Disaggregated household energy supply measurement to support equitable municipal energy planning in rural Nepal

The data available to municipal energy planners

Since 2015, Nepal's local governments have taken on more responsibility for providing and managing public services. Municipal energy planners are now to be consulted and given budgets to improve energy access in their constituencies. Yet the newest energy access data describes the country and its provinces, and the last census dates from 2011. Municipal planners remain without actionable data on their own constituents.

In rural Nepal, nationally representative survey data shows that just 67% of households are connected to the national grid, while 27% rely on off-grid technologies such as minigrids and standalone solar systems. At the same time, 72% of households rely on biomass for cooking and heating. Aggregates of this kind establish the scale of a supply deficit without locating it. They cannot tell a municipal leader which settlements are unserved, or whether poor supply falls disproportionately on poorer constituents. The paper asks how survey sampling, multi-tier supply measurement and spatial modelling can be combined to produce baselines at the scale at which planning decisions are now made.

Gridded sampling with spatial and wealth disaggregation

A total of 2,357 households were surveyed across six rural municipalities in the hilly region of Sudurpashchim Pradesh, far-western Nepal, in June and July 2018. Enumeration clusters of ten households were drawn with probability proportionate to estimated size from gridded population data (the GridSample approach). Enumerators then selected households at random within each cluster and recorded responses on tablets using Kobo Toolbox.

Supply is measured along two dimensions adapted from the multi-tier framework (MTF), a widely used scale that scores supply on tiers from 0 to 5. Here, both dimensions are kept continuous rather than binned into tiers. Nominal capacity comes from fuse amperage or peak solar capacity, and duration from daily supply hours. A supply–access index combines them: mean daily supply hours of electrified households, multiplied by the access rate and divided by 24. Appliances are grouped into three ordinal service levels, and stoves into unimproved, improved and liquefied petroleum gas (LPG).

Disaggregation runs along two axes. Households are split into expenditure terciles within each municipality, as a proxy for wealth. Spatially, survey data is binned into hexagons of about 5 km². Boosted regression trees, a machine-learning method, then predict grid and minigrid connection likelihood in every hexagon, including those not surveyed. Predictors include travel time to cities, night-time lights, population, substation distance and a hydropower proximity factor.

Severe geographic and wealth-related supply inequality

The surveys reveal severe geographic and wealth-related inequalities in energy supply provision and associated burden. Grid connection rates among surveyed households were 17% in Chure and 31% in Dogadaker, and zero in the other four municipalities. Minigrid provision follows an almost inverse and equally uneven pattern: 59% of surveyed households in Badimalika, 10% in Kedarseu, 3% in Baddikedar and Bithadchir, and none in Chure or Dogadaker. In Dogadaker, Chure and Badimalika, supply technology rises with expenditure group, so inequities are more pronounced where average supply quality is higher. The direction of the causal relationship between supply quality and wealth is not identified.

Most households use no electrical services beyond lighting and communications, regardless of supply quality. Electricity cost alone is unlikely to explain this, as households in all expenditure groups typically spend less than 5% of annual expenditures on electricity. Median electricity expenditures stay constant or rise by under a third from the bottom to the top expenditure group. Cooking offers a bleaker picture: most households burn collected firewood in unimproved stoves regardless of wealth, and fewer than 10% use more than one stove.

The spatial models predict connection reasonably well on held-out test data, with an area under the curve (AUC, where 1 is perfect and 0.5 is chance) of 0.841 for grid and 0.817 for minigrid. The population clusters used for mapping capture 91–97% of each municipality's remotely sensed population estimate.

Disaggregation as the basis for equitable planning

The data gives municipal leaders accurate information on the supply challenges faced by all of their constituents. It is a transparent starting point for municipal energy access improvement plans, enables assessment of improvement over time, and can support calibration of top-down plans in discussion with the federal government. The spatial estimates show where connection intensification, through the national grid or standalone solar systems, is necessary. They also bring to light the plight of poorer 'undergrid' households that live near the grid without being connected to it. Disaggregated measurement is crucial for transparent and equitable energy supply improvement planning.