An AI-Driven Satellite Imagery Approach to Map Electricity Access Disparities in Mozambique 01/02/2018

Francesco Paolo Antinori developed an artificial intelligence and high-resolution satellite imagery approach to map and analyze electricity access disparities in Mozambique over the period 2015–2024. His work was developed at IN+ under his MSc thesis at Poli Milano, Italy.

The study integrates nighttime satellite imagery with socioeconomic, infrastructural, and conflict variables in a dataset of approximately 38.5 million observations at 1 km² spatial resolution, covering the entire Mozambican territory over 2015–2024. The study trains a Random Forest algorithm on 2015–2021 observations and evaluates its predictive capacity on the 2022–2024 period, which remains entirely unseen during training. Model interpretability is ensured through SHAP analysis, an interpretable AI tool quantifying each variable's contribution to predictions, applied at both national and provincial level. Results are translated into an operational planning tool through a gap analysis identifying areas with unrealised structural electrification potential.

The model achieves an R² of 0.785 on the out-of-time test set. SHAP analysis reveals that electricity access is primarily a socioeconomic phenomenon: individual wealth and population density jointly explain approximately 74% of total feature importance. Drivers vary geographically in a systematic manner: Maputo, Cabo Delgado, and Niassa present anomalous profiles interpretable respectively as effects of urban maturity, armed conflict, and extreme geographic marginalisation. The gap analysis identifies 26,874 priority pixels organised into three intervention clusters: grid extension (11–21 km), connection subsidies for Maputo, and off-grid solutions for Tete and Niassa.

A uniform national electrification strategy would be methodologically inefficient. Province-differentiated approaches are necessary to achieve universal access targets by 2030. The framework is annually updatable and represents a decision support tool directly usable by EDM and international organisations operating in Mozambique's energy sector.

Click here to read the full work >>>