evidoria

← Back to browse

Good practice Imported

DANE's Satellite-and-Machine-Learning Poverty Map — Colombia's 70-Fold Leap in Statistical Granularity

Colombia · Bogotá · See the Colombia profile · See the Bogotá profile

Evidence: Observational / pre–post Top 66% 53/100 · Ask Evidence Copilot about this practice

Colombia's DANE, World Data Lab and PARIS21 combined satellite day/night imagery with ML to raise poverty-estimate granularity from 1,123 to 78,000 data points — a 70-fold increase — cross-checked against census data, though policy impact hasn't been independently evaluated.

70 x (fold)
Increase in poverty-estimate data-point granularity (2020–2021)
1123 data points
Poverty-estimate data points before project
78000 data points
Poverty-estimate data points after project
1–999 people per grid cell
Population-density range in Bogotá (100x100m grid)
DANE's Satellite-and-Machine-Learning Poverty Map — Colombia's 70-Fold Leap in Statistical Granularity

Details

Maturity
Pilot
Promoter
DANE (Departamento Administrativo Nacional de Estadística), with World Data Lab and PARIS21 (OECD)
Period
2020–2021
Keywords
national statistics, poverty measurement, geospatial analysis, social policy

Context

Colombia's national statistics office DANE partnered with World Data Lab and the OECD-hosted PARIS21 initiative to enrich national poverty and population statistics using satellite imagery and machine learning.

Objectives

The project aimed to sharply increase the geographic resolution of poverty and population estimates beyond what traditional census-based methods allow.

Activities

Researchers trained models linking daytime and night-time satellite images, then used daytime imagery alone to predict municipal poverty rates at 4x4 km resolution and population density at 100x100 m resolution, validating results against official census data.

Results

The approach increased the granularity of Colombia's poverty estimates 70-fold, from 1,123 to 78,000 data points, and revealed population-density variation invisible at municipal level, such as parts of Bogotá ranging from 1 to 999 people per 100x100 m grid cell.

Conclusions

The methodological leap in resolution is documented and cross-validated against census figures, but no independent evaluation yet shows whether the finer-grained data has changed specific policy decisions, subsidy targeting, or resource allocation.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
DANE director-general (Juan Daniel Oviedo), World Data Lab (Katharina Fenz, Simon Riedl), PARIS21 deputy head (François Fonteneau)

Conditions for success

  • Partnership between a national statistics office and external data-science organisations
  • Access to satellite day/night imagery
  • Official census data available for cross-validation

Common failure modes

  • No independent evaluation yet exists of whether the finer-grained data has changed policy decisions, subsidy targeting or resource allocation

Where it fits

Governance type
national statistics office / international partnership
Scale
national

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful