Scalable spatial design of electricity access systems

This thesis explores spatially specific data and methods to design community-tailored electricity access systems at scale. It is motivated by the need to close the electricity access gap in rural low- and middle-income country contexts quickly and cheaply in line with the Sustainable Development Goals. The majority of the 760 million people currently lacking electricity access live in rural areas of sub-Saharan Africa and South Asia. Electrifying these areas is challenging given their cultural diversity, remote nature, and sensitivity to affordability. Context-specific electrical designs are required to achieve uptake in these communities; however, such specificity can come at high cost. This thesis therefore tackles the challenge of the local specificity and global scale of the electricity access problem through practical spatial design methods. Two key data gaps are identified which must be filled in order to design least-cost energy access systems: locations of potential connection points, and their anticipated demands. Home-level location data which are publicly available for potential connection points in off-grid areas are either aggregated at inadquate resolution for topology design or contain significant gaps, particularly in off-grid areas. To address this, citizen science and computer vision are applied to accelerate accurate home detection in satellite imagery. Through a large-scale online citizen science project, approximately 1,267 km2 of rural Kenya, Sierra Leone and Uganda was mapped at an average rate of 7 km2/day and an estimated cost of $20.84/km2. Home annotations produced through this work achieve a recall of 93% and precision of 49%, which can be increased to 69% through clustering. The clustered annotations were used to train a Faster R-CNN object detection algorithm, which achieves a precision of 67% and recall of 36%; this can be increased to 57% by training on raw annotations instead of clusters. The trained detector was found to map at a rate of 42,938 km2/day, proving the rapid mapping of rural unelectrified areas to be feasible at a global scale and low cost. High-resolution residential demand data are similarly scarce in off-grid communities. Costly local surveys to understand electrical aspirations tend to produce inaccurate results, given the unfamiliarity of respondents with electricity. To overcome this, a bottom-up demand estimation approach rooted in existing empirical data is developed to achieve spatially-specific and realistic demand estimates for off-grid areas. A case study application of this approach in Sierra Leone was undertaken using Multiple Indicator Cluster Survey data. The results of this work validate the underlying premise of spatial variance of demand amplitude. The load profiles generated using this approach are found to approximate a Tier 3 load, despite a lack of Tier 3 appliances, leading to a critique of the definition of the Multi-Tier Framework for Measuring Energy Access. These location and demand data are finally applied to home-level spatial grid design. Home locations are clustered into electricity communities, grid topologies are estimated using graph theory, generation types are selected through spatial analysis, and generation and storage are optimally sized for least cost. An approach is also developed to map design pathways through modular community grid expansions which allow for demand growth and autonomy. This framework was applied in a case study region in the Northern Province of Sierra Leone. In this region, 12 local micro-grids are identified as the best electrification solution in the near-term, with 11 outlying homes receiving solar home systems. Infrastructure sizing is presented for one micro-grid, where an initial design point of 50 kW of PV and 108 kWh of battery storage is found to meet anticipated low-end and mid-term demands without energy poverty risk. Three modular expansions of 30 kW PV and 65 kWh of storage are specified for installation in this micro-grid as demands evolve. The work in this thesis focuses on sub-Saharan Africa, with particular attention paid to Kenya, Uganda, and Sierra Leone. Case studies in the thesis primarily focus on Sierra Leone; however, the methods are purposefully intended to be practical and generalizable across low- and middle-income countries requiring electricity access system design.