Geospatial Qn2_Rmarkdown

2) Ethiopian population density and ERSS sample villages

This section replicates and reconstructs Figure 4 from the paper “Rural electrification, migration and structural transformation: Evidence from Ethiopia”, titled Ethiopian population density and ERSS sample villages.

Due to data access limitations, we use a combination of open geospatial datasets and survey-derived geographic variables rather than the restricted datasets used by the original authors.

Data Sources

1. Population Density Raster (2012)

• Source: WorldPop / LandScan-style gridded population data (2012, 1km resolution)
• File: eth_pd_2012_1km.tif
• Purpose: Represent spatial population concentration across Ethiopia

2. Administrative Boundaries (2026)

• Source: Natural Earth (country boundary) and GADM (regional and zonal boundaries)
• Purpose: Provide geographic context and administrative layering

3. Electricity Transmission Network (2017)

• Source: World Bank – Ethiopia Electricity Transmission Network shapefile
• Purpose: Represent high-voltage grid lines
• Note: This replaces OpenStreetMap grid data, which follows roads too closely.

4. Road Infrastructure (2018)

• Source: OpenStreetMap (via Geofabrik shapefiles)
• Purpose: Visualize major transport corridors
• Filtered to: motorway, trunk, primary roads

5. Household Geographic Variables (2012)

• Source: ERSS public geographic covariates
• File: pub_eth_householdgeovariables_y1.csv
• Purpose: Plot sample household locations used in the survey

##1. Load and smooth population raster
pop <- rast("~/Desktop/eth_pd_2012_1km.tif")

# Aggregate to smoother resolution (~5km)
pop_smooth <- aggregate(pop, fact = 5, fun = mean, na.rm = TRUE)

##2. Load Ethipia Border and neighbouring countries
ethiopia_border <- ne_countries(country = "Ethiopia", scale = "medium", returnclass = "sf")
ethiopia_border <- st_transform(ethiopia_border, crs(pop_smooth))
ethiopia_vect <- vect(ethiopia_border)

neighbors <- ne_countries(scale = "medium", returnclass = "sf") %>%
  st_transform(st_crs(ethiopia_border)) %>%
  st_make_valid()

##3. Clip raster to Ethiopia and convert to DF

pop_crop   <- crop(pop_smooth, ethiopia_vect)
pop_masked <- mask(pop_crop, ethiopia_vect)
pop_df <- as.data.frame(pop_masked, xy = TRUE, na.rm = TRUE)
colnames(pop_df)[3] <- "pop"

##4. Classify population density (quantiles)

breaks <- quantile(pop_df$pop, probs = seq(0, 1, length.out = 7), na.rm = TRUE)
pop_df$pop_class <- cut(pop_df$pop, breaks = breaks, include.lowest = TRUE)

##5. Local boundaries

eth_l1_sf <- st_read("~/Desktop/gadm41_ETH_shp/gadm41_ETH_1.shp", quiet = TRUE)
eth_l2_sf <- st_read("~/Desktop/gadm41_ETH_shp/gadm41_ETH_2.shp", quiet = TRUE)

eth_l1_sf <- st_transform(eth_l1_sf, st_crs(ethiopia_border))
eth_l2_sf <- st_transform(eth_l2_sf, st_crs(ethiopia_border))

##6. Load WB power transmission network
wb_power <- st_read("~/Desktop/ethiopia-electricity-transmission-network/Ethiopia Electricity Transmission Network.shp", quiet = TRUE)
wb_power <- st_transform(wb_power, st_crs(ethiopia_border))
wb_power <- st_intersection(wb_power, ethiopia_border)

##7. Load and filter major roads
roads <- st_read("~/Desktop/ethiopia-180101-free.shp/gis_osm_roads_free_1.shp", quiet = TRUE)

roads_major <- roads %>%
  filter(fclass %in% c("motorway", "trunk", "primary")) %>%
  st_transform(st_crs(ethiopia_border)) %>%
  st_intersection(ethiopia_border)

##8. Load household variables for village 

hh_geo <- read_csv("~/Desktop/pub_eth_householdgeovariables_y1.csv")

hh_sf <- st_as_sf(hh_geo, coords = c("LON_DD_MOD", "LAT_DD_MOD"), crs = 4326) %>%
  st_transform(st_crs(ethiopia_border))

##9. Admin for map

bb <- st_bbox(ethiopia_border)

xpad <- (bb$xmax - bb$xmin) * 0.15
ypad <- (bb$ymax - bb$ymin) * 0.15

paper_blues <- c(
  "#f2f6fa",
  "#dbe7f2",
  "#b7d2e8",
  "#8fb9da",
  "#5f9ec7",
  "#2f6da4"
)

Transformations Done and Final Output

Several transformations were applied to ensure clarity and visual comparability:

• Raster Aggregation:The original 1km raster was aggregated to ~5km resolution to reduce noise and match the visual smoothness of the paper.
• Quantile Classification:Population density was classified using quantiles to highlight relative spatial concentration rather than absolute values.
• CRS Harmonization:All layers were transformed into a common coordinate reference system to ensure spatial alignment.
• Boundary Masking: Population raster was clipped strictly to Ethiopia’s borders to avoid misleading density outside the study area.
• Road Filtering: Only major roads were retained to avoid clutter and to match the visual emphasis of the paper.

These are the differences from the paper:

• Electrification Status Not Publicly Available: The paper classifies villages as electrified vs non-electrified. In this replication, household points are plotted without classification, represented by a yellow dot.
• Transmission Network Source: The paper uses utility-grade grid data, while we use World Bank open grid data, which approximates the structure.
• Population Density Representation: The paper aggregates density at the district level, while ee use a smoothed raster-based representation for higher spatial resolution.
• Temporal Differences: The paper studies changes between 2012 and 2014, while our map reflects a static 2012 snapshot overlayed with data from 2017, 2018 and 2026. Some historical data was unavailable.

Original Figure

knitr::include_graphics("~/Desktop/Qn2 actual chart.png")

End