Data Guide

This guide describes the files of release 2026.10.4 and how to use them. For how the numbers are made, read the Methods page. Files are on the Downloads page or directly at

https://data.demandgrid.net/releases/2026.10.4/exports/<cc>/<file>

where <cc> is the country code: gb Great Britain (with Northern Ireland for 2021 to 2023), fr metropolitan France, gp Guadeloupe, mq Martinique, gf French Guiana, re Réunion, yt Mayotte, ca Canada, kr South Korea, au Australia. Years are 2021 to 2024 everywhere except Australia, which covers financial years 2017 to 2025.

The files

Each country has the same set. Sizes are per file: Great Britain’s pixel table is about 33 MB a year, France’s about 70 MB, Canada’s and Australia’s larger still.

Hourly profiles exist for Great Britain on the map only; they are not published as downloads yet.

Units, sectors and bases

Every consumption value is GWh per year per 1 km cell, which is numerically the same as GWh per km². Values are consumption at the customer’s meter: network losses are not included. year is the calendar year, except in Australia, where it is the financial year ending 30 June: 2025 means July 2024 to June 2025.

Three sectors, two bases, and one identity that holds in every cell:

elec_tot_gwh = elec_res_gwh + elec_non_res_gwh + elec_facility_gwh   (gross)
net_tot_gwh  = net_res_gwh  + net_non_res_gwh  + net_facility_gwh    (net)

elec_ columns are gross; net_ columns are net. Net is what the grid delivers to the premises. Gross is net plus the self-supply we can estimate: rooftop solar used on site where a rooftop-solar model exists, and a large site’s own generation where known. Residential is homes; non-residential is everything else that is not a large facility; large facilities are the individually identified sites of 10 GWh a year or more.

Which basis to use

Where a country has no rooftop-solar model (no_btm_model = true: France, its overseas territories and South Korea), gross equals net for homes and buildings. Canada’s model covers Ontario only: no_btm_model is false for the whole country, so check btm_covered in the region metadata file before using gross minus net there. Large facilities can differ on the two bases anywhere, by the electricity they generate for themselves.

The pixel table

One row per cell and year. GeoParquet 1.1 in longitude and latitude (EPSG:4326) with a covering bounding box, so a reader can fetch just the rows in an area.

Blank is not zero. The table holds only the cells inside a covered region (there are no rows for the sea). A value is blank (null) where the region has no official figure: Northern Ireland in 2024, and a few French communes outside any intercommunality in every year. A value of exactly zero means no modeled consumption; the map draws it as transparent.

Cell ids. pixel_id = row × 36,082 + column on the global 1 km Mollweide grid (ESRI:54009), with column = (x − (−18,041,000) − 500) / 1,000 and row = (9,021,000 − y − 500) / 1,000 for the cell center (x, y) in meters. You never need this to use the files, since the geometry column carries each cell, but it lets you compute ids from your own grid.

The rasters

The GeoTIFFs stay on the equal-area grid (ESRI:54009, Mollweide) at 1,000 m, float32, no-data as NaN. Each file holds four bands in this order:

1. residential
2. non-residential
3. large facilities
4. total (the sum of the first three)

The gross and net files have the same layout. Band values are GWh per cell, the same numbers as the pixel table to float32 precision. The ratio files carry the five per-person and per-built-area columns of one basis as float32 bands, in this order: total per person, residential per person, total per built m², residential per residential built m², non-residential per non-residential built m². Each band’s name metadata tag is the column name. No file mixes GWh with kWh per person or per m². Reproject your own layers onto the raster rather than the raster onto them, so each cell keeps its value and its identity.

The large-facility points

One row per facility and year, as points in longitude and latitude. The columns to read first:

Two rules before you add things up. Do not add the points to the raster: the raster already contains each site’s mapped value, spread over about 2 km around it within its region. And when you compare a site with a regional total, use the mapped column on the same basis as the total (mapped_elec_facility_gwh for gross, mapped_net_facility_gwh for net), which is what the total contains.

For a network model that wants sites as point loads: take elec_res_gwh + elec_non_res_gwh (or the net twins) from the pixels as the surface, and attach each site’s mapped_elec_facility_gwh (or mapped_net_facility_gwh) at its point. Per region and per country this adds up to the total columns.

The region metadata file

One row per official region and year: what a region’s total follows and how to cite it. A region is a local authority district in Great Britain and Northern Ireland (361 for 2021 to 2023, 350 in 2024), an intercommunality in metropolitan France (1,236, including four communes outside any intercommunality), a commune in the overseas territories, a Forward Sortation Area in Canada (1,634, the first three characters of the postal code), a district (sigungu) in South Korea (274 polygons sharing 229 published figures), and a local government area in Australia (544).

• region_id and region_desc name the official unit; region_uid is the join key to the pixel table.
• level_source and level_source_citation give the official statistics the region’s totals follow, with publisher, license and URL. shape_source and its citation name any finer official source used to shape the split within the region.
• res_anchor_basis, non_res_anchor_basis and basis_note say on what basis the official figure was published and any caveat about it.
• btm_covered is true where a rooftop-solar estimate was applied in that region and year.
• res_level_correction_factor is the factor by which finer official residential figures were scaled to match the coarser official total, where both exist (Canada, Australia, South Korea). Blank elsewhere.
• official_res_gwh and official_non_res_gwh are the published figures for the unit, with official_source and official_basis saying where they come from and on what basis; served_res_gwh and served_non_res_gwh are the map’s comparable sums (net residential; net non-residential plus the large facilities the figure counts, in_frame_facility_gwh); facility_outside_official_gwh is what the map adds on top, on the gross basis; official_relation is equal where the served sums reproduce the figure and shape where the figure only shapes the split of a coarser total. Where several regions share one published figure (Korean districts), official_unit_id names it: deduplicate on it before summing. Where a region has no published figure of its own (Canada outside Ontario, Australian council areas without billing), all of these columns are blank.
• The parquet version carries the region polygon; the CSV has the same columns without geometry.

Reconciling with the official figures you already use

Where your area is an official region, you do not need a spatial join: open the region metadata file (the CSV opens in a spreadsheet), filter to your region and year, and read the published figures, the map’s comparable sums and official_relation side by side; to get the map’s totals by sector and basis, sum the pixel rows that carry the region’s region_uid. What you will find, by country:

• Great Britain, France and the overseas territories (official_relation = equal). net_res_gwh summed over a local authority, intercommunality or commune equals the published residential figure. net_non_res_gwh plus the grid draw of the large facilities the statistics count (in_frame_facility_gwh) equals the published non-residential figure. What the map adds on top, on the gross basis, is in facility_outside_official_gwh: in Great Britain, transmission-connected steelworks and smelters and every site’s own generation; in France, own generation only. On the net basis, France’s totals equal the published figures and Great Britain’s exceed them only by the grid draw of the transmission-connected sites. On the gross basis, rooftop solar used on site is added to both sectors where it is modeled.
• Canada, Australia, South Korea (official_relation = shape). The published figure for a region shapes the split of a coarser official total (province, state, nation), so a region’s served sum differs from its figure by design: in the latest year the median gap is about 6% for Ontario postal areas and 5% to 9% for Australian council areas (residential to non-residential), while Korean districts reproduce KEPCO’s figures. The province, state and national sums follow the official figure on the basis it was published, to within a fraction of a percent. In Australia, as in Great Britain, net residential is the metered level and gross adds the rooftop-solar estimate; non-residential follows the Australian Energy Statistics as published.

The country metadata JSON

A small file worth opening once per country: bases, sectors and additivity in machine-readable form; facility_layer (the size threshold, classes present, how many sites have estimates and how many are scaled to fit, the yearly amount outside the official statistics in out_of_frame_gwh_by_year and label_gap_gwh_by_year, and the sources behind the estimates); official_statistics (one plain statement of which published figure the region file’s official columns carry and how the map relates to it); derived_metrics (the per-person and per-built-area columns, their denominators, thresholds and sources); level_source_citation; and license, citation, attribution and upstream_attributions, which lists the credit lines upstream sources require.

Adding pixels up to your own areas

One rule, used by the site and every recipe here: a cell counts once, in the polygon that contains its center. Cells are exactly 1 km² on the equal-area grid, so there is no area weighting to do. If your area is an official region, skip the join and sum by region_uid (previous section). Open the recipe for your tool:

import fsspec
import geopandas as gpd

RELEASE = "2026.10.4"
url = f"https://data.demandgrid.net/releases/{RELEASE}/exports/gb/gb_electricity_pixels_2024.parquet"

areas = gpd.read_file("my_areas.gpkg").to_crs("EPSG:4326")
pixels = gpd.read_parquet(
    url,
    bbox=tuple(areas.total_bounds),  # fetches only the row groups that overlap
    columns=["pixel_id", "region_uid", "elec_tot_gwh", "net_tot_gwh", "geometry"],
    filesystem=fsspec.filesystem("https"),
)
# each cell counts once, in the polygon that holds its center
centers = pixels.set_geometry(pixels.geometry.to_crs("ESRI:54009").centroid.to_crs("EPSG:4326"))
joined = gpd.sjoin(centers, areas, predicate="within")
totals = joined.groupby("area_name")[["elec_tot_gwh", "net_tot_gwh"]].sum()

library(sf)
library(sfarrow)
library(dplyr)

pixels <- st_read_parquet("gb_electricity_pixels_2024.parquet")
areas <- st_read("my_areas.gpkg") |> st_transform(st_crs(pixels))

centers <- pixels |>
  st_transform("ESRI:54009") |>
  st_centroid() |>
  st_transform(st_crs(pixels))
joined <- st_join(centers, areas, join = st_within, left = FALSE)
totals <- joined |>
  st_drop_geometry() |>
  group_by(area_name) |>
  summarise(gross_gwh = sum(elec_tot_gwh), net_gwh = sum(net_tot_gwh))

INSTALL spatial; LOAD spatial;
INSTALL httpfs;  LOAD httpfs;

CREATE TABLE areas AS SELECT * FROM ST_Read('my_areas.gpkg');  -- in longitude/latitude

-- the bounds of your areas, as literals (west, south, east, north), so only the overlapping row groups are read
WITH pixels AS (
  SELECT * FROM read_parquet('https://data.demandgrid.net/releases/2026.10.4/exports/gb/gb_electricity_pixels_2024.parquet')
  WHERE bbox.xmin <= -1.3 AND bbox.xmax >= -1.8 AND bbox.ymin <= 53.9 AND bbox.ymax >= 53.7
)
SELECT a.area_name,
       SUM(p.elec_tot_gwh) AS gross_gwh,
       SUM(p.net_tot_gwh)  AS net_gwh
FROM pixels p
JOIN areas a ON ST_Within(ST_Centroid(p.geometry), a.geom)
GROUP BY a.area_name;

Comparing places

Before you put two places side by side, especially across countries:

• Use the same basis on both sides.
• Check the year convention: calendar years, or financial years in Australia.
• Check the region metadata for whether each area’s total is an official figure or an allocated share of a coarser one (see “Reconciling” above), and training_tier for whether local data trained the model there.
• Check btm_covered, not the country-level no_btm_model, before comparing gross minus net.
• Use the residential columns, or set elec_facility_gwh aside, when one large site would swamp a comparison of ordinary demand.

Known issues in this release

Known in release 2026.10.4:

• Great Britain. Northern Ireland has no official figure for 2024 yet, so its cells are blank that year and its facility points stop at 2023.
• France, Martinique, South Korea. The region metadata file’s basis_note says “published gross == net”. That holds for residential and non-residential, which have no rooftop-solar model there, but not for large facilities, whose own generation makes gross higher than net.
• Australia. Sixty mines have estimate_status = no_estimate and add nothing to the raster.
• Australia. Where a shaping source starts partway through the series (Ausgrid residential figures from 2019), a council’s residential split can step that year; check shape_source by year before charting a trend.
• Everywhere. The facility points do not add up to the raster wherever a region’s sites are scaled to fit its official figure, and only the sites held below half their estimate carry scaled_to_fit; use the mapped columns for any comparison with the raster.

Anything found after publication is listed on the Changelog.

Versions, citing and license

Every release has an id such as 2026.10.4 and stays published unchanged at its own URL; the id is in the file path and in the metadata JSON, not inside the tables or rasters. Pin the id in anything you publish: a later release can change values everywhere, and the Changelog says what moved and why.

The data is licensed CC BY 4.0. Cite it as:

Add the date you accessed it where your style requires one, and use the release id in the URL of the files you actually downloaded.

Attribution also covers the sources named in each country’s metadata: the official statistics in level_source_citation (for Great Britain, DESNZ subnational electricity consumption statistics, Open Government Licence v3.0), Climate TRACE (CC BY 4.0) behind most facility estimates, and, where a facility value was sized from an OpenStreetMap power plant (Great Britain, France, Australia), the credit line © OpenStreetMap contributors. Each country’s upstream_attributions lists what applies.

← Back to Downloads