The emburdensynth pipeline turns open data into
cell-level household energy expenditure estimates in five phases plus a
post-processing step. Each stage is implemented as a single exported
function; each can be run independently for debugging.
impute_physical_characteristics()
Probabilistic mapping from NREL ResStock building physical attributes onto ACS PUMS demographic households. Produces per-household dwelling area, vintage, heating fuel, insulation profile.
calibrate_expenditures()
Quantile mapping via RECS empirical CDF stratified by income bracket and building type. Assigns utility-expenditure quantiles to each household.
optimize_spatial_allocation()
Multi-level simulated annealing across ~2,300 PUMAs / GADM level-2
units, matching cell-level marginals to national totals. Runs in
parallel via furrr.
reverse_engineer_consumption()
Piecewise inverse-tariff root-finding via uniroot. Converts
household dollar expenditure to physical kWh consumption using
per-country utility rate structures from URDB.
generate_load_profiles()
Load-profile morphing to 8,760 hourly bins, weighted by end-use and
season using NREL EULP baselines. Output:
l1–l8760 kWh/hour columns per household in
parquet.
calculate_energy_insecurity_gap()
Joins the physics baseline (degree-day derived required kWh) with the pipeline’s estimated consumption. Flags cells where estimated < 0.9 × physics-required.
Aggregating household-level burden to country-level requires a
well-formed weighted mean. The naive choice —
weighted.mean(spending/income, weights) — is biased. The
mathematically correct formulation, derived from the Net-Energy-Burden
framework in the emburden package, computes
Nh = (G-S)/Se per cell (where Se is energy
spending), takes the weighted mean of Nh, and inverts
as:
NEB = 1 / (1 + weighted_mean(Nh))Across the 128 countries covered in our release, naive aggregation over-estimates country burden by a median 5.72%. The bias is small for stable economies but becomes very large under hyperinflation. For VEN, where cell-level energy spending/income ratios span several orders of magnitude, the naive-vs-proper divergence reaches 1414%. This result establishes that any prior cross-country comparison of household energy burden using weighted-mean-of-ratios has been systematically biased in a direction that inflates estimates in the world’s most economically volatile settings.
Cell-level physics-required kWh is derived from:
with country-specific HDD and CDD from World Bank Climate Change Knowledge Portal and a global-average building envelope (60 m², COP 2.5 for cooling, 2.8 for heating). Per-country building-stock corrections are on the roadmap.
Bootstrap 90% confidence intervals via 500 Beta-perturbation
replicates on GCD sector shares (n_eff = 100) plus
boundary-jitter sensitivity across GADM level-2 boundaries. Cell-level
CI width is aggregated to country-mean CI width via variance
decomposition.
Country-level burden is joined to 11 external signals for the PCA + k-means archetype analysis:
See data provenance for full URLs + retrieval dates.