Rewiring
the WorldBeta · v2026.10.4
Countries & validation
Release 2026.10.4

Countries & validation

Info on model training, official statistics used for anchoring to regional levels, and model performance by country. A country with no local training data is scored by a model that never saw it (leave-one-country-out).

CountryYearsLocal training dataCaseLevel sourceLarge facilitiesHourly
Australia2017–25Partly trainedT2-A1+SAustralian Energy Statistics 2026 flat tables (Tables A–S), Department of Climate Change, Energy, the Environment and Water (DCCEEW) (CC BY 4.0)212Annual only
Canada2021–24Partly trainedT2-A1+SSupply and demand of primary and secondary energy in natural units, Statistics Canada (Statistics Canada Open Licence)258Annual only
France (metropolitan)2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))206Annual only
French Guiana2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))—Annual only
Great Britain2021–24Trained on local dataT3-A1Subnational electricity consumption statistics (LAD), UK Department for Energy Security and Net Zero (DESNZ) (Open Government Licence v3.0)Subnational total final energy consumption, UK Department for Energy Security and Net Zero (DESNZ) (Open Government Licence v3.0)DUKES 5.1 — Electricity commodity balance (UK, annual), UK Department for Energy Security and Net Zero (DESNZ) (Open Government Licence v3.0)116Experimental · 2024
Guadeloupe2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))—Annual only
Martinique2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))1Annual only
Mayotte2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))—Annual only
Réunion2021–24Trained on local dataT3-A1Consommation annuelle d'électricité et de gaz par IRIS, Agence ORE (Opérateurs de Réseaux d'Énergie) (Licence Ouverte (Etalab V2.0))—Annual only
South Korea2021–24No local dataT0-A1+SKEPCO electricity sales by sigungu and contract type (monthly workbooks), Korea Electric Power Corporation (한국전력공사) (KOGL Type 1 (공공누리 제1유형, attribution) — "이용허락범위 제한 없음" per data.go.kr 15069677)46Annual only
Trained on local data

The model learned from consumption published for small areas here, and every region’s total is matched to official statistics. Sometimes only residential data is available, sometimes residential and non-residential, sometimes only the two combined.

Partly trained

Local small-area consumption data covers part of this area, and the model learned from it there; elsewhere the same model, trained in other places, draws the pattern. Every region’s total is matched to official statistics.

No local data

No local small-area consumption data: a model trained in other countries draws the pattern inside each region, and official statistics set every region’s total. Its accuracy is checked by holding whole countries out of training (leave-one-country-out).

Validation by country or territory

Our model is compared against a population-only counterfactual, meaning consumption is allocated proportional to population. Both methods receive correction to the same known totals of compared areas to make the comparison fairer. We also use LOCO (leave-one-country-out) validation to see how well our model would perform on countries it has not seen yet by training several versions of our model that hold out entire countries from the training data.

Sectors validated: Residential / Non-residential / Total. Large facilities are estimated site by site from facility reports, production figures and published electricity use, so they are not part of the model checks; the comparison of the largest sites against grid operators’ own figures is below.

Metrics for validation: RMSE, root-mean-square error per area, in gigawatt-hours (GWh) per year; nRMSE, the RMSE as a share of the mean known demand of the scored areas; mdAPE, median absolute percentage error, the percentage half of the areas fall within. None of the three is the training objective: the model is fitted with a Tweedie loss, which suits consumption that is near zero in many areas and very large in a few. RMSE and nRMSE weigh the largest areas and the largest errors most, so they are the most sensitive to outliers; mdAPE, the error of a typical area, is the least.

Chart

Overall (all validated countries) · Scored on areas kept hidden from the model while training, and by a model that never saw the country or territory

Pooled over all regions with training data. Since every held-out area is scored together, countries with more held-out areas are weighted more heavily here.

Held-out areas are whole neighboring subregions the model never saw during training.

RMSE by scale, centered on the mean known demand

All validated countries · Residential

Neighborhood scale

9,945 areas · mean known demand 4.6 GWh

GWh per neighborhood per year

neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.4 km²): 9,945 areas · mean known demand 4.6 GWhOur estimate: RMSE ±1.9 GWh around the mean known demand. Population-only prediction: RMSE ±2.7 GWh around the mean known demandOur estimate: RMSE ±1.9 GWh around the mean known demand±1.9Population-only prediction: RMSE ±2.7 GWh around the mean known demand±2.7

Municipality scale

159 areas · mean known demand 150 GWh

GWh per municipality per year

municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 340 km²): 159 areas · mean known demand 150 GWhOur estimate: RMSE ±3.3 GWh around the mean known demand. Population-only prediction: RMSE ±12 GWh around the mean known demandOur estimate: RMSE ±3.3 GWh around the mean known demand±3.3Population-only prediction: RMSE ±12 GWh around the mean known demand±12

Sub-region scale

117 areas · mean known demand 370 GWh

GWh per sub-region per year

sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial) (Great Britain); FSA 2-character prefix (Canada); hidden EPCI Intercommunalities grouped by Département (partial) (France (metropolitan)); Statistical Area Level 4 (Australia), median 720 km²): 117 areas · mean known demand 370 GWhOur estimate: RMSE ±78 GWh around the mean known demand. Population-only prediction: RMSE ±180 GWh around the mean known demandOur estimate: RMSE ±78 GWh around the mean known demand±78Population-only prediction: RMSE ±180 GWh around the mean known demand±180

Country scale

Not scored¹

Neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.4 km²).

Municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 340 km²).

Sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial) (Great Britain); FSA 2-character prefix (Canada); hidden EPCI Intercommunalities grouped by Département (partial) (France (metropolitan)); Statistical Area Level 4 (Australia), median 720 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

All validated countries · Residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: All validated countries · Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.4 km²): ours 42%, poponly 59%; municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 340 km²): ours 2%, poponly 8%; sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial) (Great Britain); FSA 2-character prefix (Canada); hidden EPCI Intercommunalities grouped by Département (partial) (France (metropolitan)); Statistical Area Level 4 (Australia), median 720 km²): ours 21%, poponly 48%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.4 km²): nRMSE 59% of mean known demand; RMSE ±2.7 GWhPopulation-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 340 km²): nRMSE 8.3% of mean known demand; RMSE ±12 GWhPopulation-only prediction · sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial) (Great Britain); FSA 2-character prefix (Canada); hidden EPCI Intercommunalities grouped by Département (partial) (France (metropolitan)); Statistical Area Level 4 (Australia), median 720 km²): nRMSE 48% of mean known demand; RMSE ±180 GWhOur estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.4 km²): nRMSE 42% of mean known demand; RMSE ±1.9 GWhOur estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 340 km²): nRMSE 2.2% of mean known demand; RMSE ±3.3 GWhOur estimate · sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial) (Great Britain); FSA 2-character prefix (Canada); hidden EPCI Intercommunalities grouped by Département (partial) (France (metropolitan)); Statistical Area Level 4 (Australia), median 720 km²): nRMSE 21% of mean known demand; RMSE ±78 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

All validated countries · Residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: All validated countries · ResidentialAreas per bin at the neighborhood scale: 2,487 · 2,486 · 2,486 · 2,486. Areas per bin at the municipality scale: 40 · 40 · 39 · 40. Areas per bin at the sub-region scale: 30 · 29 · 29 · 29.Population-only prediction · neighborhood scale, areas 0.011–0.30 km² (2,487 areas): nRMSE 44% of mean known demandPopulation-only prediction · neighborhood scale, areas 0.30–1.4 km² (2,486 areas): nRMSE 38% of mean known demandPopulation-only prediction · neighborhood scale, areas 1.4–11 km² (2,486 areas): nRMSE 73% of mean known demandPopulation-only prediction · neighborhood scale, areas 11–490,000 km² (2,486 areas): nRMSE 44% of mean known demandPopulation-only prediction · municipality scale, areas 15–200 km² (40 areas): nRMSE 7.4% of mean known demandPopulation-only prediction · municipality scale, areas 200–340 km² (40 areas): nRMSE 9.5% of mean known demandPopulation-only prediction · municipality scale, areas 340–540 km² (39 areas): nRMSE 9.5% of mean known demandPopulation-only prediction · municipality scale, areas 540–26,000 km² (40 areas): nRMSE 7.4% of mean known demandPopulation-only prediction · sub-region scale, areas 6.7–320 km² (30 areas): nRMSE 68% of mean known demandPopulation-only prediction · sub-region scale, areas 320–720 km² (29 areas): nRMSE 53% of mean known demandPopulation-only prediction · sub-region scale, areas 720–1,300 km² (29 areas): nRMSE 40% of mean known demandPopulation-only prediction · sub-region scale, areas 1,300–320,000 km² (29 areas): nRMSE 25% of mean known demandOur estimate · neighborhood scale, areas 0.011–0.30 km² (2,487 areas): nRMSE 23% of mean known demandOur estimate · neighborhood scale, areas 0.30–1.4 km² (2,486 areas): nRMSE 21% of mean known demandOur estimate · neighborhood scale, areas 1.4–11 km² (2,486 areas): nRMSE 54% of mean known demandOur estimate · neighborhood scale, areas 11–490,000 km² (2,486 areas): nRMSE 31% of mean known demandOur estimate · municipality scale, areas 15–200 km² (40 areas): nRMSE 1.7% of mean known demandOur estimate · municipality scale, areas 200–340 km² (40 areas): nRMSE 2.2% of mean known demandOur estimate · municipality scale, areas 340–540 km² (39 areas): nRMSE 2.2% of mean known demandOur estimate · municipality scale, areas 540–26,000 km² (40 areas): nRMSE 2.4% of mean known demandOur estimate · sub-region scale, areas 6.7–320 km² (30 areas): nRMSE 36% of mean known demandOur estimate · sub-region scale, areas 320–720 km² (29 areas): nRMSE 10% of mean known demandOur estimate · sub-region scale, areas 720–1,300 km² (29 areas): nRMSE 7.4% of mean known demandOur estimate · sub-region scale, areas 1,300–320,000 km² (29 areas): nRMSE 15% of mean known demandneighborhoodmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: All validated countries · ResidentialAreas per class at the neighborhood scale: city centers 3,603 · towns and suburbs 1,795 · rural areas 4,547. Areas per class at the municipality scale: city centers 14 · rural areas 144.Our estimate · city centers, neighborhood scale (3,603 areas): nRMSE 53% of mean known demandPopulation-only prediction · city centers, neighborhood scale (3,603 areas): nRMSE 86% of mean known demandOur estimate · towns and suburbs, neighborhood scale (1,795 areas): nRMSE 29% of mean known demandPopulation-only prediction · towns and suburbs, neighborhood scale (1,795 areas): nRMSE 54% of mean known demandOur estimate · rural areas, neighborhood scale (4,547 areas): nRMSE 37% of mean known demandPopulation-only prediction · rural areas, neighborhood scale (4,547 areas): nRMSE 42% of mean known demandOur estimate · city centers, municipality scale (14 areas): nRMSE 0.97% of mean known demandPopulation-only prediction · city centers, municipality scale (14 areas): nRMSE 5.7% of mean known demandOur estimate · rural areas, municipality scale (144 areas): nRMSE 2.5% of mean known demandPopulation-only prediction · rural areas, municipality scale (144 areas): nRMSE 8.7% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

All held-out countries and territories · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — All held-out countries and territories, Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.8 km²): poponly 54%, ours 43%; municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 350 km²): poponly 22%, ours 10%; sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): poponly 33%, ours 14%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.8 km²): nRMSE 54% of mean known demand; RMSE ±2.0 GWh (mean known demand 3.7 GWh) — corrected to the known totals one level up (the municipality in Great Britain and France (metropolitan), the sub-region in Canada and Australia): what a user who knows those local totals would getPopulation-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 350 km²): nRMSE 22% of mean known demand; RMSE ±35 GWh (mean known demand 160 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): nRMSE 33% of mean known demand; RMSE ±470 GWh (mean known demand 1,400 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan)); Local Government Area (Australia), median 1.8 km²): nRMSE 43% of mean known demand; RMSE ±1.6 GWh (mean known demand 3.7 GWh) — corrected to the known totals one level up (the municipality in Great Britain and France (metropolitan), the sub-region in Canada and Australia): what a user who knows those local totals would getOur estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)), median 350 km²): nRMSE 9.6% of mean known demand; RMSE ±15 GWh (mean known demand 160 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): nRMSE 14% of mean known demand; RMSE ±190 GWh (mean known demand 1,400 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — All held-out countries and territories, Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan), Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 1.8 km²): poponly 131%, ours 132%; municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte), median 340 km²): poponly 99%, ours 25%; sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): poponly 33%, ours 14%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan), Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 1.8 km²): nRMSE 130% of mean known demand; RMSE ±4.8 GWh (mean known demand 3.7 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte), median 340 km²): nRMSE 99% of mean known demand; RMSE ±150 GWh (mean known demand 150 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): nRMSE 33% of mean known demand; RMSE ±470 GWh (mean known demand 1,400 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022 (Great Britain); Forward Sortation Area (Canada); IRIS (France (metropolitan), Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 1.8 km²): nRMSE 130% of mean known demand; RMSE ±4.9 GWh (mean known demand 3.7 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte), median 340 km²): nRMSE 25% of mean known demand; RMSE ±37 GWh (mean known demand 150 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion (Great Britain); FSA 2-character prefix (Canada); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,400 km²): nRMSE 14% of mean known demand; RMSE ±190 GWh (mean known demand 1,400 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

All validated countries · Non-residential

Municipality scale

218 areas · mean known demand 150 GWh

GWh per municipality per year

municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): 218 areas · mean known demand 150 GWhOur estimate: RMSE ±96 GWh around the mean known demand. Population-only prediction: RMSE ±170 GWh around the mean known demandOur estimate: RMSE ±96 GWh around the mean known demand±96Population-only prediction: RMSE ±170 GWh around the mean known demand±170

Sub-region scale

21 areas · mean known demand 1,500 GWh

GWh per sub-region per year

sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): 21 areas · mean known demand 1,500 GWhOur estimate: RMSE ±660 GWh around the mean known demand. Population-only prediction: RMSE ±880 GWh around the mean known demandOur estimate: RMSE ±660 GWh around the mean known demand±660Population-only prediction: RMSE ±880 GWh around the mean known demand±880

Country scale

Not scored¹

Municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²).

Sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²).

  • Where a lower bar touches zero, the RMSE is larger than the mean known demand at that scale.
  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

All validated countries · Non-residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: All validated countries · Non-residentialmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): ours 63%, poponly 111%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): ours 43%, poponly 57%Population-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 110% of mean known demand; RMSE ±170 GWhPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): nRMSE 57% of mean known demand; RMSE ±880 GWhOur estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 63% of mean known demand; RMSE ±96 GWhOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): nRMSE 43% of mean known demand; RMSE ±660 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

All validated countries · Non-residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: All validated countries · Non-residentialAreas per bin at the municipality scale: 55 · 54 · 54 · 55. Areas per bin at the sub-region scale: 6 · 5 · 5 · 5.Population-only prediction · municipality scale, areas 3.1–220 km² (55 areas): nRMSE 160% of mean known demandPopulation-only prediction · municipality scale, areas 220–430 km² (54 areas): nRMSE 55% of mean known demandPopulation-only prediction · municipality scale, areas 430–890 km² (54 areas): nRMSE 28% of mean known demandPopulation-only prediction · municipality scale, areas 890–99,000 km² (55 areas): nRMSE 36% of mean known demandPopulation-only prediction · sub-region scale, areas 110–3,800 km² (6 areas): nRMSE 68% of mean known demandPopulation-only prediction · sub-region scale, areas 3,800–6,000 km² (5 areas): nRMSE 23% of mean known demandPopulation-only prediction · sub-region scale, areas 6,000–7,300 km² (5 areas): nRMSE 12% of mean known demandPopulation-only prediction · sub-region scale, areas 7,300–320,000 km² (5 areas): nRMSE 29% of mean known demandOur estimate · municipality scale, areas 3.1–220 km² (55 areas): nRMSE 90% of mean known demandOur estimate · municipality scale, areas 220–430 km² (54 areas): nRMSE 43% of mean known demandOur estimate · municipality scale, areas 430–890 km² (54 areas): nRMSE 23% of mean known demandOur estimate · municipality scale, areas 890–99,000 km² (55 areas): nRMSE 28% of mean known demandOur estimate · sub-region scale, areas 110–3,800 km² (6 areas): nRMSE 49% of mean known demandOur estimate · sub-region scale, areas 3,800–6,000 km² (5 areas): nRMSE 21% of mean known demandOur estimate · sub-region scale, areas 6,000–7,300 km² (5 areas): nRMSE 11% of mean known demandOur estimate · sub-region scale, areas 7,300–320,000 km² (5 areas): nRMSE 42% of mean known demandmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: All validated countries · Non-residentialAreas per class at the municipality scale: city centers 20 · rural areas 196.Our estimate · city centers, municipality scale (20 areas): nRMSE 54% of mean known demandPopulation-only prediction · city centers, municipality scale (20 areas): nRMSE 100% of mean known demandOur estimate · rural areas, municipality scale (196 areas): nRMSE 40% of mean known demandPopulation-only prediction · rural areas, municipality scale (196 areas): nRMSE 48% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

All held-out countries and territories · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — All held-out countries and territories, Non-residentialmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): poponly 63%, ours 47%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): poponly 33%, ours 28%Population-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 63% of mean known demand; RMSE ±160 GWh (mean known demand 250 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 33% of mean known demand; RMSE ±750 GWh (mean known demand 2,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 47% of mean known demand; RMSE ±120 GWh (mean known demand 250 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 28% of mean known demand; RMSE ±640 GWh (mean known demand 2,300 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — All held-out countries and territories, Non-residentialmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): poponly 83%, ours 62%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): poponly 33%, ours 28%Population-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): nRMSE 83% of mean known demand; RMSE ±200 GWh (mean known demand 240 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 33% of mean known demand; RMSE ±750 GWh (mean known demand 2,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): nRMSE 62% of mean known demand; RMSE ±150 GWh (mean known demand 240 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 28% of mean known demand; RMSE ±640 GWh (mean known demand 2,300 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

All validated countries · Total

Municipality scale

218 areas · mean known demand 260 GWh

GWh per municipality per year

municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): 218 areas · mean known demand 260 GWhOur estimate: RMSE ±100 GWh around the mean known demand. Population-only prediction: RMSE ±180 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±180 GWh around the mean known demandOur estimate: RMSE ±100 GWh around the mean known demand±100Population-only prediction: RMSE ±180 GWh around the mean known demand±180Population-only prediction plus facilities: RMSE ±180 GWh around the mean known demand±180

Sub-region scale

21 areas · mean known demand 2,600 GWh

GWh per sub-region per year

sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): 21 areas · mean known demand 2,600 GWhOur estimate: RMSE ±780 GWh around the mean known demand. Population-only prediction: RMSE ±890 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±900 GWh around the mean known demandOur estimate: RMSE ±780 GWh around the mean known demand±780Population-only prediction: RMSE ±890 GWh around the mean known demand±890Population-only prediction plus facilities: RMSE ±900 GWh around the mean known demand±900

Country scale

Not scored¹

Municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²).

Sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

All validated countries · Total

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: All validated countries · Totalmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): ours 39%, poponly 68%, poponly_plus_facilities 68%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): ours 29%, poponly 34%, poponly_plus_facilities 34%Population-only prediction plus facilities · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 68% of mean known demand; RMSE ±180 GWhPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): nRMSE 34% of mean known demand; RMSE ±900 GWhPopulation-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 68% of mean known demand; RMSE ±180 GWhPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): nRMSE 34% of mean known demand; RMSE ±890 GWhOur estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 39% of mean known demand; RMSE ±100 GWhOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,500 km²): nRMSE 29% of mean known demand; RMSE ±780 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

All validated countries · Total

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: All validated countries · TotalAreas per bin at the municipality scale: 55 · 54 · 54 · 55. Areas per bin at the sub-region scale: 6 · 5 · 5 · 5.Population-only prediction plus facilities · municipality scale, areas 3.1–220 km² (55 areas): nRMSE 120% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 220–430 km² (54 areas): nRMSE 34% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 430–890 km² (54 areas): nRMSE 20% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 890–99,000 km² (55 areas): nRMSE 19% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 110–3,800 km² (6 areas): nRMSE 45% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 3,800–6,000 km² (5 areas): nRMSE 8.2% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 6,000–7,300 km² (5 areas): nRMSE 8.8% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 7,300–320,000 km² (5 areas): nRMSE 18% of mean known demandPopulation-only prediction · municipality scale, areas 3.1–220 km² (55 areas): nRMSE 120% of mean known demandPopulation-only prediction · municipality scale, areas 220–430 km² (54 areas): nRMSE 39% of mean known demandPopulation-only prediction · municipality scale, areas 430–890 km² (54 areas): nRMSE 21% of mean known demandPopulation-only prediction · municipality scale, areas 890–99,000 km² (55 areas): nRMSE 21% of mean known demandPopulation-only prediction · sub-region scale, areas 110–3,800 km² (6 areas): nRMSE 44% of mean known demandPopulation-only prediction · sub-region scale, areas 3,800–6,000 km² (5 areas): nRMSE 9.0% of mean known demandPopulation-only prediction · sub-region scale, areas 6,000–7,300 km² (5 areas): nRMSE 8.3% of mean known demandPopulation-only prediction · sub-region scale, areas 7,300–320,000 km² (5 areas): nRMSE 16% of mean known demandOur estimate · municipality scale, areas 3.1–220 km² (55 areas): nRMSE 65% of mean known demandOur estimate · municipality scale, areas 220–430 km² (54 areas): nRMSE 25% of mean known demandOur estimate · municipality scale, areas 430–890 km² (54 areas): nRMSE 14% of mean known demandOur estimate · municipality scale, areas 890–99,000 km² (55 areas): nRMSE 16% of mean known demandOur estimate · sub-region scale, areas 110–3,800 km² (6 areas): nRMSE 35% of mean known demandOur estimate · sub-region scale, areas 3,800–6,000 km² (5 areas): nRMSE 11% of mean known demandOur estimate · sub-region scale, areas 6,000–7,300 km² (5 areas): nRMSE 7.0% of mean known demandOur estimate · sub-region scale, areas 7,300–320,000 km² (5 areas): nRMSE 34% of mean known demandmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: All validated countries · TotalAreas per class at the municipality scale: city centers 20 · rural areas 196.Our estimate · city centers, municipality scale (20 areas): nRMSE 39% of mean known demandPopulation-only prediction · city centers, municipality scale (20 areas): nRMSE 72% of mean known demandPopulation-only prediction plus facilities · city centers, municipality scale (20 areas): nRMSE 72% of mean known demandOur estimate · rural areas, municipality scale (196 areas): nRMSE 24% of mean known demandPopulation-only prediction · rural areas, municipality scale (196 areas): nRMSE 30% of mean known demandPopulation-only prediction plus facilities · rural areas, municipality scale (196 areas): nRMSE 27% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

All held-out countries and territories · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — All held-out countries and territories, Totalmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): poponly_plus_facilities 38%, poponly 50%, ours 29%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): poponly_plus_facilities 26%, poponly 30%, ours 19%Population-only prediction plus facilities · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 38% of mean known demand; RMSE ±160 GWh (mean known demand 430 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 26% of mean known demand; RMSE ±1,000 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 50% of mean known demand; RMSE ±220 GWh (mean known demand 430 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 30% of mean known demand; RMSE ±1,200 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Local Government Area (Australia), median 390 km²): nRMSE 29% of mean known demand; RMSE ±120 GWh (mean known demand 430 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 19% of mean known demand; RMSE ±730 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — All held-out countries and territories, Totalmunicipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): poponly_plus_facilities 71%, poponly 86%, ours 39%; sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): poponly_plus_facilities 26%, poponly 30%, ours 19%Population-only prediction plus facilities · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): nRMSE 71% of mean known demand; RMSE ±290 GWh (mean known demand 410 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 26% of mean known demand; RMSE ±1,000 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): nRMSE 86% of mean known demand; RMSE ±350 GWh (mean known demand 410 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 30% of mean known demand; RMSE ±1,200 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District (Great Britain); EPCI Intercommunality (France (metropolitan)); Commune (Guadeloupe, Martinique, French Guiana, Réunion, Mayotte); Local Government Area (Australia), median 370 km²): nRMSE 39% of mean known demand; RMSE ±160 GWh (mean known demand 410 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion (Great Britain); Département (France (metropolitan)); Statistical Area Level 4 (Australia), median 5,700 km²): nRMSE 19% of mean known demand; RMSE ±730 GWh (mean known demand 3,900 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Australia · Scored on areas kept hidden from the model while training, and by a model that never saw the country or territory

Held-out areas are whole neighboring subregions the model never saw during training.

RMSE by scale, centered on the mean known demand

Australia · Residential

Neighborhood scale

253 areas · mean known demand 46 GWh

GWh per neighborhood per year

neighborhood scale (Local Government Area, median 45 km²): 253 areas · mean known demand 46 GWhOur estimate: RMSE ±4.4 GWh around the mean known demand. Population-only prediction: RMSE ±8.7 GWh around the mean known demandOur estimate: RMSE ±4.4 GWh around the mean known demand±4.4Population-only prediction: RMSE ±8.7 GWh around the mean known demand±8.7

Municipality scale

Not scored¹

Sub-region scale

10 areas · mean known demand 630 GWh

GWh per sub-region per year

sub-region scale (Statistical Area Level 4, median 590 km²): 10 areas · mean known demand 630 GWhOur estimate: RMSE ±49 GWh around the mean known demand. Population-only prediction: RMSE ±92 GWh around the mean known demandOur estimate: RMSE ±49 GWh around the mean known demand±49Population-only prediction: RMSE ±92 GWh around the mean known demand±92

Country scale

Not scored²

Neighborhood scale (Local Government Area, median 45 km²).

Sub-region scale (Statistical Area Level 4, median 590 km²).

  1. Not scored at the municipality scale: the small-zone tier is the Local Government Area tier here; this scale is the same as small_zone
  2. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Australia · Residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Australia · Residentialneighborhood scale (Local Government Area, median 45 km²): ours 10%, poponly 19%; sub-region scale (Statistical Area Level 4, median 590 km²): ours 8%, poponly 15%Population-only prediction · neighborhood scale (Local Government Area, median 45 km²): nRMSE 19% of mean known demand; RMSE ±8.7 GWhPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 590 km²): nRMSE 15% of mean known demand; RMSE ±92 GWhOur estimate · neighborhood scale (Local Government Area, median 45 km²): nRMSE 9.5% of mean known demand; RMSE ±4.4 GWhOur estimate · sub-region scale (Statistical Area Level 4, median 590 km²): nRMSE 7.7% of mean known demand; RMSE ±49 GWh

Median area size, km² (log scale)

  • Not scored at the municipality scale: the small-zone tier is the Local Government Area tier here; this scale is the same as small_zone
  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Australia · Residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Australia · ResidentialAreas per bin at the neighborhood scale: 64 · 63 · 63 · 63. Areas per bin at the sub-region scale: 5 · 5.Population-only prediction · neighborhood scale, areas 0.23–5.0 km² (64 areas): nRMSE 39% of mean known demandPopulation-only prediction · neighborhood scale, areas 5.0–22 km² (63 areas): nRMSE 18% of mean known demandPopulation-only prediction · neighborhood scale, areas 22–100 km² (63 areas): nRMSE 36% of mean known demandPopulation-only prediction · neighborhood scale, areas 100–99,000 km² (63 areas): nRMSE 9.6% of mean known demandPopulation-only prediction · sub-region scale, areas 56–2,800 km² (5 areas): nRMSE 14% of mean known demandPopulation-only prediction · sub-region scale, areas 2,800–320,000 km² (5 areas): nRMSE 15% of mean known demandOur estimate · neighborhood scale, areas 0.23–5.0 km² (64 areas): nRMSE 23% of mean known demandOur estimate · neighborhood scale, areas 5.0–22 km² (63 areas): nRMSE 22% of mean known demandOur estimate · neighborhood scale, areas 22–100 km² (63 areas): nRMSE 9.4% of mean known demandOur estimate · neighborhood scale, areas 100–99,000 km² (63 areas): nRMSE 4.8% of mean known demandOur estimate · sub-region scale, areas 56–2,800 km² (5 areas): nRMSE 8.3% of mean known demandOur estimate · sub-region scale, areas 2,800–320,000 km² (5 areas): nRMSE 6.6% of mean known demandneighborhoodsub-region

Median area size, km² (log scale)

By urbanization

By urbanization: Australia · ResidentialAreas per class at the neighborhood scale: city centers 44 · towns and suburbs 28 · rural areas 181.Our estimate · city centers, neighborhood scale (44 areas): nRMSE 11% of mean known demandPopulation-only prediction · city centers, neighborhood scale (44 areas): nRMSE 23% of mean known demandOur estimate · towns and suburbs, neighborhood scale (28 areas): nRMSE 15% of mean known demandPopulation-only prediction · towns and suburbs, neighborhood scale (28 areas): nRMSE 28% of mean known demandOur estimate · rural areas, neighborhood scale (181 areas): nRMSE 7.0% of mean known demandPopulation-only prediction · rural areas, neighborhood scale (181 areas): nRMSE 13% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Australia · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Australia, Residentialneighborhood scale (Local Government Area, median 3,500 km²): poponly 10%, ours 8%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly 27%, ours 26%Population-only prediction · neighborhood scale (Local Government Area, median 3,500 km²): nRMSE 9.6% of mean known demand; RMSE ±19 GWh (mean known demand 200 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 27% of mean known demand; RMSE ±200 GWh (mean known demand 730 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Local Government Area, median 3,500 km²): nRMSE 7.8% of mean known demand; RMSE ±15 GWh (mean known demand 200 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 26% of mean known demand; RMSE ±190 GWh (mean known demand 730 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Australia, Residentialneighborhood scale (Local Government Area, median 3,500 km²): poponly 37%, ours 46%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly 27%, ours 26%Population-only prediction · neighborhood scale (Local Government Area, median 3,500 km²): nRMSE 37% of mean known demand; RMSE ±74 GWh (mean known demand 200 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 27% of mean known demand; RMSE ±200 GWh (mean known demand 730 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Local Government Area, median 3,500 km²): nRMSE 46% of mean known demand; RMSE ±91 GWh (mean known demand 200 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 26% of mean known demand; RMSE ±190 GWh (mean known demand 730 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the municipality scale: the small-zone tier is the Local Government Area tier here; this scale is the same as small_zone
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

Australia · Non-residential

Municipality scale

63 areas · mean known demand 57 GWh

GWh per municipality per year

municipality scale (Local Government Area, median 360 km²): 63 areas · mean known demand 57 GWhOur estimate: RMSE ±46 GWh around the mean known demand. Population-only prediction: RMSE ±54 GWh around the mean known demandOur estimate: RMSE ±46 GWh around the mean known demand±46Population-only prediction: RMSE ±54 GWh around the mean known demand±54

Sub-region scale

6 areas · mean known demand 570 GWh

GWh per sub-region per year

sub-region scale (Statistical Area Level 4, median 5,100 km²): 6 areas · mean known demand 570 GWhOur estimate: RMSE ±210 GWh around the mean known demand. Population-only prediction: RMSE ±230 GWh around the mean known demandOur estimate: RMSE ±210 GWh around the mean known demand±210Population-only prediction: RMSE ±230 GWh around the mean known demand±230

Country scale

Not scored¹

Municipality scale (Local Government Area, median 360 km²).

Sub-region scale (Statistical Area Level 4, median 5,100 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Australia · Non-residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Australia · Non-residentialmunicipality scale (Local Government Area, median 360 km²): ours 80%, poponly 94%; sub-region scale (Statistical Area Level 4, median 5,100 km²): ours 37%, poponly 39%Population-only prediction · municipality scale (Local Government Area, median 360 km²): nRMSE 94% of mean known demand; RMSE ±54 GWhPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 5,100 km²): nRMSE 39% of mean known demand; RMSE ±230 GWhOur estimate · municipality scale (Local Government Area, median 360 km²): nRMSE 80% of mean known demand; RMSE ±46 GWhOur estimate · sub-region scale (Statistical Area Level 4, median 5,100 km²): nRMSE 37% of mean known demand; RMSE ±210 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Australia · Non-residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Australia · Non-residentialAreas per bin at the municipality scale: 16 · 16 · 15 · 16.Population-only prediction · municipality scale, areas 3.2–46 km² (16 areas): nRMSE 130% of mean known demandPopulation-only prediction · municipality scale, areas 46–460 km² (16 areas): nRMSE 45% of mean known demandPopulation-only prediction · municipality scale, areas 460–3,800 km² (15 areas): nRMSE 16% of mean known demandPopulation-only prediction · municipality scale, areas 3,800–99,000 km² (16 areas): nRMSE 60% of mean known demandOur estimate · municipality scale, areas 3.2–46 km² (16 areas): nRMSE 110% of mean known demandOur estimate · municipality scale, areas 46–460 km² (16 areas): nRMSE 69% of mean known demandOur estimate · municipality scale, areas 460–3,800 km² (15 areas): nRMSE 17% of mean known demandOur estimate · municipality scale, areas 3,800–99,000 km² (16 areas): nRMSE 120% of mean known demandmunicipality

Median area size, km² (log scale)

By urbanization

By urbanization: Australia · Non-residentialAreas per class at the municipality scale: city centers 8 · rural areas 53.Our estimate · city centers, municipality scale (8 areas): nRMSE 51% of mean known demandPopulation-only prediction · city centers, municipality scale (8 areas): nRMSE 96% of mean known demandOur estimate · rural areas, municipality scale (53 areas): nRMSE 91% of mean known demandPopulation-only prediction · rural areas, municipality scale (53 areas): nRMSE 95% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Australia · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Australia, Non-residentialmunicipality scale (Local Government Area, median 3,500 km²): poponly 40%, ours 32%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly 48%, ours 21%Population-only prediction · municipality scale (Local Government Area, median 3,500 km²): nRMSE 40% of mean known demand; RMSE ±120 GWh (mean known demand 300 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 48% of mean known demand; RMSE ±530 GWh (mean known demand 1,100 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Government Area, median 3,500 km²): nRMSE 32% of mean known demand; RMSE ±97 GWh (mean known demand 300 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 21% of mean known demand; RMSE ±230 GWh (mean known demand 1,100 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Australia, Non-residentialmunicipality scale (Local Government Area, median 3,500 km²): poponly 89%, ours 43%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly 48%, ours 21%Population-only prediction · municipality scale (Local Government Area, median 3,500 km²): nRMSE 89% of mean known demand; RMSE ±270 GWh (mean known demand 300 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 48% of mean known demand; RMSE ±530 GWh (mean known demand 1,100 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Government Area, median 3,500 km²): nRMSE 43% of mean known demand; RMSE ±130 GWh (mean known demand 300 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 21% of mean known demand; RMSE ±230 GWh (mean known demand 1,100 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

Australia · Total

Municipality scale

63 areas · mean known demand 110 GWh

GWh per municipality per year

municipality scale (Local Government Area, median 360 km²): 63 areas · mean known demand 110 GWhOur estimate: RMSE ±52 GWh around the mean known demand. Population-only prediction: RMSE ±60 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±53 GWh around the mean known demandOur estimate: RMSE ±52 GWh around the mean known demand±52Population-only prediction: RMSE ±60 GWh around the mean known demand±60Population-only prediction plus facilities: RMSE ±53 GWh around the mean known demand±53

Sub-region scale

6 areas · mean known demand 1,100 GWh

GWh per sub-region per year

sub-region scale (Statistical Area Level 4, median 5,100 km²): 6 areas · mean known demand 1,100 GWhOur estimate: RMSE ±260 GWh around the mean known demand. Population-only prediction: RMSE ±210 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±210 GWh around the mean known demandOur estimate: RMSE ±260 GWh around the mean known demand±260Population-only prediction: RMSE ±210 GWh around the mean known demand±210Population-only prediction plus facilities: RMSE ±210 GWh around the mean known demand±210

Country scale

Not scored¹

Municipality scale (Local Government Area, median 360 km²).

Sub-region scale (Statistical Area Level 4, median 5,100 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Australia · Total

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Australia · Totalmunicipality scale (Local Government Area, median 360 km²): ours 47%, poponly 54%, poponly_plus_facilities 47%; sub-region scale (Statistical Area Level 4, median 5,100 km²): ours 24%, poponly 19%, poponly_plus_facilities 18%Population-only prediction plus facilities · municipality scale (Local Government Area, median 360 km²): nRMSE 47% of mean known demand; RMSE ±53 GWhPopulation-only prediction plus facilities · sub-region scale (Statistical Area Level 4, median 5,100 km²): nRMSE 18% of mean known demand; RMSE ±210 GWhPopulation-only prediction · municipality scale (Local Government Area, median 360 km²): nRMSE 54% of mean known demand; RMSE ±60 GWhPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 5,100 km²): nRMSE 19% of mean known demand; RMSE ±210 GWhOur estimate · municipality scale (Local Government Area, median 360 km²): nRMSE 47% of mean known demand; RMSE ±52 GWhOur estimate · sub-region scale (Statistical Area Level 4, median 5,100 km²): nRMSE 24% of mean known demand; RMSE ±260 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Australia · Total

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Australia · TotalAreas per bin at the municipality scale: 16 · 16 · 15 · 16.Population-only prediction plus facilities · municipality scale, areas 3.2–46 km² (16 areas): nRMSE 83% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 46–460 km² (16 areas): nRMSE 21% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 460–3,800 km² (15 areas): nRMSE 6.8% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 3,800–99,000 km² (16 areas): nRMSE 22% of mean known demandPopulation-only prediction · municipality scale, areas 3.2–46 km² (16 areas): nRMSE 92% of mean known demandPopulation-only prediction · municipality scale, areas 46–460 km² (16 areas): nRMSE 21% of mean known demandPopulation-only prediction · municipality scale, areas 460–3,800 km² (15 areas): nRMSE 10% of mean known demandPopulation-only prediction · municipality scale, areas 3,800–99,000 km² (16 areas): nRMSE 38% of mean known demandOur estimate · municipality scale, areas 3.2–46 km² (16 areas): nRMSE 78% of mean known demandOur estimate · municipality scale, areas 46–460 km² (16 areas): nRMSE 36% of mean known demandOur estimate · municipality scale, areas 460–3,800 km² (15 areas): nRMSE 7.2% of mean known demandOur estimate · municipality scale, areas 3,800–99,000 km² (16 areas): nRMSE 53% of mean known demandmunicipality

Median area size, km² (log scale)

By urbanization

By urbanization: Australia · TotalAreas per class at the municipality scale: city centers 8 · rural areas 53.Our estimate · city centers, municipality scale (8 areas): nRMSE 31% of mean known demandPopulation-only prediction · city centers, municipality scale (8 areas): nRMSE 58% of mean known demandPopulation-only prediction plus facilities · city centers, municipality scale (8 areas): nRMSE 54% of mean known demandOur estimate · rural areas, municipality scale (53 areas): nRMSE 53% of mean known demandPopulation-only prediction · rural areas, municipality scale (53 areas): nRMSE 54% of mean known demandPopulation-only prediction plus facilities · rural areas, municipality scale (53 areas): nRMSE 47% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Australia · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Australia, Totalmunicipality scale (Local Government Area, median 3,500 km²): poponly_plus_facilities 24%, poponly 28%, ours 19%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly_plus_facilities 33%, poponly 35%, ours 20%Population-only prediction plus facilities · municipality scale (Local Government Area, median 3,500 km²): nRMSE 24% of mean known demand; RMSE ±120 GWh (mean known demand 510 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction plus facilities · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 33% of mean known demand; RMSE ±610 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Government Area, median 3,500 km²): nRMSE 28% of mean known demand; RMSE ±140 GWh (mean known demand 510 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 35% of mean known demand; RMSE ±660 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Government Area, median 3,500 km²): nRMSE 19% of mean known demand; RMSE ±99 GWh (mean known demand 510 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 20% of mean known demand; RMSE ±380 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Australia, Totalmunicipality scale (Local Government Area, median 3,500 km²): poponly_plus_facilities 56%, poponly 60%, ours 40%; sub-region scale (Statistical Area Level 4, median 7,200 km²): poponly_plus_facilities 33%, poponly 35%, ours 20%Population-only prediction plus facilities · municipality scale (Local Government Area, median 3,500 km²): nRMSE 56% of mean known demand; RMSE ±290 GWh (mean known demand 510 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction plus facilities · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 33% of mean known demand; RMSE ±610 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Government Area, median 3,500 km²): nRMSE 60% of mean known demand; RMSE ±310 GWh (mean known demand 510 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 35% of mean known demand; RMSE ±660 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Government Area, median 3,500 km²): nRMSE 40% of mean known demand; RMSE ±200 GWh (mean known demand 510 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Statistical Area Level 4, median 7,200 km²): nRMSE 20% of mean known demand; RMSE ±380 GWh (mean known demand 1,900 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Canada · Scored on areas kept hidden from the model while training, and by a model that never saw the country or territory

Held-out areas are whole neighboring subregions the model never saw during training.

RMSE by scale, centered on the mean known demand

Canada · Residential

Neighborhood scale

122 areas · mean known demand 98 GWh

GWh per neighborhood per year

neighborhood scale (Forward Sortation Area, median 63 km²): 122 areas · mean known demand 98 GWhOur estimate: RMSE ±15 GWh around the mean known demand. Population-only prediction: RMSE ±19 GWh around the mean known demandOur estimate: RMSE ±15 GWh around the mean known demand±15Population-only prediction: RMSE ±19 GWh around the mean known demand±19

Municipality scale

Not scored¹

Sub-region scale

19 areas · mean known demand 630 GWh

GWh per sub-region per year

sub-region scale (FSA 2-character prefix, median 3,400 km²): 19 areas · mean known demand 630 GWhOur estimate: RMSE ±200 GWh around the mean known demand. Population-only prediction: RMSE ±430 GWh around the mean known demandOur estimate: RMSE ±200 GWh around the mean known demand±200Population-only prediction: RMSE ±430 GWh around the mean known demand±430

Country scale

Not scored²

Neighborhood scale (Forward Sortation Area, median 63 km²).

Sub-region scale (FSA 2-character prefix, median 3,400 km²).

  1. Not scored at the municipality scale: the small-zone tier is the Forward Sortation Area tier here; this scale is the same as small_zone
  2. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Canada · Residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Canada · Residentialneighborhood scale (Forward Sortation Area, median 63 km²): ours 16%, poponly 19%; sub-region scale (FSA 2-character prefix, median 3,400 km²): ours 31%, poponly 68%Population-only prediction · neighborhood scale (Forward Sortation Area, median 63 km²): nRMSE 19% of mean known demand; RMSE ±19 GWhPopulation-only prediction · sub-region scale (FSA 2-character prefix, median 3,400 km²): nRMSE 68% of mean known demand; RMSE ±430 GWhOur estimate · neighborhood scale (Forward Sortation Area, median 63 km²): nRMSE 16% of mean known demand; RMSE ±15 GWhOur estimate · sub-region scale (FSA 2-character prefix, median 3,400 km²): nRMSE 31% of mean known demand; RMSE ±200 GWh

Median area size, km² (log scale)

  • Not scored at the municipality scale: the small-zone tier is the Forward Sortation Area tier here; this scale is the same as small_zone
  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Canada · Residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Canada · ResidentialAreas per bin at the neighborhood scale: 31 · 30 · 30 · 31. Areas per bin at the sub-region scale: 6 · 6 · 7.Population-only prediction · neighborhood scale, areas 1.4–16 km² (31 areas): nRMSE 31% of mean known demandPopulation-only prediction · neighborhood scale, areas 16–63 km² (30 areas): nRMSE 22% of mean known demandPopulation-only prediction · neighborhood scale, areas 63–350 km² (30 areas): nRMSE 21% of mean known demandPopulation-only prediction · neighborhood scale, areas 350–490,000 km² (31 areas): nRMSE 11% of mean known demandPopulation-only prediction · sub-region scale, areas 6.7–520 km² (6 areas): nRMSE 75% of mean known demandPopulation-only prediction · sub-region scale, areas 520–4,700 km² (6 areas): nRMSE 46% of mean known demandPopulation-only prediction · sub-region scale, areas 4,700–250,000 km² (7 areas): nRMSE 34% of mean known demandOur estimate · neighborhood scale, areas 1.4–16 km² (31 areas): nRMSE 19% of mean known demandOur estimate · neighborhood scale, areas 16–63 km² (30 areas): nRMSE 16% of mean known demandOur estimate · neighborhood scale, areas 63–350 km² (30 areas): nRMSE 15% of mean known demandOur estimate · neighborhood scale, areas 350–490,000 km² (31 areas): nRMSE 12% of mean known demandOur estimate · sub-region scale, areas 6.7–520 km² (6 areas): nRMSE 33% of mean known demandOur estimate · sub-region scale, areas 520–4,700 km² (6 areas): nRMSE 10% of mean known demandOur estimate · sub-region scale, areas 4,700–250,000 km² (7 areas): nRMSE 27% of mean known demandneighborhoodsub-region

Median area size, km² (log scale)

By urbanization

By urbanization: Canada · ResidentialAreas per class at the neighborhood scale: city centers 32 · rural areas 87.Our estimate · city centers, neighborhood scale (32 areas): nRMSE 18% of mean known demandPopulation-only prediction · city centers, neighborhood scale (32 areas): nRMSE 27% of mean known demandOur estimate · rural areas, neighborhood scale (87 areas): nRMSE 14% of mean known demandPopulation-only prediction · rural areas, neighborhood scale (87 areas): nRMSE 14% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Canada · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Canada, Residentialneighborhood scale (Forward Sortation Area, median 26 km²): poponly 20%, ours 17%; sub-region scale (FSA 2-character prefix, median 1,000 km²): poponly 37%, ours 17%Population-only prediction · neighborhood scale (Forward Sortation Area, median 26 km²): nRMSE 20% of mean known demand; RMSE ±17 GWh (mean known demand 85 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (FSA 2-character prefix, median 1,000 km²): nRMSE 37% of mean known demand; RMSE ±370 GWh (mean known demand 980 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Forward Sortation Area, median 26 km²): nRMSE 17% of mean known demand; RMSE ±14 GWh (mean known demand 85 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (FSA 2-character prefix, median 1,000 km²): nRMSE 17% of mean known demand; RMSE ±170 GWh (mean known demand 980 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Canada, Residentialneighborhood scale (Forward Sortation Area, median 26 km²): poponly 44%, ours 25%; sub-region scale (FSA 2-character prefix, median 1,000 km²): poponly 37%, ours 17%Population-only prediction · neighborhood scale (Forward Sortation Area, median 26 km²): nRMSE 44% of mean known demand; RMSE ±38 GWh (mean known demand 85 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (FSA 2-character prefix, median 1,000 km²): nRMSE 37% of mean known demand; RMSE ±370 GWh (mean known demand 980 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Forward Sortation Area, median 26 km²): nRMSE 25% of mean known demand; RMSE ±22 GWh (mean known demand 85 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (FSA 2-character prefix, median 1,000 km²): nRMSE 17% of mean known demand; RMSE ±170 GWh (mean known demand 980 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the municipality scale: the small-zone tier is the Forward Sortation Area tier here; this scale is the same as small_zone
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Non-residential

Not scored in this release:

  • Not scored at the municipality scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score
  • Not scored at the sub-region scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score
  • Not scored at the country scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score

Scored as a country without local data: a model that never saw it

Canada · scored by a model that never saw the country or territory

Non-residential

Not scored in this release:

  1. Not scored at the municipality scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score
  2. Not scored at the sub-region scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score
  3. Not scored at the country scale: Canada publishes no non-residential consumption at the Forward Sortation Area tier; nothing to score

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Total

Not scored in this release:

  • Not scored at the municipality scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score
  • Not scored at the sub-region scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score
  • Not scored at the country scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score

Scored as a country without local data: a model that never saw it

Canada · scored by a model that never saw the country or territory

Total

Not scored in this release:

  1. Not scored at the municipality scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score
  2. Not scored at the sub-region scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score
  3. Not scored at the country scale: Canada publishes no total consumption at the Forward Sortation Area tier; nothing to score

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

France (metropolitan) · Scored on areas kept hidden from the model while training, and by a model that never saw the country or territory

Held-out areas are whole neighboring subregions the model never saw during training.

RMSE by scale, centered on the mean known demand

France (metropolitan) · Residential

Neighborhood scale

4,361 areas · mean known demand 3.0 GWh

GWh per neighborhood per year

neighborhood scale (IRIS, median 8.2 km²): 4,361 areas · mean known demand 3.0 GWhOur estimate: RMSE ±1.0 GWh around the mean known demand. Population-only prediction: RMSE ±1.7 GWh around the mean known demandOur estimate: RMSE ±1.0 GWh around the mean known demand±1.0Population-only prediction: RMSE ±1.7 GWh around the mean known demand±1.7

Municipality scale

124 areas · mean known demand 110 GWh

GWh per municipality per year

municipality scale (EPCI Intercommunality, median 370 km²): 124 areas · mean known demand 110 GWhOur estimate: RMSE ±3.0 GWh around the mean known demand. Population-only prediction: RMSE ±11 GWh around the mean known demandOur estimate: RMSE ±3.0 GWh around the mean known demand±3.0Population-only prediction: RMSE ±11 GWh around the mean known demand±11

Sub-region scale

63 areas · mean known demand 210 GWh

GWh per sub-region per year

sub-region scale (hidden EPCI Intercommunalities grouped by Département (partial), median 760 km²): 63 areas · mean known demand 210 GWhOur estimate: RMSE ±16 GWh around the mean known demand. Population-only prediction: RMSE ±82 GWh around the mean known demandOur estimate: RMSE ±16 GWh around the mean known demand±16Population-only prediction: RMSE ±82 GWh around the mean known demand±82

Country scale

Not scored¹

Neighborhood scale (IRIS, median 8.2 km²).

Municipality scale (EPCI Intercommunality, median 370 km²).

Sub-region scale (hidden EPCI Intercommunalities grouped by Département (partial), median 760 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

France (metropolitan) · Residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: France (metropolitan) · Residentialneighborhood scale (IRIS, median 8.2 km²): ours 34%, poponly 55%; municipality scale (EPCI Intercommunality, median 370 km²): ours 3%, poponly 10%; sub-region scale (hidden EPCI Intercommunalities grouped by Département (partial), median 760 km²): ours 8%, poponly 39%Population-only prediction · neighborhood scale (IRIS, median 8.2 km²): nRMSE 55% of mean known demand; RMSE ±1.7 GWhPopulation-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 10% of mean known demand; RMSE ±11 GWhPopulation-only prediction · sub-region scale (hidden EPCI Intercommunalities grouped by Département (partial), median 760 km²): nRMSE 39% of mean known demand; RMSE ±82 GWhOur estimate · neighborhood scale (IRIS, median 8.2 km²): nRMSE 34% of mean known demand; RMSE ±1.0 GWhOur estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 2.8% of mean known demand; RMSE ±3.0 GWhOur estimate · sub-region scale (hidden EPCI Intercommunalities grouped by Département (partial), median 760 km²): nRMSE 7.5% of mean known demand; RMSE ±16 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

France (metropolitan) · Residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: France (metropolitan) · ResidentialAreas per bin at the neighborhood scale: 1,091 · 1,090 · 1,090 · 1,090. Areas per bin at the municipality scale: 31 · 31 · 31 · 31. Areas per bin at the sub-region scale: 16 · 16 · 15 · 16.Population-only prediction · neighborhood scale, areas 0.042–3.3 km² (1,091 areas): nRMSE 48% of mean known demandPopulation-only prediction · neighborhood scale, areas 3.3–8.2 km² (1,090 areas): nRMSE 69% of mean known demandPopulation-only prediction · neighborhood scale, areas 8.2–15 km² (1,090 areas): nRMSE 78% of mean known demandPopulation-only prediction · neighborhood scale, areas 15–180 km² (1,090 areas): nRMSE 38% of mean known demandPopulation-only prediction · municipality scale, areas 16–270 km² (31 areas): nRMSE 12% of mean known demandPopulation-only prediction · municipality scale, areas 270–370 km² (31 areas): nRMSE 11% of mean known demandPopulation-only prediction · municipality scale, areas 370–540 km² (31 areas): nRMSE 12% of mean known demandPopulation-only prediction · municipality scale, areas 540–1,300 km² (31 areas): nRMSE 7.7% of mean known demandPopulation-only prediction · sub-region scale, areas 66–390 km² (16 areas): nRMSE 52% of mean known demandPopulation-only prediction · sub-region scale, areas 390–760 km² (16 areas): nRMSE 38% of mean known demandPopulation-only prediction · sub-region scale, areas 760–1,000 km² (15 areas): nRMSE 42% of mean known demandPopulation-only prediction · sub-region scale, areas 1,000–2,900 km² (16 areas): nRMSE 15% of mean known demandOur estimate · neighborhood scale, areas 0.042–3.3 km² (1,091 areas): nRMSE 26% of mean known demandOur estimate · neighborhood scale, areas 3.3–8.2 km² (1,090 areas): nRMSE 27% of mean known demandOur estimate · neighborhood scale, areas 8.2–15 km² (1,090 areas): nRMSE 64% of mean known demandOur estimate · neighborhood scale, areas 15–180 km² (1,090 areas): nRMSE 20% of mean known demandOur estimate · municipality scale, areas 16–270 km² (31 areas): nRMSE 2.8% of mean known demandOur estimate · municipality scale, areas 270–370 km² (31 areas): nRMSE 3.1% of mean known demandOur estimate · municipality scale, areas 370–540 km² (31 areas): nRMSE 2.3% of mean known demandOur estimate · municipality scale, areas 540–1,300 km² (31 areas): nRMSE 2.7% of mean known demandOur estimate · sub-region scale, areas 66–390 km² (16 areas): nRMSE 5.9% of mean known demandOur estimate · sub-region scale, areas 390–760 km² (16 areas): nRMSE 13% of mean known demandOur estimate · sub-region scale, areas 760–1,000 km² (15 areas): nRMSE 6.3% of mean known demandOur estimate · sub-region scale, areas 1,000–2,900 km² (16 areas): nRMSE 4.3% of mean known demandneighborhoodmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: France (metropolitan) · ResidentialAreas per class at the neighborhood scale: city centers 581 · towns and suburbs 472 · rural areas 3,308. Areas per class at the municipality scale: rural areas 121.Our estimate · city centers, neighborhood scale (581 areas): nRMSE 23% of mean known demandPopulation-only prediction · city centers, neighborhood scale (581 areas): nRMSE 58% of mean known demandOur estimate · towns and suburbs, neighborhood scale (472 areas): nRMSE 23% of mean known demandPopulation-only prediction · towns and suburbs, neighborhood scale (472 areas): nRMSE 49% of mean known demandOur estimate · rural areas, neighborhood scale (3,308 areas): nRMSE 41% of mean known demandPopulation-only prediction · rural areas, neighborhood scale (3,308 areas): nRMSE 52% of mean known demandOur estimate · rural areas, municipality scale (121 areas): nRMSE 2.9% of mean known demandPopulation-only prediction · rural areas, municipality scale (121 areas): nRMSE 10% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

France (metropolitan) · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — France (metropolitan), Residentialneighborhood scale (IRIS, median 7.4 km²): poponly 49%, ours 36%; municipality scale (EPCI Intercommunality, median 370 km²): poponly 29%, ours 11%; sub-region scale (Département, median 6,000 km²): poponly 40%, ours 9%Population-only prediction · neighborhood scale (IRIS, median 7.4 km²): nRMSE 49% of mean known demand; RMSE ±1.6 GWh (mean known demand 3.2 GWh) — corrected to each municipality's known total: what a user who knows the municipality totals would getPopulation-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 29% of mean known demand; RMSE ±36 GWh (mean known demand 120 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 40% of mean known demand; RMSE ±660 GWh (mean known demand 1,700 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 7.4 km²): nRMSE 36% of mean known demand; RMSE ±1.1 GWh (mean known demand 3.2 GWh) — corrected to each municipality's known total: what a user who knows the municipality totals would getOur estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 11% of mean known demand; RMSE ±13 GWh (mean known demand 120 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 9.4% of mean known demand; RMSE ±160 GWh (mean known demand 1,700 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — France (metropolitan), Residentialneighborhood scale (IRIS, median 7.4 km²): poponly 65%, ours 41%; municipality scale (EPCI Intercommunality, median 370 km²): poponly 139%, ours 27%; sub-region scale (Département, median 6,000 km²): poponly 40%, ours 9%Population-only prediction · neighborhood scale (IRIS, median 7.4 km²): nRMSE 65% of mean known demand; RMSE ±2.1 GWh (mean known demand 3.2 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 140% of mean known demand; RMSE ±170 GWh (mean known demand 120 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 40% of mean known demand; RMSE ±660 GWh (mean known demand 1,700 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 7.4 km²): nRMSE 41% of mean known demand; RMSE ±1.3 GWh (mean known demand 3.2 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 27% of mean known demand; RMSE ±33 GWh (mean known demand 120 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 9.4% of mean known demand; RMSE ±160 GWh (mean known demand 1,700 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

France (metropolitan) · Non-residential

Municipality scale

119 areas · mean known demand 120 GWh

GWh per municipality per year

municipality scale (EPCI Intercommunality, median 420 km²): 119 areas · mean known demand 120 GWhOur estimate: RMSE ±38 GWh around the mean known demand. Population-only prediction: RMSE ±45 GWh around the mean known demandOur estimate: RMSE ±38 GWh around the mean known demand±38Population-only prediction: RMSE ±45 GWh around the mean known demand±45

Sub-region scale

10 areas · mean known demand 1,400 GWh

GWh per sub-region per year

sub-region scale (Département, median 6,600 km²): 10 areas · mean known demand 1,400 GWhOur estimate: RMSE ±200 GWh around the mean known demand. Population-only prediction: RMSE ±180 GWh around the mean known demandOur estimate: RMSE ±200 GWh around the mean known demand±200Population-only prediction: RMSE ±180 GWh around the mean known demand±180

Country scale

Not scored¹

Municipality scale (EPCI Intercommunality, median 420 km²).

Sub-region scale (Département, median 6,600 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

France (metropolitan) · Non-residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: France (metropolitan) · Non-residentialmunicipality scale (EPCI Intercommunality, median 420 km²): ours 32%, poponly 38%; sub-region scale (Département, median 6,600 km²): ours 14%, poponly 13%Population-only prediction · municipality scale (EPCI Intercommunality, median 420 km²): nRMSE 38% of mean known demand; RMSE ±45 GWhPopulation-only prediction · sub-region scale (Département, median 6,600 km²): nRMSE 13% of mean known demand; RMSE ±180 GWhOur estimate · municipality scale (EPCI Intercommunality, median 420 km²): nRMSE 32% of mean known demand; RMSE ±38 GWhOur estimate · sub-region scale (Département, median 6,600 km²): nRMSE 14% of mean known demand; RMSE ±200 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

France (metropolitan) · Non-residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: France (metropolitan) · Non-residentialAreas per bin at the municipality scale: 30 · 30 · 29 · 30. Areas per bin at the sub-region scale: 5 · 5.Population-only prediction · municipality scale, areas 3.2–290 km² (30 areas): nRMSE 44% of mean known demandPopulation-only prediction · municipality scale, areas 290–420 km² (30 areas): nRMSE 41% of mean known demandPopulation-only prediction · municipality scale, areas 420–700 km² (29 areas): nRMSE 25% of mean known demandPopulation-only prediction · municipality scale, areas 700–2,000 km² (30 areas): nRMSE 36% of mean known demandPopulation-only prediction · sub-region scale, areas 3,800–6,600 km² (5 areas): nRMSE 19% of mean known demandPopulation-only prediction · sub-region scale, areas 6,600–7,400 km² (5 areas): nRMSE 8.3% of mean known demandOur estimate · municipality scale, areas 3.2–290 km² (30 areas): nRMSE 24% of mean known demandOur estimate · municipality scale, areas 290–420 km² (30 areas): nRMSE 44% of mean known demandOur estimate · municipality scale, areas 420–700 km² (29 areas): nRMSE 27% of mean known demandOur estimate · municipality scale, areas 700–2,000 km² (30 areas): nRMSE 28% of mean known demandOur estimate · sub-region scale, areas 3,800–6,600 km² (5 areas): nRMSE 14% of mean known demandOur estimate · sub-region scale, areas 6,600–7,400 km² (5 areas): nRMSE 14% of mean known demandmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: France (metropolitan) · Non-residentialAreas per class at the municipality scale: rural areas 119.Our estimate · rural areas, municipality scale (119 areas): nRMSE 32% of mean known demandPopulation-only prediction · rural areas, municipality scale (119 areas): nRMSE 38% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

France (metropolitan) · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — France (metropolitan), Non-residentialmunicipality scale (EPCI Intercommunality, median 370 km²): poponly 77%, ours 61%; sub-region scale (Département, median 6,000 km²): poponly 30%, ours 25%Population-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 77% of mean known demand; RMSE ±150 GWh (mean known demand 190 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 30% of mean known demand; RMSE ±780 GWh (mean known demand 2,500 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 61% of mean known demand; RMSE ±120 GWh (mean known demand 190 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 25% of mean known demand; RMSE ±630 GWh (mean known demand 2,500 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — France (metropolitan), Non-residentialmunicipality scale (EPCI Intercommunality, median 370 km²): poponly 95%, ours 72%; sub-region scale (Département, median 6,000 km²): poponly 30%, ours 25%Population-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 95% of mean known demand; RMSE ±190 GWh (mean known demand 190 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 30% of mean known demand; RMSE ±780 GWh (mean known demand 2,500 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 72% of mean known demand; RMSE ±140 GWh (mean known demand 190 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 25% of mean known demand; RMSE ±630 GWh (mean known demand 2,500 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

France (metropolitan) · Total

Municipality scale

119 areas · mean known demand 220 GWh

GWh per municipality per year

municipality scale (EPCI Intercommunality, median 420 km²): 119 areas · mean known demand 220 GWhOur estimate: RMSE ±38 GWh around the mean known demand. Population-only prediction: RMSE ±54 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±48 GWh around the mean known demandOur estimate: RMSE ±38 GWh around the mean known demand±38Population-only prediction: RMSE ±54 GWh around the mean known demand±54Population-only prediction plus facilities: RMSE ±48 GWh around the mean known demand±48

Sub-region scale

10 areas · mean known demand 2,600 GWh

GWh per sub-region per year

sub-region scale (Département, median 6,600 km²): 10 areas · mean known demand 2,600 GWhOur estimate: RMSE ±220 GWh around the mean known demand. Population-only prediction: RMSE ±230 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±230 GWh around the mean known demandOur estimate: RMSE ±220 GWh around the mean known demand±220Population-only prediction: RMSE ±230 GWh around the mean known demand±230Population-only prediction plus facilities: RMSE ±230 GWh around the mean known demand±230

Country scale

Not scored¹

Municipality scale (EPCI Intercommunality, median 420 km²).

Sub-region scale (Département, median 6,600 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

France (metropolitan) · Total

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: France (metropolitan) · Totalmunicipality scale (EPCI Intercommunality, median 420 km²): ours 18%, poponly 25%, poponly_plus_facilities 22%; sub-region scale (Département, median 6,600 km²): ours 9%, poponly 9%, poponly_plus_facilities 9%Population-only prediction plus facilities · municipality scale (EPCI Intercommunality, median 420 km²): nRMSE 22% of mean known demand; RMSE ±48 GWhPopulation-only prediction plus facilities · sub-region scale (Département, median 6,600 km²): nRMSE 9.1% of mean known demand; RMSE ±230 GWhPopulation-only prediction · municipality scale (EPCI Intercommunality, median 420 km²): nRMSE 25% of mean known demand; RMSE ±54 GWhPopulation-only prediction · sub-region scale (Département, median 6,600 km²): nRMSE 8.8% of mean known demand; RMSE ±230 GWhOur estimate · municipality scale (EPCI Intercommunality, median 420 km²): nRMSE 18% of mean known demand; RMSE ±38 GWhOur estimate · sub-region scale (Département, median 6,600 km²): nRMSE 8.5% of mean known demand; RMSE ±220 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

France (metropolitan) · Total

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: France (metropolitan) · TotalAreas per bin at the municipality scale: 30 · 30 · 29 · 30. Areas per bin at the sub-region scale: 5 · 5.Population-only prediction plus facilities · municipality scale, areas 3.2–290 km² (30 areas): nRMSE 22% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 290–420 km² (30 areas): nRMSE 19% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 420–700 km² (29 areas): nRMSE 20% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 700–2,000 km² (30 areas): nRMSE 21% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 3,800–6,600 km² (5 areas): nRMSE 12% of mean known demandPopulation-only prediction plus facilities · sub-region scale, areas 6,600–7,400 km² (5 areas): nRMSE 7.5% of mean known demandPopulation-only prediction · municipality scale, areas 3.2–290 km² (30 areas): nRMSE 29% of mean known demandPopulation-only prediction · municipality scale, areas 290–420 km² (30 areas): nRMSE 24% of mean known demandPopulation-only prediction · municipality scale, areas 420–700 km² (29 areas): nRMSE 21% of mean known demandPopulation-only prediction · municipality scale, areas 700–2,000 km² (30 areas): nRMSE 23% of mean known demandPopulation-only prediction · sub-region scale, areas 3,800–6,600 km² (5 areas): nRMSE 13% of mean known demandPopulation-only prediction · sub-region scale, areas 6,600–7,400 km² (5 areas): nRMSE 6.3% of mean known demandOur estimate · municipality scale, areas 3.2–290 km² (30 areas): nRMSE 13% of mean known demandOur estimate · municipality scale, areas 290–420 km² (30 areas): nRMSE 23% of mean known demandOur estimate · municipality scale, areas 420–700 km² (29 areas): nRMSE 16% of mean known demandOur estimate · municipality scale, areas 700–2,000 km² (30 areas): nRMSE 16% of mean known demandOur estimate · sub-region scale, areas 3,800–6,600 km² (5 areas): nRMSE 7.4% of mean known demandOur estimate · sub-region scale, areas 6,600–7,400 km² (5 areas): nRMSE 8.5% of mean known demandmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: France (metropolitan) · TotalAreas per class at the municipality scale: rural areas 119.Our estimate · rural areas, municipality scale (119 areas): nRMSE 18% of mean known demandPopulation-only prediction · rural areas, municipality scale (119 areas): nRMSE 25% of mean known demandPopulation-only prediction plus facilities · rural areas, municipality scale (119 areas): nRMSE 22% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

France (metropolitan) · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — France (metropolitan), Totalmunicipality scale (EPCI Intercommunality, median 370 km²): poponly_plus_facilities 47%, poponly 69%, ours 36%; sub-region scale (Département, median 6,000 km²): poponly_plus_facilities 27%, poponly 34%, ours 14%Population-only prediction plus facilities · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 47% of mean known demand; RMSE ±160 GWh (mean known demand 330 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction plus facilities · sub-region scale (Département, median 6,000 km²): nRMSE 27% of mean known demand; RMSE ±1,200 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 69% of mean known demand; RMSE ±230 GWh (mean known demand 330 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 34% of mean known demand; RMSE ±1,500 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 36% of mean known demand; RMSE ±120 GWh (mean known demand 330 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 14% of mean known demand; RMSE ±600 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — France (metropolitan), Totalmunicipality scale (EPCI Intercommunality, median 370 km²): poponly_plus_facilities 95%, poponly 119%, ours 40%; sub-region scale (Département, median 6,000 km²): poponly_plus_facilities 27%, poponly 34%, ours 14%Population-only prediction plus facilities · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 95% of mean known demand; RMSE ±310 GWh (mean known demand 330 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction plus facilities · sub-region scale (Département, median 6,000 km²): nRMSE 27% of mean known demand; RMSE ±1,200 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 120% of mean known demand; RMSE ±400 GWh (mean known demand 330 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (Département, median 6,000 km²): nRMSE 34% of mean known demand; RMSE ±1,500 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (EPCI Intercommunality, median 370 km²): nRMSE 40% of mean known demand; RMSE ±130 GWh (mean known demand 330 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (Département, median 6,000 km²): nRMSE 14% of mean known demand; RMSE ±600 GWh (mean known demand 4,300 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

French Guiana · Scored only by a model that never saw the country or territory (a single split block)

Too few districts to hold any out: the whole territory is one block.

Scored as a country without local data: a model that never saw it

French Guiana · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — French Guiana, Residentialneighborhood scale (IRIS, median 2.7 km²): poponly 82%, ours 59%; municipality scale (Commune, median 2,100 km²): poponly 68%, ours 44%Population-only prediction · neighborhood scale (IRIS, median 2.7 km²): nRMSE 82% of mean known demand; RMSE ±3.5 GWh (mean known demand 4.3 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 2,100 km²): nRMSE 68% of mean known demand; RMSE ±11 GWh (mean known demand 17 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 2.7 km²): nRMSE 59% of mean known demand; RMSE ±2.5 GWh (mean known demand 4.3 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 2,100 km²): nRMSE 44% of mean known demand; RMSE ±7.3 GWh (mean known demand 17 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: French Guiana is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

French Guiana · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — French Guiana, Non-residentialmunicipality scale (Commune, median 2,200 km²): poponly 123%, ours 95%Population-only prediction · municipality scale (Commune, median 2,200 km²): nRMSE 120% of mean known demand; RMSE ±32 GWh (mean known demand 26 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 2,200 km²): nRMSE 95% of mean known demand; RMSE ±25 GWh (mean known demand 26 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: French Guiana is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

French Guiana · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — French Guiana, Totalmunicipality scale (Commune, median 2,200 km²): poponly_plus_facilities 97%, poponly 97%, ours 68%Population-only prediction plus facilities · municipality scale (Commune, median 2,200 km²): nRMSE 97% of mean known demand; RMSE ±41 GWh (mean known demand 42 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 2,200 km²): nRMSE 97% of mean known demand; RMSE ±41 GWh (mean known demand 42 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 2,200 km²): nRMSE 68% of mean known demand; RMSE ±29 GWh (mean known demand 42 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: French Guiana is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Great Britain · Scored on areas kept hidden from the model while training, and by a model that never saw the country or territory

Held-out areas are whole neighboring subregions the model never saw during training.

RMSE by scale, centered on the mean known demand

Great Britain · Residential

Neighborhood scale

5,209 areas · mean known demand 2.0 GWh

GWh per neighborhood per year

neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.34 km²): 5,209 areas · mean known demand 2.0 GWhOur estimate: RMSE ±0.36 GWh around the mean known demand. Population-only prediction: RMSE ±0.70 GWh around the mean known demandOur estimate: RMSE ±0.36 GWh around the mean known demand±0.36Population-only prediction: RMSE ±0.70 GWh around the mean known demand±0.70

Municipality scale

35 areas · mean known demand 300 GWh

GWh per municipality per year

municipality scale (Local Authority District, median 200 km²): 35 areas · mean known demand 300 GWhOur estimate: RMSE ±4.1 GWh around the mean known demand. Population-only prediction: RMSE ±16 GWh around the mean known demandOur estimate: RMSE ±4.1 GWh around the mean known demand±4.1Population-only prediction: RMSE ±16 GWh around the mean known demand±16

Sub-region scale

25 areas · mean known demand 420 GWh

GWh per sub-region per year

sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial), median 330 km²): 25 areas · mean known demand 420 GWhOur estimate: RMSE ±14 GWh around the mean known demand. Population-only prediction: RMSE ±72 GWh around the mean known demandOur estimate: RMSE ±14 GWh around the mean known demand±14Population-only prediction: RMSE ±72 GWh around the mean known demand±72

Country scale

Not scored¹

Neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.34 km²).

Municipality scale (Local Authority District, median 200 km²).

Sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial), median 330 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Great Britain · Residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Great Britain · Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.34 km²): ours 18%, poponly 34%; municipality scale (Local Authority District, median 200 km²): ours 1%, poponly 5%; sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial), median 330 km²): ours 3%, poponly 17%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.34 km²): nRMSE 34% of mean known demand; RMSE ±0.70 GWhPopulation-only prediction · municipality scale (Local Authority District, median 200 km²): nRMSE 5.3% of mean known demand; RMSE ±16 GWhPopulation-only prediction · sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial), median 330 km²): nRMSE 17% of mean known demand; RMSE ±72 GWhOur estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.34 km²): nRMSE 18% of mean known demand; RMSE ±0.36 GWhOur estimate · municipality scale (Local Authority District, median 200 km²): nRMSE 1.4% of mean known demand; RMSE ±4.1 GWhOur estimate · sub-region scale (hidden Local Authority Districts grouped by ITL2 subregion (partial), median 330 km²): nRMSE 3.3% of mean known demand; RMSE ±14 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Great Britain · Residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Great Britain · ResidentialAreas per bin at the neighborhood scale: 1,303 · 1,302 · 1,302 · 1,302. Areas per bin at the municipality scale: 9 · 9 · 8 · 9. Areas per bin at the sub-region scale: 7 · 6 · 6 · 6.Population-only prediction · neighborhood scale, areas 0.011–0.19 km² (1,303 areas): nRMSE 53% of mean known demandPopulation-only prediction · neighborhood scale, areas 0.19–0.34 km² (1,302 areas): nRMSE 32% of mean known demandPopulation-only prediction · neighborhood scale, areas 0.34–0.88 km² (1,302 areas): nRMSE 28% of mean known demandPopulation-only prediction · neighborhood scale, areas 0.88–1,200 km² (1,302 areas): nRMSE 31% of mean known demandPopulation-only prediction · municipality scale, areas 15–97 km² (9 areas): nRMSE 9.2% of mean known demandPopulation-only prediction · municipality scale, areas 97–200 km² (9 areas): nRMSE 2.4% of mean known demandPopulation-only prediction · municipality scale, areas 200–460 km² (8 areas): nRMSE 3.1% of mean known demandPopulation-only prediction · municipality scale, areas 460–26,000 km² (9 areas): nRMSE 5.9% of mean known demandPopulation-only prediction · sub-region scale, areas 15–170 km² (7 areas): nRMSE 18% of mean known demandPopulation-only prediction · sub-region scale, areas 170–330 km² (6 areas): nRMSE 10% of mean known demandPopulation-only prediction · sub-region scale, areas 330–820 km² (6 areas): nRMSE 5.1% of mean known demandPopulation-only prediction · sub-region scale, areas 820–26,000 km² (6 areas): nRMSE 22% of mean known demandOur estimate · neighborhood scale, areas 0.011–0.19 km² (1,303 areas): nRMSE 27% of mean known demandOur estimate · neighborhood scale, areas 0.19–0.34 km² (1,302 areas): nRMSE 16% of mean known demandOur estimate · neighborhood scale, areas 0.34–0.88 km² (1,302 areas): nRMSE 16% of mean known demandOur estimate · neighborhood scale, areas 0.88–1,200 km² (1,302 areas): nRMSE 15% of mean known demandOur estimate · municipality scale, areas 15–97 km² (9 areas): nRMSE 1.9% of mean known demandOur estimate · municipality scale, areas 97–200 km² (9 areas): nRMSE 0.47% of mean known demandOur estimate · municipality scale, areas 200–460 km² (8 areas): nRMSE 0.88% of mean known demandOur estimate · municipality scale, areas 460–26,000 km² (9 areas): nRMSE 1.8% of mean known demandOur estimate · sub-region scale, areas 15–170 km² (7 areas): nRMSE 4.5% of mean known demandOur estimate · sub-region scale, areas 170–330 km² (6 areas): nRMSE 3.2% of mean known demandOur estimate · sub-region scale, areas 330–820 km² (6 areas): nRMSE 2.4% of mean known demandOur estimate · sub-region scale, areas 820–26,000 km² (6 areas): nRMSE 2.6% of mean known demandneighborhoodmunicipalitysub-region

Median area size, km² (log scale)

By urbanization

By urbanization: Great Britain · ResidentialAreas per class at the neighborhood scale: city centers 2,946 · towns and suburbs 1,292 · rural areas 971. Areas per class at the municipality scale: city centers 12 · rural areas 23.Our estimate · city centers, neighborhood scale (2,946 areas): nRMSE 20% of mean known demandPopulation-only prediction · city centers, neighborhood scale (2,946 areas): nRMSE 36% of mean known demandOur estimate · towns and suburbs, neighborhood scale (1,292 areas): nRMSE 15% of mean known demandPopulation-only prediction · towns and suburbs, neighborhood scale (1,292 areas): nRMSE 32% of mean known demandOur estimate · rural areas, neighborhood scale (971 areas): nRMSE 15% of mean known demandPopulation-only prediction · rural areas, neighborhood scale (971 areas): nRMSE 32% of mean known demandOur estimate · city centers, municipality scale (12 areas): nRMSE 1.1% of mean known demandPopulation-only prediction · city centers, municipality scale (12 areas): nRMSE 5.2% of mean known demandOur estimate · rural areas, municipality scale (23 areas): nRMSE 1.5% of mean known demandPopulation-only prediction · rural areas, municipality scale (23 areas): nRMSE 5.4% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Great Britain · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Great Britain, Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): poponly 30%, ours 24%; municipality scale (Local Authority District, median 250 km²): poponly 11%, ours 8%; sub-region scale (ITL2 subregion, median 3,700 km²): poponly 14%, ours 12%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): nRMSE 30% of mean known demand; RMSE ±0.70 GWh (mean known demand 2.3 GWh) — corrected to each municipality's known total: what a user who knows the municipality totals would getPopulation-only prediction · municipality scale (Local Authority District, median 250 km²): nRMSE 11% of mean known demand; RMSE ±32 GWh (mean known demand 280 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,700 km²): nRMSE 14% of mean known demand; RMSE ±300 GWh (mean known demand 2,200 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): nRMSE 24% of mean known demand; RMSE ±0.55 GWh (mean known demand 2.3 GWh) — corrected to each municipality's known total: what a user who knows the municipality totals would getOur estimate · municipality scale (Local Authority District, median 250 km²): nRMSE 7.6% of mean known demand; RMSE ±21 GWh (mean known demand 280 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion, median 3,700 km²): nRMSE 12% of mean known demand; RMSE ±260 GWh (mean known demand 2,200 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Great Britain, Residentialneighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): poponly 36%, ours 28%; municipality scale (Local Authority District, median 250 km²): poponly 21%, ours 19%; sub-region scale (ITL2 subregion, median 3,700 km²): poponly 14%, ours 12%Population-only prediction · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): nRMSE 36% of mean known demand; RMSE ±0.82 GWh (mean known demand 2.3 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · municipality scale (Local Authority District, median 250 km²): nRMSE 21% of mean known demand; RMSE ±58 GWh (mean known demand 280 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,700 km²): nRMSE 14% of mean known demand; RMSE ±300 GWh (mean known demand 2,200 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (Lower Layer Super Output Area 2021 / Scottish Data Zone 2022, median 0.41 km²): nRMSE 28% of mean known demand; RMSE ±0.64 GWh (mean known demand 2.3 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · municipality scale (Local Authority District, median 250 km²): nRMSE 19% of mean known demand; RMSE ±53 GWh (mean known demand 280 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion, median 3,700 km²): nRMSE 12% of mean known demand; RMSE ±260 GWh (mean known demand 2,200 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

Great Britain · Non-residential

Municipality scale

36 areas · mean known demand 620 GWh

GWh per municipality per year

municipality scale (Local Authority District, median 360 km²): 36 areas · mean known demand 620 GWhOur estimate: RMSE ±250 GWh around the mean known demand. Population-only prediction: RMSE ±470 GWh around the mean known demandOur estimate: RMSE ±250 GWh around the mean known demand±250Population-only prediction: RMSE ±470 GWh around the mean known demand±470

Sub-region scale

5 areas · mean known demand 4,300 GWh

GWh per sub-region per year

sub-region scale (ITL2 subregion, median 2,000 km²): 5 areas · mean known demand 4,300 GWhOur estimate: RMSE ±1,500 GWh around the mean known demand. Population-only prediction: RMSE ±2,000 GWh around the mean known demandOur estimate: RMSE ±1,500 GWh around the mean known demand±1,500Population-only prediction: RMSE ±2,000 GWh around the mean known demand±2,000

Country scale

Not scored¹

Municipality scale (Local Authority District, median 360 km²).

Sub-region scale (ITL2 subregion, median 2,000 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Great Britain · Non-residential

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Great Britain · Non-residentialmunicipality scale (Local Authority District, median 360 km²): ours 41%, poponly 76%; sub-region scale (ITL2 subregion, median 2,000 km²): ours 35%, poponly 47%Population-only prediction · municipality scale (Local Authority District, median 360 km²): nRMSE 76% of mean known demand; RMSE ±470 GWhPopulation-only prediction · sub-region scale (ITL2 subregion, median 2,000 km²): nRMSE 47% of mean known demand; RMSE ±2,000 GWhOur estimate · municipality scale (Local Authority District, median 360 km²): nRMSE 41% of mean known demand; RMSE ±250 GWhOur estimate · sub-region scale (ITL2 subregion, median 2,000 km²): nRMSE 35% of mean known demand; RMSE ±1,500 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Great Britain · Non-residential

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Great Britain · Non-residentialAreas per bin at the municipality scale: 9 · 9 · 9 · 9.Population-only prediction · municipality scale, areas 3.1–170 km² (9 areas): nRMSE 91% of mean known demandPopulation-only prediction · municipality scale, areas 170–390 km² (9 areas): nRMSE 30% of mean known demandPopulation-only prediction · municipality scale, areas 390–1,000 km² (9 areas): nRMSE 20% of mean known demandPopulation-only prediction · municipality scale, areas 1,000–3,000 km² (9 areas): nRMSE 21% of mean known demandOur estimate · municipality scale, areas 3.1–170 km² (9 areas): nRMSE 48% of mean known demandOur estimate · municipality scale, areas 170–390 km² (9 areas): nRMSE 23% of mean known demandOur estimate · municipality scale, areas 390–1,000 km² (9 areas): nRMSE 15% of mean known demandOur estimate · municipality scale, areas 1,000–3,000 km² (9 areas): nRMSE 13% of mean known demandmunicipality

Median area size, km² (log scale)

By urbanization

By urbanization: Great Britain · Non-residentialAreas per class at the municipality scale: city centers 12 · rural areas 24.Our estimate · city centers, municipality scale (12 areas): nRMSE 37% of mean known demandPopulation-only prediction · city centers, municipality scale (12 areas): nRMSE 69% of mean known demandOur estimate · rural areas, municipality scale (24 areas): nRMSE 19% of mean known demandPopulation-only prediction · rural areas, municipality scale (24 areas): nRMSE 27% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Great Britain · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Great Britain, Non-residentialmunicipality scale (Local Authority District, median 270 km²): poponly 48%, ours 33%; sub-region scale (ITL2 subregion, median 3,900 km²): poponly 28%, ours 29%Population-only prediction · municipality scale (Local Authority District, median 270 km²): nRMSE 48% of mean known demand; RMSE ±200 GWh (mean known demand 420 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 28% of mean known demand; RMSE ±920 GWh (mean known demand 3,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District, median 270 km²): nRMSE 33% of mean known demand; RMSE ±140 GWh (mean known demand 420 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 29% of mean known demand; RMSE ±960 GWh (mean known demand 3,300 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Great Britain, Non-residentialmunicipality scale (Local Authority District, median 270 km²): poponly 52%, ours 48%; sub-region scale (ITL2 subregion, median 3,900 km²): poponly 28%, ours 29%Population-only prediction · municipality scale (Local Authority District, median 270 km²): nRMSE 52% of mean known demand; RMSE ±220 GWh (mean known demand 420 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 28% of mean known demand; RMSE ±920 GWh (mean known demand 3,300 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District, median 270 km²): nRMSE 48% of mean known demand; RMSE ±200 GWh (mean known demand 420 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 29% of mean known demand; RMSE ±960 GWh (mean known demand 3,300 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

RMSE by scale, centered on the mean known demand

Great Britain · Total

Municipality scale

36 areas · mean known demand 950 GWh

GWh per municipality per year

municipality scale (Local Authority District, median 360 km²): 36 areas · mean known demand 950 GWhOur estimate: RMSE ±260 GWh around the mean known demand. Population-only prediction: RMSE ±490 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±480 GWh around the mean known demandOur estimate: RMSE ±260 GWh around the mean known demand±260Population-only prediction: RMSE ±490 GWh around the mean known demand±490Population-only prediction plus facilities: RMSE ±480 GWh around the mean known demand±480

Sub-region scale

5 areas · mean known demand 6,700 GWh

GWh per sub-region per year

sub-region scale (ITL2 subregion, median 2,000 km²): 5 areas · mean known demand 6,700 GWhOur estimate: RMSE ±1,800 GWh around the mean known demand. Population-only prediction: RMSE ±2,100 GWh around the mean known demand. Population-only prediction plus facilities: RMSE ±2,100 GWh around the mean known demandOur estimate: RMSE ±1,800 GWh around the mean known demand±1,800Population-only prediction: RMSE ±2,100 GWh around the mean known demand±2,100Population-only prediction plus facilities: RMSE ±2,100 GWh around the mean known demand±2,100

Country scale

Not scored¹

Municipality scale (Local Authority District, median 360 km²).

Sub-region scale (ITL2 subregion, median 2,000 km²).

  1. Not scored at the country scale: the error would be zero by construction

Error by aggregation scale

Great Britain · Total

nRMSE (RMSE ÷ mean known demand)

Error by aggregation scale: Great Britain · Totalmunicipality scale (Local Authority District, median 360 km²): ours 28%, poponly 51%, poponly_plus_facilities 51%; sub-region scale (ITL2 subregion, median 2,000 km²): ours 26%, poponly 31%, poponly_plus_facilities 31%Population-only prediction plus facilities · municipality scale (Local Authority District, median 360 km²): nRMSE 51% of mean known demand; RMSE ±480 GWhPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion, median 2,000 km²): nRMSE 31% of mean known demand; RMSE ±2,100 GWhPopulation-only prediction · municipality scale (Local Authority District, median 360 km²): nRMSE 51% of mean known demand; RMSE ±490 GWhPopulation-only prediction · sub-region scale (ITL2 subregion, median 2,000 km²): nRMSE 31% of mean known demand; RMSE ±2,100 GWhOur estimate · municipality scale (Local Authority District, median 360 km²): nRMSE 28% of mean known demand; RMSE ±260 GWhOur estimate · sub-region scale (ITL2 subregion, median 2,000 km²): nRMSE 26% of mean known demand; RMSE ±1,800 GWh

Median area size, km² (log scale)

  • Not scored at the country scale: the error would be zero by construction

How error varies with area size and urbanization

Great Britain · Total

nRMSE (RMSE ÷ mean known demand)

By area size

By area size: Great Britain · TotalAreas per bin at the municipality scale: 9 · 9 · 9 · 9.Population-only prediction plus facilities · municipality scale, areas 3.1–170 km² (9 areas): nRMSE 73% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 170–390 km² (9 areas): nRMSE 19% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 390–1,000 km² (9 areas): nRMSE 14% of mean known demandPopulation-only prediction plus facilities · municipality scale, areas 1,000–3,000 km² (9 areas): nRMSE 13% of mean known demandPopulation-only prediction · municipality scale, areas 3.1–170 km² (9 areas): nRMSE 73% of mean known demandPopulation-only prediction · municipality scale, areas 170–390 km² (9 areas): nRMSE 21% of mean known demandPopulation-only prediction · municipality scale, areas 390–1,000 km² (9 areas): nRMSE 14% of mean known demandPopulation-only prediction · municipality scale, areas 1,000–3,000 km² (9 areas): nRMSE 14% of mean known demandOur estimate · municipality scale, areas 3.1–170 km² (9 areas): nRMSE 39% of mean known demandOur estimate · municipality scale, areas 170–390 km² (9 areas): nRMSE 14% of mean known demandOur estimate · municipality scale, areas 390–1,000 km² (9 areas): nRMSE 9.7% of mean known demandOur estimate · municipality scale, areas 1,000–3,000 km² (9 areas): nRMSE 7.5% of mean known demandmunicipality

Median area size, km² (log scale)

By urbanization

By urbanization: Great Britain · TotalAreas per class at the municipality scale: city centers 12 · rural areas 24.Our estimate · city centers, municipality scale (12 areas): nRMSE 27% of mean known demandPopulation-only prediction · city centers, municipality scale (12 areas): nRMSE 50% of mean known demandPopulation-only prediction plus facilities · city centers, municipality scale (12 areas): nRMSE 50% of mean known demandOur estimate · rural areas, municipality scale (24 areas): nRMSE 10% of mean known demandPopulation-only prediction · rural areas, municipality scale (24 areas): nRMSE 18% of mean known demandPopulation-only prediction plus facilities · rural areas, municipality scale (24 areas): nRMSE 16% of mean known demand

Urbanization class

Error falls as areas get bigger, and small areas are mostly urban while large ones are mostly rural, so read the two breakdowns together rather than as independent effects.

Scored as a country without local data: a model that never saw it

Great Britain · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to local totals one level up

Corrected to local totals one level up — Great Britain, Totalmunicipality scale (Local Authority District, median 270 km²): poponly_plus_facilities 29%, poponly 30%, ours 20%; sub-region scale (ITL2 subregion, median 3,900 km²): poponly_plus_facilities 19%, poponly 19%, ours 21%Population-only prediction plus facilities · municipality scale (Local Authority District, median 270 km²): nRMSE 29% of mean known demand; RMSE ±210 GWh (mean known demand 720 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 19% of mean known demand; RMSE ±1,000 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Authority District, median 270 km²): nRMSE 30% of mean known demand; RMSE ±210 GWh (mean known demand 720 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 19% of mean known demand; RMSE ±1,100 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District, median 270 km²): nRMSE 20% of mean known demand; RMSE ±140 GWh (mean known demand 720 GWh) — corrected to each sub-region's known total: what a user who knows the sub-region totals would getOur estimate · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 21% of mean known demand; RMSE ±1,200 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)

Corrected to the national total only

Corrected to the national total only — Great Britain, Totalmunicipality scale (Local Authority District, median 270 km²): poponly_plus_facilities 34%, poponly 35%, ours 31%; sub-region scale (ITL2 subregion, median 3,900 km²): poponly_plus_facilities 19%, poponly 19%, ours 21%Population-only prediction plus facilities · municipality scale (Local Authority District, median 270 km²): nRMSE 34% of mean known demand; RMSE ±240 GWh (mean known demand 720 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction plus facilities · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 19% of mean known demand; RMSE ±1,000 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Local Authority District, median 270 km²): nRMSE 35% of mean known demand; RMSE ±250 GWh (mean known demand 720 GWh) — corrected to the national total only: the situation of a country with no local consumption dataPopulation-only prediction · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 19% of mean known demand; RMSE ±1,100 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Local Authority District, median 270 km²): nRMSE 31% of mean known demand; RMSE ±230 GWh (mean known demand 720 GWh) — corrected to the national total only: the situation of a country with no local consumption dataOur estimate · sub-region scale (ITL2 subregion, median 3,900 km²): nRMSE 21% of mean known demand; RMSE ±1,200 GWh (mean known demand 5,600 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Modeled national demand against NESO metered consumption, 2024Experimental

Great Britain · average day and peak day of each season · hours in UTC (GMT, no summer-time shift)

Hourly profiles are experimental. Shapes come from sampled smart-meter and standard load profiles and have not been fully validated. Please use with caution and conduct your own validation before use.

Hour of day (UTC)

Sum of every pixel's hourly series against NESO's national consumption on the same days; modeled residential and non-residential series carry a weather correction. Annual totals: modeled gross 250 TWh, NESO 270 TWh (ratio 0.94). Counting large facilities by what they draw from the grid, which is what NESO can see, the ratio is 0.93.

The modeled series sums to the DESNZ meter-derived statistics the annual product is anchored to, plus the large sites connected straight to the transmission grid; the NESO series is national consumption reconstructed from transmission demand (which includes those directly connected sites), imports, a loss factor and estimated embedded wind and solar. The remaining gap between the two annual totals is a grid-level versus meter-level accounting difference (unmetered supplies, energy-industry own use and embedded generation sit on the NESO side); the scaled NESO series removes it so the two shapes can be compared directly.

The large-facility series is a standard industrial profile applied to each site’s annual estimate, not a measured shape.

  • MW (mean over the hour), UTC hour starts
  • days chosen on the modeled gross national series; NESO is shown on the same days

Guadeloupe · Scored only by a model that never saw the country or territory (a single split block)

Too few districts to hold any out: the whole territory is one block.

Scored as a country without local data: a model that never saw it

Guadeloupe · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Guadeloupe, Residentialneighborhood scale (IRIS, median 7.4 km²): poponly 31%, ours 23%; municipality scale (Commune, median 51 km²): poponly 22%, ours 15%Population-only prediction · neighborhood scale (IRIS, median 7.4 km²): nRMSE 31% of mean known demand; RMSE ±1.6 GWh (mean known demand 5.1 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 51 km²): nRMSE 22% of mean known demand; RMSE ±4.8 GWh (mean known demand 22 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 7.4 km²): nRMSE 23% of mean known demand; RMSE ±1.2 GWh (mean known demand 5.1 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 51 km²): nRMSE 15% of mean known demand; RMSE ±3.3 GWh (mean known demand 22 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Guadeloupe is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Guadeloupe · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Guadeloupe, Non-residentialmunicipality scale (Commune, median 51 km²): poponly 108%, ours 42%Population-only prediction · municipality scale (Commune, median 51 km²): nRMSE 110% of mean known demand; RMSE ±26 GWh (mean known demand 24 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 51 km²): nRMSE 42% of mean known demand; RMSE ±10 GWh (mean known demand 24 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Guadeloupe is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Guadeloupe · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Guadeloupe, Totalmunicipality scale (Commune, median 51 km²): poponly_plus_facilities 58%, poponly 58%, ours 22%Population-only prediction plus facilities · municipality scale (Commune, median 51 km²): nRMSE 58% of mean known demand; RMSE ±27 GWh (mean known demand 46 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 51 km²): nRMSE 58% of mean known demand; RMSE ±27 GWh (mean known demand 46 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 51 km²): nRMSE 22% of mean known demand; RMSE ±10 GWh (mean known demand 46 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Guadeloupe is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Martinique · Scored only by a model that never saw the country or territory (a single split block)

Too few districts to hold any out: the whole territory is one block.

Scored as a country without local data: a model that never saw it

Martinique · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Martinique, Residentialneighborhood scale (IRIS, median 4.7 km²): poponly 45%, ours 33%; municipality scale (Commune, median 29 km²): poponly 20%, ours 17%Population-only prediction · neighborhood scale (IRIS, median 4.7 km²): nRMSE 45% of mean known demand; RMSE ±1.9 GWh (mean known demand 4.3 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 29 km²): nRMSE 20% of mean known demand; RMSE ±3.5 GWh (mean known demand 17 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 4.7 km²): nRMSE 33% of mean known demand; RMSE ±1.4 GWh (mean known demand 4.3 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 29 km²): nRMSE 17% of mean known demand; RMSE ±2.9 GWh (mean known demand 17 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Martinique is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Martinique · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Martinique, Non-residentialmunicipality scale (Commune, median 29 km²): poponly 117%, ours 46%Population-only prediction · municipality scale (Commune, median 29 km²): nRMSE 120% of mean known demand; RMSE ±27 GWh (mean known demand 23 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 29 km²): nRMSE 46% of mean known demand; RMSE ±10 GWh (mean known demand 23 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Martinique is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Martinique · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Martinique, Totalmunicipality scale (Commune, median 29 km²): poponly_plus_facilities 73%, poponly 75%, ours 33%Population-only prediction plus facilities · municipality scale (Commune, median 29 km²): nRMSE 73% of mean known demand; RMSE ±30 GWh (mean known demand 41 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 29 km²): nRMSE 75% of mean known demand; RMSE ±31 GWh (mean known demand 41 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 29 km²): nRMSE 33% of mean known demand; RMSE ±13 GWh (mean known demand 41 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Martinique is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Mayotte · Scored only by a model that never saw the country or territory (a single split block)

Too few districts to hold any out: the whole territory is one block.

Scored as a country without local data: a model that never saw it

Mayotte · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Mayotte, Residentialneighborhood scale (IRIS, median 1.4 km²): poponly 50%, ours 35%; municipality scale (Commune, median 20 km²): poponly 27%, ours 18%Population-only prediction · neighborhood scale (IRIS, median 1.4 km²): nRMSE 50% of mean known demand; RMSE ±0.86 GWh (mean known demand 1.7 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 20 km²): nRMSE 27% of mean known demand; RMSE ±3.3 GWh (mean known demand 12 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 1.4 km²): nRMSE 35% of mean known demand; RMSE ±0.60 GWh (mean known demand 1.7 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 20 km²): nRMSE 18% of mean known demand; RMSE ±2.2 GWh (mean known demand 12 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Mayotte is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Mayotte · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Mayotte, Non-residentialmunicipality scale (Commune, median 20 km²): poponly 76%, ours 47%Population-only prediction · municipality scale (Commune, median 20 km²): nRMSE 76% of mean known demand; RMSE ±7.7 GWh (mean known demand 10 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 20 km²): nRMSE 47% of mean known demand; RMSE ±4.8 GWh (mean known demand 10 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Mayotte is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Mayotte · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Mayotte, Totalmunicipality scale (Commune, median 20 km²): poponly_plus_facilities 35%, poponly 35%, ours 17%Population-only prediction plus facilities · municipality scale (Commune, median 20 km²): nRMSE 35% of mean known demand; RMSE ±7.5 GWh (mean known demand 22 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 20 km²): nRMSE 35% of mean known demand; RMSE ±7.5 GWh (mean known demand 22 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 20 km²): nRMSE 17% of mean known demand; RMSE ±3.7 GWh (mean known demand 22 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Mayotte is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Réunion · Scored only by a model that never saw the country or territory (a single split block)

Too few districts to hold any out: the whole territory is one block.

Scored as a country without local data: a model that never saw it

Réunion · scored by a model that never saw the country or territory

Residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Réunion, Residentialneighborhood scale (IRIS, median 3.0 km²): poponly 43%, ours 31%; municipality scale (Commune, median 83 km²): poponly 11%, ours 17%Population-only prediction · neighborhood scale (IRIS, median 3.0 km²): nRMSE 43% of mean known demand; RMSE ±1.7 GWh (mean known demand 4.0 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 83 km²): nRMSE 11% of mean known demand; RMSE ±5.7 GWh (mean known demand 53 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · neighborhood scale (IRIS, median 3.0 km²): nRMSE 31% of mean known demand; RMSE ±1.2 GWh (mean known demand 4.0 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 83 km²): nRMSE 17% of mean known demand; RMSE ±9.0 GWh (mean known demand 53 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Réunion is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Réunion · scored by a model that never saw the country or territory

Non-residential

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Réunion, Non-residentialmunicipality scale (Commune, median 92 km²): poponly 53%, ours 69%Population-only prediction · municipality scale (Commune, median 92 km²): nRMSE 53% of mean known demand; RMSE ±35 GWh (mean known demand 65 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 92 km²): nRMSE 69% of mean known demand; RMSE ±45 GWh (mean known demand 65 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Réunion is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

Scored as a country without local data: a model that never saw it

Réunion · scored by a model that never saw the country or territory

Total

nRMSE (RMSE ÷ mean known demand)

Corrected to the national total only

Corrected to the national total only — Réunion, Totalmunicipality scale (Commune, median 92 km²): poponly_plus_facilities 30%, poponly 30%, ours 41%Population-only prediction plus facilities · municipality scale (Commune, median 92 km²): nRMSE 30% of mean known demand; RMSE ±36 GWh (mean known demand 120 GWh) — corrected to the national total (the level above this scale is the country here)Population-only prediction · municipality scale (Commune, median 92 km²): nRMSE 30% of mean known demand; RMSE ±36 GWh (mean known demand 120 GWh) — corrected to the national total (the level above this scale is the country here)Our estimate · municipality scale (Commune, median 92 km²): nRMSE 41% of mean known demand; RMSE ±48 GWh (mean known demand 120 GWh) — corrected to the national total (the level above this scale is the country here)

Aggregation scale

  1. Not scored at the sub-region scale: Réunion is a single split block, so the block scale is the country scale (national control only)
  2. Not scored at the country scale: the error would be zero by construction

Every area is scored by a model that never saw this country or territory: the accuracy a newly added country would get.

  • Great Britain, Canada, France (metropolitan) and Australia are scored two ways: corrected to the known totals one level up, and corrected to the national total only. Guadeloupe, Martinique, French Guiana, Réunion and Mayotte have too few districts to hold any out, so they are corrected to the national total only.

South Korea · No local consumption data to score against

The map is scaled to the official totals of each district, and nothing finer is published, so no local figure is left over to score it against. The model never trains on this country; the leave-one-country-out scores on this page are the closest guide to its accuracy inside a district.

External benchmark: JRC EU Energy Atlas

The Joint Research Centre's EU Energy Atlas is a similar public product at the annual, 1km level that we can validate against known consumption labels for any EU countries where we have training support (currently only France for our model). The atlas does not have sector-specific layers for electricity consumption, so this validation is against our total consumption estimates, not our residential model.

119 hidden EPCIs of metropolitan France · total electricity · 2021 labels · gross basis
ProductmdAPE (median absolute percentage error)RMSE (root-mean-square error)
Our estimate11.1%44.7 GWh
JRC atlas 2019, as published24.6%113.0 GWh
JRC atlas 2019, rescaled to the same total23.1%114.8 GWh