Project 25006 AI-Driven Mapping of Solar Panel Adoption Potential Across Africa
Project Title
AI-Driven Mapping of Solar Panel Adoption Potential Across Africa
Nan Jia
Program Research Fellow in Stanford University and Michigan State University
Organization Stanford University and Michigan State University
Email: cannonjia17@gmail.com
Project Description
This project will integrate Google Earth Engine (GEE) with supervised machine-learning models to generate Africa-wide, high-resolution maps of residential and small-scale solar-photovoltaic (PV) adoption potential. First, we will compile multi-source satellite and ancillary data in GEE, including (i) long-term solar irradiance and cloud-cover composites, (ii) land-surface temperature, (iii) building-footprint masks, and (iv) distance to transmission lines and roads. Second, we will enrich these geospatial layers with socio-economic covariates—population density, household income proxies, electrification status, and policy incentives—sourced from open repositories (e.g., WorldPop, DHS, IEA).
Using a stratified sample of verified PV installations as training data, we will fine-tune gradient-boosted decision trees and convolutional neural networks to predict two key outputs for every 1 × 1 km cell: (1) technically feasible rooftop/ground-mount PV capacity (kW) and (2) likelihood of near-term adoption (probability score). Model performance will be validated through spatial cross-validation and sensitivity analysis. Outputs will be aggregated to administrative levels (province, district) and shared via an interactive GEE web app and downloadable GIS layers.
The resulting atlas will help governments, NGOs, and private investors identify high-impact intervention zones, optimize incentive schemes, and accelerate progress toward SDG 7 (affordable and clean energy)."
Tasks and Responsibilities "The intern will begin by gathering high-quality satellite and meteorological layers directly available in Google Earth Engine (GEE)—for example, multisource solar irradiance composites, cloud-cover frequency, land-surface temperature, and recent building-footprint masks. In parallel, they will obtain socio-economic covariates (population density, electrification status, household-income proxies, policy incentive indices) from external open repositories such as WorldPop, the Demographic and Health Surveys (DHS), and the International Energy Agency (IEA). These external raster or vector layers will be cleaned, re-projected, and uploaded to the team’s GEE asset space (or linked via cloud storage) so that all inputs share a consistent spatial grid and metadata standard.
Next, the intern will assemble a training dataset by collating verified coordinates of existing solar-PV installations from public databases and partner field surveys. They will produce a stratified sample that represents diverse climatic and socio-economic settings across Africa, save it as labelled FeatureCollections (GeoJSON/CSV), and document the sampling protocol in a brief technical memo.
With data ingested, the intern will support model prototyping in the GEE Python API. They will implement baseline algorithms—Random Forests, gradient-boosted trees, and preliminary convolutional neural networks—run hyper-parameter sweeps, and log performance metrics in fully reproducible Jupyter notebooks.
To assess robustness, the intern will design a spatial cross-validation scheme, map residuals, identify systematic biases, and recommend feature refinements. Findings will be summarised in a concise validation report that includes confusion matrices, MAE/R² tables, and bias heat maps.
Minimum Qualifications
Enrolled in, or recently graduated from, a bachelor’s or master’s program in geography, computer science, environmental engineering, or a related field.
Proven GIS and remote-sensing experience—ideally with Google Earth Engine or comparable platforms.
Solid Python skills, including basic use of scikit-learn or other machine-learning libraries.
Ability to clean, manage, and document large geospatial datasets.
Familiarity with Git-based version control and reproducible-research workflows.
Clear written and spoken communication in English.
Terms of the Project
one year
Key Words
artificial intelligence; machine learning; geospatial analysis; satellite imagery; solar photovoltaics; energy access; renewable energy; Africa; adoption potential
Research Topics
Solar photovoltaic adoption modeling; machine learning
Deliverables
Reproducible ML toolkit (GEE + Python) with validation brief and Africa-wide prediction rasters.
Curated geospatial + socio-economic dataset with full metadata and provenance note.
Open training set of verified solar-PV locations (GeoJSON/CSV) with sampling memo.
Reproducible ML codebase (GEE + Python) with validation brief.
Continental prediction rasters (capacity and adoption likelihood) in cloud-optimised GeoTIFFs, plus national summaries.
Interactive web map and concise final report (policy-focused, figure-ready).