A geospatial machine-learning framework for regional electrification estimation mapping and trend analysis using...

Area of Science:

• Geospatial analysis
• Machine learning applications
• Energy access studies

Background:

• Significant electricity access disparities exist between urban and rural areas in Ethiopia's Amhara region.
• Traditional surveys for electrification are expensive and inconsistent.
• Existing Nighttime Light (NTL)-population threshold methods have limitations including arbitrary thresholds and failure to detect off-grid areas.

Purpose of the Study:

• To develop and validate a geospatial machine learning framework for mapping electrification status in rural Ethiopia.
• To estimate regional electrification rates from 2018 to 2024.
• To compare the performance of the developed framework against traditional NTL-population threshold methods.

Main Methods:

• Utilized Visible Infrared Imaging Radiometer Suite-Day/Night Band (VIIRS-DNB) nighttime lights, population density, and proximity to grid lines and roads.
• Employed machine learning algorithms including Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT).
• Validated model outputs against GPS-recorded transformer locations.

Main Results:

• The Random Forest model achieved 86.8% overall accuracy and a 71.3% Kappa coefficient.
• The RF model accurately reconstructed electrified networks, outperforming NTL-population threshold methods.
• Population density was the leading predictor (27.49% importance) for electrification status.

Conclusions:

• The developed geospatial machine learning framework provides an accurate and reliable method for mapping rural electrification.
• The study highlights persistent zone-level electrification disparities and estimates a steady growth to 41% by 2024.
• This framework is valuable for supporting Sustainable Development Goal 7 (SDG 7) planning and investment in underserved regions.