven recorded births or deaths per resident can reveal incompatible geographic assignment or coverage. The diagnostic below is not a comparison of resident health risks; source-specific residence versus occurrence definitions still need confirmation.
vital_year <- analysis_latest_common_year(c("DT_NSO_2100_027V1", "DT_NSO_2100_018V6", pop_tbl))
births <- get_ub_data(
"DT_NSO_2100_018V6",
list(Indicator = "0", Year = vital_year) # Indicator "0" = both sexes
) |>
select(Region, District, count = value) |>
mutate(measure = "Live births per 1,000 residents")
deaths <- get_ub_data(
"DT_NSO_2100_027V1",
list(Indicator = "0", Year = vital_year) # Indicator "0" = all deaths
) |>
select(Region, District, count = value) |>
mutate(measure = "Deaths per 1,000 residents")
residents <- get_ub_data(pop_tbl, list(Year = vital_year)) |>
select(Region, residents = value)
vital_rates <- bind_rows(births, deaths) |>
left_join(residents, by = join_by(Region), relationship = "many-to-one") |>
mutate(rate = count / residents * 1000)
stopifnot(
all(is.finite(vital_rates$count)), all(is.finite(vital_rates$residents)),
all(vital_rates$residents > 0)
)
vital_rates |>
ggplot(aes(x = rate, y = reorder(District, residents))) +
geom_col(fill = "#5B6E9A", width = 0.7) +
geom_text(aes(label = round(rate, 1)), hjust = -0.1, size = 3) +
facet_wrap(~measure, scales = "free_x") +
scale_x_continuous(expand = expansion(mult = c(0, 0.2))) +
labs(
title = paste0("Recorded Births and Deaths per 1,000 Residents, ", vital_year),
subtitle = "Districts ordered by number of residents",
x = NULL, y = NULL,
caption = "Source: NSO Mongolia, DT_NSO_2100_018V6, DT_NSO_2100_027V1 and DT_NSO_0300_002V4"
) +
theme(panel.grid.major.y = element_blank(), panel.grid.minor = element_blank())The observed differences are consistent with occurrence-based reporting or another mismatch in coverage. They do not establish the precise assignment rule. Do not interpret the diagnostic ratios as resident birth or death rates until NSO or the Health Development Center confirms the definitions for each table.
Do not correlate these diagnostic ratios with density or environmental measures as though they describe residents. Obtain compatible residence-coded events and populations for district comparisons. City-wide aggregation may reduce internal district mismatch, but does not by itself resolve referrals into the capital or inconsistent source coverage.
4. Air Quality Monitoring Network
The following is an illustrative inventory of 18 approximate station locations, not a verified list of active monitoring sites. The original compilation referred to public listings, including the World Air Quality Index project, but did not retain a per-coordinate URL or verification date. Those fields remain unknown; verify official coordinates, operating periods and pollutant coverage before exposure assignment or network planning.
ub_aq_stations <- tribble(
~Station, ~Latitude, ~Longitude,
"Misheel-Expo center", 47.89428, 106.8825,
"West crossroad", 47.91576, 106.8940,
"1st micro district", 47.91798, 106.8481,
"13th micro district", 47.91761, 106.9374,
"32nd Toirog", 47.939333, 106.914333,
"Ofitseruudiin ordon", 47.916611, 106.972556,
"Kharkhorin market", 47.910583, 106.837111,
"Urgakh naran micro district", 47.873778, 107.115028,
"Dambdarjaa", 47.988250, 106.951167,
"Khailaast", 47.963528, 106.893889,
"Tolgoit", 47.920278, 106.799250,
"Zuragt", 47.92973, 106.8886,
"Amgalan", 47.906250, 107.016556,
"Nisekh", 47.86394, 106.7791,
"Tavan buudal", 47.954500, 106.914833,
"Bayankhoshuu", 47.945528, 106.822917,
"Sharkhad", 47.93375, 107.0103,
"100 ail", 47.93291, 106.9214
)
# Keep missing provenance explicit rather than inventing sources or dates.
ub_aq_stations <- ub_aq_stations |>
mutate(
coordinate_source = "Historical illustrative compilation; unverified",
verified_at = NA_character_, operating_period = NA_character_,
pollutant_coverage = NA_character_
)
# Only unique polygon assignments count toward a district's inventory.
station_points <- st_as_sf(ub_aq_stations, coords = c("Longitude", "Latitude"), crs = 4326)
polygon_matches <- st_intersects(station_points, ub_districts)
station_points$assignment <- if_else(lengths(polygon_matches) == 1L, "Unique", "Unresolved")
# Assign each uniquely located station to a district
ub_stations_sf <- station_points |>
filter(assignment == "Unique") |>
st_join(select(ub_districts, District = name))
analysis_keys(st_drop_geometry(ub_stations_sf), "Station")
unresolved_stations <- station_points |>
filter(assignment != "Unique") |>
st_drop_geometry() |>
select(Station, assignment)
# Station count for every district, including those with none
station_count <- ub_districts |>
st_drop_geometry() |>
select(District = name) |>
left_join(count(st_drop_geometry(ub_stations_sf), District, name = "Stations"), by = join_by(District)) |>
mutate(Stations = coalesce(Stations, 0L)) |>
arrange(desc(Stations))All listed station locations have a unique district assignment.
| District | Listed locations |
|---|---|
| Bayanzurkh | 5 |
| Bayangol | 3 |
| Songinokhairkhan | 3 |
| Sukhbaatar | 3 |
| Khan-Uul | 2 |
| Chingeltei | 2 |
| Baganuur | 0 |
| Bagakhangai | 0 |
| Nalaikh | 0 |
Stations are assigned by their coordinates, using mongolmaps’ simplified district boundaries. A station within a few hundred metres of a district border may fall on the other side.
Spatial Distribution
# Zoom to the stations, with a 3 km margin, in UTM zone 48N (metres)
utm48n <- 32648
station_box <- ub_stations_sf |>
st_transform(utm48n) |>
st_buffer(3000) |>
st_bbox()
mn_map(filter(ub_districts, name %in% central),
label = FALSE, crs = utm48n,
title = "Air Quality Monitoring Stations in Central Ulaanbaatar"
) +
geom_sf(data = ub_stations_sf, color = "#C2623D", size = 2.5, alpha = 0.9) +
ggrepel::geom_text_repel(
data = ub_stations_sf, aes(label = Station, geometry = geometry), stat = "sf_coordinates",
size = 2.4, color = "grey20", min.segment.length = 0, seed = 1
) +
coord_sf(
crs = utm48n, datum = NA,
xlim = station_box[c("xmin", "xmax")], ylim = station_box[c("ymin", "ymax")]
) +
labs(subtitle = "Red points = monitoring stations (approximate locations)") +
analysis_map_theme()Coverage
station_coverage <- station_count |>
left_join(select(pop_data, District, Population = value), by = join_by(District)) |>
mutate(Pop_per_Station = if_else(Stations > 0, Population / Stations, NA_real_))
station_coverage |>
mutate(Pop_per_Station = if_else(Stations > 0,
scales::comma(Pop_per_Station, accuracy = 1), "Not applicable"
)) |>
knitr::kable(
col.names = c("District", "Listed locations", "Population", "Residents per listed location"),
digits = 0, format.args = list(big.mark = ","),
caption = "All districts, including those without a listed location"
)| District | Listed locations | Population | Residents per listed location |
|---|---|---|---|
| Bayanzurkh | 5 | 452,084 | 90,417 |
| Bayangol | 3 | 257,391 | 85,797 |
| Songinokhairkhan | 3 | 341,746 | 113,915 |
| Sukhbaatar | 3 | 142,356 | 47,452 |
| Khan-Uul | 2 | 278,532 | 139,266 |
| Chingeltei | 2 | 131,278 | 65,639 |
| Baganuur | 0 | 29,076 | Not applicable |
| Bagakhangai | 0 | 4,401 | Not applicable |
| Nalaikh | 0 | 40,781 | Not applicable |
station_coverage |>
filter(Stations > 0) |>
ggplot(aes(x = Pop_per_Station / 1000, y = reorder(District, Pop_per_Station))) +
geom_col(fill = "#167D8D", width = 0.7) +
geom_text(aes(label = paste0(round(Pop_per_Station / 1000), "k (", Stations, ")")), hjust = -0.1, size = 3) +
scale_x_continuous(expand = expansion(mult = c(0, 0.25))) +
labs(
title = "Residents per Listed Station Location",
subtitle = "Thousands of residents per station (number of stations in brackets)",
x = "Residents per station (thousands)", y = NULL,
caption = paste("Population year", pop_year, "| Districts with no listed location remain in the table")
) +
theme(panel.grid.major.y = element_blank(), panel.grid.minor = element_blank())3 districts have no uniquely assigned location in this inventory. This does not establish absence of monitoring. Residents per listed location does not measure representativeness; operating status, pollutant coverage, terrain and distance from sources matter. These ratios cannot determine where another monitor would add the most information.
Conclusion
The tables show differences in recorded population and mean school enrolment. Fine-scale density remains illustrative until boundary vintages are checked. Health comparisons require compatible residence-coded events, and the station inventory needs official verification before assessing exposure or monitoring adequacy.