geo4cambodia

Subnational development data for Cambodia, ready to analyse, at every spatial scale.

geo4cambodia is the Python package of the GeoDevelopment Observatory of Cambodia. It loads the observatory's indicators (district GDP per capita, province population, night-time lights, PM2.5, Global Data Lab HDI...) as tidy panels, with boundaries for every scale. It also lets you build your own indicators: link your tables to the same unit IDs, summarise rasters over units, move data across scales and combine datasets. It works out of the box in Google Colab: every example below runs in a new Colab notebook (replace the placeholder files, such as your own table or raster, with yours).

!pip install -q geo4cambodia mapclassify           # mapclassify: for scheme="quantiles"
import geo4cambodia as geo

geo.datasets()                                       # what is available, at which scale and years
gdf = geo.load("gdppc", years=2022, geometry=True)   # 202 districts, with boundaries
gdf.plot(column="y", scheme="quantiles", k=5, legend=True)

What it does

One ID for every unit, at every scale

Every unit has a geo ID that nests by prefix (GADM 4.1 hierarchy): district KH-16-01 lies in province KH-16 (Phnom Penh), which lies in Cambodia KH. Global Data Lab regions (KH-GDL-02) are unions of whole provinces. Each dataset is a long panel keyed by geo_id and year.

Build your own indicator in five lines

!pip install -q geo4cambodia
import pandas as pd
import geo4cambodia as geo

mine = pd.read_csv("schools_by_province.csv")             # e.g. columns Province, year, schools
mine = geo.link(mine, scale="province", on="Province")     # "BântéayMéanchey" -> KH-01
panel = geo.combine(["env"], scale="province", extra=[mine])
panel["schools_per_10k"] = panel["schools"] / panel["total_GlobPOP_Count"] * 1e4
geo.save(panel, "schools", scale="province", directory="my_data")

From a raster:

!pip install -q "geo4cambodia[raster]"            # [raster] adds rasterio
import geo4cambodia as geo

lights = geo.zonal("viirs_2020.tif", scale="district", stats=["sum", "mean"], name="ntl", year=2020)

Then explore and model the panel with expdpy:

!pip install -q geo4cambodia expdpy
import geo4cambodia as geo
import expdpy as ex

df, df_def = geo.for_expdpy("hdi")
df = ex.set_labels(df, df_def, set_panel=True)
ex.analyze_panel_table(df, "HDI", ["lnNTLpc"])            # pooled, between, fixed and random effects

Or analyse it spatially with geometrics:

!pip install -q geo4cambodia geometrics
import geo4cambodia as geo
import geometrics as gm

gdf, df, df_dict = geo.for_geometrics("gdppc")
df = gm.set_labels(df, df_dict, set_panel=True)
gm.analyze_beta_convergence(df, "y", gdf=gdf, w=gm.make_weights(gdf)).interpret()   # levels: logged internally

Without the package

Every file is also a plain URL, readable with pandas alone (preinstalled in Colab, so nothing to install):

import pandas as pd
df = pd.read_csv("https://raw.githubusercontent.com/quarcs-lab/geo4cambodia/main/data/district/gdppc.csv")

Installation

In a Google Colab cell:

!pip install -q geo4cambodia                    # core: pandas, numpy, geopandas, pyarrow, requests, pyyaml (Colab has them)
!pip install -q "geo4cambodia[raster]"          # + rasterio, for zonal()
!pip install -q git+https://github.com/quarcs-lab/geo4cambodia   # development version

Python 3.11 or later. The data are not inside the package. It reads them from the GitHub repository, so installation is quick and you always get the files the website links to. For reproducible research, pin a release of the data with geo.load(..., ref="v0.3.0") or the environment variable GEO4CAMBODIA_REF=v0.3.0.

Documentation

• How the data infrastructure works: scales, IDs, metadata, aggregation rules, and recipes for building indicators
• Dataset pages: variable dictionaries, coverage, sources and citations
• Website: indicators, notebooks, web apps and insights

Citation

Mendez, C., Khoun, T., & Poortinga, A. (2026). GeoDevelopment Observatory of Cambodia. https://quarcs-lab.github.io/geo4cambodia/

MIT License. Original data sources are credited on each dataset page.