OptimalSolution/databank-ember

Ember electricity data

Global electricity generation, capacity, demand and emissions from Ember's public bulk releases, reshaped onto one long schema and stored as plain parquet. Area codes, labels and the geoscale that joins them are in scales/.

Terms

• Licence: CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/)
• Attribution: Data: Ember - CC-BY-4.0
• Cite as: Yearly and monthly electricity data, Ember. https://ember-energy.org/data/
• Source: https://ember-energy.org/data/

Redistributed by OptimalSolution LLC -- see DISCLAIMER.md. Redistribution only: no responsibility for the content, no endorsement or position, no affiliation with the publisher, no warranty. What was changed from the published data is recorded per dataset under bank.modifications and bank.changes.

Datasets

dataset	rows	coverage
ember_generation_yearly	543,302	1985 to 2025, annual
ember_generation_monthly	838,831	1998-12 to 2026-08, monthly
ember_capacity_wind_solar_monthly	52,910	2016-01 to 2026-08, monthly
ember_price_monthly	4,004	2015-01 to 2026-09, month
ember_price_daily	121,418	2015-01-01 to 2026-09-17, day
ember_price_hourly	2,924,722	2015-01-01 to 2026-09-17, hour
ember_targets_2030	2,064	2022 to 2050, year
ember_targets_sources	223	reference table
ember_cleantech_exports_monthly	293,804	2018-01 to 2026-07, month
ember_cleantech_exports_by_code_monthly	1,183,132	2018-01 to 2026-06, month
ember_india_yearly	17,055	2019 to 2024, year
ember_india_monthly	234,687	2019-01 to 2025-11, month
ember_us_states_yearly	86,553	2001 to 2025, year
ember_us_states_monthly	1,052,031	2001-01 to 2026-06, month
ember_interconnection_ntc	1,702	2024 to 2040
ember_interconnection_flows	22,500	scenarios 2024/2030/2040
ember_interconnection_country_profile	8,648	scenarios 2024/2030/2040
ember_interconnection_import_potential	368	target years 2030/2040
ember_interconnection_peak_demand	134	2024 to 2040

Read this before you aggregate

• is_derived marks Ember's OWN aggregate series -- Total generation, Clean, Fossil, Renewables, Demand. They sit in the same series column as the fuels they summarise, so summing the column without filtering counts the fuels twice.
• entity_kind separates the 209 countries from Ember's 15 precomputed aggregates, and those aggregates OVERLAP -- every EU member is also in OECD. Summing entity_id double-counts differently again.
• unit is part of the key. Solar capacity is published twice, once in GWAC and once in GWDC, so (entity, period, series, variable) is NOT unique. The generation tables also mix TWh with percent in one value column.
• AU and MENA exist only as precomputed rows: Ember publishes no membership for them, so they cannot be broken down into countries. scales/ember_geoscale.json lists them under unmappable.
• Geometry is NOT here. It is Natural Earth, public domain, shared with every other source and referenced as basemaps:ne_countries_50m. Without it the geoscale still aggregates and recasts; only maps need it.
• The sub-national tables carry the national total too. India and the US both ship their country row alongside their states, coded with the ISO3 and marked entity_kind = "country". Summing every row double-counts the country exactly once. India's OT is Ember's 'Others' residual, not a territory: it has no geometry and never will.
• The China cleantech tables are keyed on the DESTINATION. Every row is an export from China to entity_id; it says nothing about that country's own production. exports_12m_rolling is a rolling twelve-month sum of the column beside it, so summing it over periods counts each month twelve times -- it is flagged is_derived. The by-code table is the same trade one level finer than the by-category table; stacking them doubles.
• The interconnection Hour and Month columns are not clocks. Ember's own README calls them hour-of-day and month-of-year averages, so period_index runs 1-24 or 1-12 and describes a PROFILE, not a timeline. The tables are also scenario output, not observation: 2024 is a modelled reference and 2030/2040 are projections under Reference, Projects and Needs.
• Three interconnection zones are not countries. NW, BW and CW are the North Sea, Baltic and Celtic offshore wind hubs. scales/ember_interconnection_zones.csv marks them offshore_hub and leaves their ISO3 empty. Ember also writes UK where ISO 3166-1 writes GB.
• ember_targets_2030 is not only about 2030. The tracker is named for it, but TARGET_YEAR in the underlying data ranges from 2022 to 2050. Filter on year rather than assuming. The targets are also stated ambitions, not outcomes, and source_id joins to ember_targets_sources for the document each one came from.
• Prices are wholesale day-ahead, not retail. They exclude taxes, levies and network charges. The hourly table carries Ember's local wall-clock reading as text beside the UTC timestamp, because a local time with no offset cannot be parsed without inventing a timezone.

Using it

# the tables (plain parquet -- no special reader needed)
library(arrow)
d <- read_parquet("datasets/ember_generation_yearly/data/part-0.parquet")

# the geoscale, rebuilt from the shipped CSV + JSON
source("scales/ember_geoscale.R")
gs <- ember_geoscale()                       # no geometry
gs <- ember_geoscale("../naturalearth")     # with geometry

# split a regional aggregate down to countries, conserving the total
geoscales::recast_geoscale(x, gs, from = "ember_region", to = "region",
                           values = "value", rule = "sum",
                           weight = "pop_est")