October 26-27 | Open Source in Energy Access Symposium Hackathon | Kigali, Rwanda
Build a market-reality layer for the Energy Access Explorer (EAE) that distinguishes where the off-grid solar market has failed to arrive from where it has quietly succeeded.
Grant capital and results-based financing for off-grid solar tend to flow toward territories that are already commercially viable, because those are the territories that are visible in the data. Places that remain dark across every available dataset also remain dark in most funding decisions.
The sector does not lack maps of where energy demand should exist; the Energy Access Explorer (EAE) already provides these, running multi-criteria analysis on the fly across normalized demand and supply layers so that planners can reweight criteria interactively and rank priority areas from their own perspective. What EAE currently doesn't include is a market-reality layer: a view of where the off-grid solar market has failed to arrive, and a means of distinguishing that from where it has quietly succeeded. This challenge asks teams to build that layer for a country of their choice, integrating sales data.
VIIRS and Black Marble have a radiance detection floor well above the emissions of a solar home system or pico-lantern, which means a village where every household owns a solar home system registers as dark, and a village with no energy access at all also registers as dark. The intuitive formulation of this challenge — in which dense population combined with zero nighttime lights is taken to indicate an underserved community — will therefore surface precisely those territories where the off-grid sector has already done its best work. Darkness indicates the absence of a grid connection, which defines the addressable market for distributed renewable energy, but it says nothing about whether that market has been served.
Actual penetration must be measured from sources that measure penetration, which is why georeferenced household survey data (DHS and MICS geo-clusters carrying solar panel ownership and electricity source, and ESMAP Multi-Tier Framework surveys carrying explicit Tier 0 to 5 access) forms the backbone of this challenge. GOGLA sales data is semi-annual, affiliate-reported, and aggregated to country and product category, with company-level detail held confidential under a three-data-point rule and no public access to the underlying platform, so it should be treated as a national control total against which a survey-derived penetration surface is calibrated — rather than as a pixel-level penetration layer in its own right.
Three core deliverables (plus an optional prototype):
- Predictive model of off-grid solar penetration: an open-source model trained on georeferenced survey labels, with blocked cross-validation performance reported alongside random cross-validation performance, and feature analysis framed as hypotheses rather than findings.
- EAE-conformant layer: at least one raster, with values normalized 0 to 1 and direction documented, accompanied by metadata covering source, year, method, and known limitations.
- Documentation and limitations statement: a brief write-up describing the datasets used, the modelling approach, the validation methodology, and accuracy metrics, together with a one-page statement of what the model cannot see, where it will be wrong, and who it would misallocate capital toward.
- Interactive prototype (optional): a map in which priority zones are visible, inspectable, and traceable back to the layers that produced them.
- Python geospatial: GeoPandas for vector processing; Rasterio / GDAL for raster analysis.
- Machine learning: Scikit-learn, XGBoost, LightGBM for modelling off-grid solar penetration.
- Geospatial and socio-economic data: NASA Black Marble; WorldPop or Meta High Resolution Settlement Layer for population; Meta Relative Wealth Index via the Humanitarian Data Exchange; Gridfinder or OpenStreetMap for grid infrastructure.
This is the suggested group that allows a team to distinguish a village the market has already served from a village it has never reached. Teams are encouraged to identify more data during the hackathon. Each source is listed with the resolution it operates at, because that determines the job it can do:
- GOGLA Sales and Impact Data: unit sales by country, region, and product category, split between cash and PAYGo, reported semi-annually and annually by GOGLA affiliates. Published as public reports with accompanying tables. The underlying platform at data.gogla.org is restricted to companies that report into it, and a three-data-point rule enforces confidentiality, so distributor-level footprints are not obtainable. Resolution: national. Job: control total.
- Off-Grid Solar Market Trends Report (GOGLA and ESMAP): market sizing, investment gap, and outlook. Resolution: regional and national. Job: context and sanity check.
- GOGLA deals database: investment flows into off-grid solar companies from 2012 onward. Resolution: company. Job: identifying which distributors are capitalized to expand into a new territory.
- PAYGo PERFORM: standardized key performance indicators for PAYGo company performance, including portfolio quality and repayment. Resolution: company, aggregated. Job: identifying which distributors could absorb and deploy a grant.
- AMDA Benchmarking Africa's Minigrids: operational and financial benchmarks aggregated across member developers and countries. Resolution: operator, aggregated. Job: context on mini-grid viability.
- ENERGYDATA.INFO (ESMAP / World Bank): open data registry with API access, carrying existing and candidate mini-grid project datasets, grid network data, and the Africa Electricity Grids Explorer. Resolution: point and line geometry. Job: the only georeferenced market and infrastructure source in this group.
- Infrastructure data available in Energy Access Explorer.
- National results-based financing and electrification programme records: connection counts and subsidy disbursements published by rural electrification agencies and programme administrators. Resolution: sub-national where available, variable quality. Job: validation and ground-truthing.
- DHS and MICS geo-clusters, and ESMAP Multi-Tier Framework surveys: household solar panel ownership, electricity source, and Tier 0 to 5 access, carried at cluster level with displaced GPS. Resolution: point. Job: this is the only source in the challenge that measures penetration at a location, and it is therefore the label.
- Python
- SQL
- JavaScript / any other for frontend
- Experience with Geospatial Analysis and Remote Sensing using, for example, Python, Google Earth Engine, PostGIS, QGIS, GeoJSON, GeoParquet, or spatial SQL.
- Experience with Machine Learning and Predictive Modelling, including feature engineering, model evaluation, and validation of spatially autocorrelated data.
- Experience with Data Science and Analytics using Python, Pandas and GeoPandas, data cleaning and integration, and statistical analysis.
- Familiarity with household survey microdata, for example DHS, MICS, LSMS, or Multi-Tier Framework surveys.
- Knowledge of off-grid solar markets and distribution models, including solar home systems, pico-solar, and results-based financing.
- Dashboard and mapping development skills.
- Preston (full name to be added by EnAccess)
- Join the OSEAS Discord server: https://community.oseas.org/
- Introduce yourself in the
#introductionschannel and join the relevant channels for this challenge. - Read the documentation: