---
license: cc-by-4.0
language:
  - en
  - es
pretty_name: "ds4bolivia — Bolivian municipalities: SDGs, satellite embeddings & nighttime lights"
tags:
  - bolivia
  - sdg
  - sustainable-development
  - satellite-embeddings
  - nighttime-lights
  - remote-sensing
  - geospatial
size_categories:
  - n<1K
---

# Bolivian Municipalities — SDGs, Satellite Embeddings & Nighttime Lights

Socioeconomic, satellite, and geographic measures for Bolivia's **339 municipalities**
across its **9 departments** — the analysis data for the project *"Predicting the
Sustainable Development of Bolivian Municipalities with Satellite Embeddings and Machine
Learning"* (Carlos Mendez, Nagoya University).

This dataset is a public **mirror** of the `data/` folder in
[`cmg777/project2026e`](https://github.com/cmg777/project2026e). Every file is keyed on
`asdf_id` (integer, `0`–`338`), the universal join key across all datasets. To attach
municipality/department labels, merge any file to `regionNames/regionNames.csv` on `asdf_id`.

## Contents

| Path | Description |
| ---- | ----------- |
| `ds4bolivia_v20250523.csv` | Analysis-ready **merged master** (339 × 351) — all subfolders joined on `asdf_id`. |
| `definitions_ds4bolivia_v20250523.csv` | **Data dictionary** for the master file (`varname` → `varlabel`). |
| `regionNames/` | Administrative IDs & names — the join foundation. |
| `sdg/` | IMDS + 14 composite SDG indices (0–100 scale). |
| `sdgVariables/` | 64 individual SDG indicators across all 17 goals. |
| `satelliteEmbeddings/` | 64-dim Google Satellite Embeddings — simple-mean (2017) and population-weighted (**2017–2025** panel). |
| `nighttimeLights/` | VIIRS nighttime lights (simple-average & population-weighted, **2017–2021**), plus `rasters/` GeoTIFF. |
| `predictions/` | Out-of-sample predictions for SDG 1, 7 and 13 (four-view: actual, lights, embeddings, combined) plus space-time forward projections. |
| `maps/` | Municipality boundary polygons (GeoJSON), keyed by `asdf_id`. |
| `sdg/`, `regionNames/`, … | Each subfolder ships its own `README.md` with a full variable dictionary. |

## Provenance & coverage

- **Spatial unit:** 339 Bolivian municipalities (9 departments); **version** `v20250523`.
- **Source:** [`quarcs-lab/ds4bolivia`](https://github.com/quarcs-lab/ds4bolivia); the 64-dim
  satellite embeddings are Google Satellite Embeddings (Google Earth Engine); SDG indices and
  the IMDS index follow Andersen et al. (2020).
- **Time coverage:** population 2001–2020; nighttime lights 2012–2021; most SDG variables 2012–2019;
  satellite embeddings 2017 (simple-mean) and 2017–2025 (population-weighted panel).

## Load from Python

Load any file straight from the Hub (no full clone needed):

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="cmg777/project2026e",
    repo_type="dataset",
    filename="satelliteEmbeddings/bolivia_pop_weighted_2017.csv",
)
df = pd.read_csv(path)
```

The companion repo ships a `code/hf_data.py` helper with a `data_path()` function that
prefers a local copy and otherwise streams the file from this dataset.