Renewables.ninja Wind & Solar PV Simulation Data

It is modeled physics, not a metered feed - and that difference decides what you can do with it. Renewables.ninja was built by researchers affiliated with ETH Zurich and Imperial College London to simulate hourly power output from wind farms and solar PV installations anywhere on the planet. Two published models do the work: the GSEE Python package for photovoltaics and the virtual wind farm (vwf) package in R for wind, both fed by weather data rather than by utility disclosures - MERRA-2 reanalysis on a 0.625 x 0.5 degree grid, with SARAH satellite irradiance on a finer 0.05 x 0.05 degree grid driving the PV model.

The output lands in two shapes. A point simulation answers "what would this exact location produce," hour by hour, in kW. The country datasets answer the same question for whole national fleets: files such as ninja_pv_country_DE_merra-2_corrected carry hourly capacity factors from 0 to 1 reaching back to 1985, bias-corrected against national generation statistics, with the correction status written straight into the filename so nobody mistakes raw model output for calibrated history. Within Datadory's catalog of 1,744 datasets across 159 viable industries, this slice scores 8/10 on our quality rubric.

Two delivery shapes, one grain - the hour. The site shape reports mean power output in kW per hour for the coordinate and technology you scoped. The country shape reports the same hourly rhythm as a fraction of installed national capacity, which is what makes a German PV winter afternoon directly comparable with a Spanish one:

time=2016-01-01T00:00Z  electricity_kW=<kW-scale hourly output>
  model=pv  weather-driver=SARAH

file=ninja_pv_country_DE_merra-2_corrected.csv
  columns=time,national_pv_capacity_factor
  span=1985-2016 hourly  values=0..1
  correction=bias-corrected vs national generation statistics

Both shapes open with leading # comment lines - two in country files, three in point files - carrying units, parameters and provenance, so every file explains itself before the first data row. A single year runs to roughly 8,760 hourly rows, and the documented German PV example spans 1985 through 2016: thirty-plus weather years in one file, enough to catch calm dark winters that a single reference year would miss.

Five verified core fields, written against the published model documentation rather than guessed. time opens every row - a UTC timestamp marking the start of each hour-long period, and the join key against whatever fleet register you bolt on. electricity carries mean power output in kW for site simulations; capacity_factor takes over for country files and expresses the same hour on a 0-to-1 scale, bias-corrected where the filename says so. An optional local_time column swaps the timezone when a request asks for it, and every file leads with # metadata header lines recording units and parameters.

Request-side knobs travel with the data too: turbine model (the Vestas V80 2000 is the documented example), hub height, tilt, azimuth, tracking mode, system loss and time averaging. Those ride along as parameters folded under "additional fields on request" rather than pretending to be independent columns.

Geography first: point simulation works for any coordinate on earth, because the underlying weather grids are global - MERRA-2 at 0.625 x 0.5 degrees for wind, SARAH at 0.05 x 0.05 degrees for PV. Ready-made country aggregation currently centers on Europe: EU-28 plus Norway and Switzerland, distributed as national hourly series.

Time is the quiet superpower here. Country series start in 1985, and the documented German example runs through 2016 with archived releases extending to 2019 - multiple decades of weather years rather than one cherry-picked reference year. Granularity is fixed at hourly, per site or per country, about 8,760 rows per year; anyone who has tried to downsample a monthly national statistic into an hourly profile knows exactly how much grief that resolution saves.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the channel your stack already speaks. The rows arrive identical either way - one row per hour with its UTC timestamp and either a kW output figure or a 0-to-1 capacity factor, plus the parameter context that produced it - normalized so nothing in your pipeline re-runs a physical model to make the numbers comparable.

Every delivery ships with the field dictionary above plus these sample rows for validation, so your first join attempt happens against evidence, not hope.

• Power-system modeling - hourly national capacity factors from 1985 onward give dispatch and adequacy models a weather-year spread instead of a single average year; see market sizing.
• ML feature engineering - a clean hourly panel with a stable timestamp key trains forecasting and nowcasting models without any scraping step; see ml model training.
• Asset screening - simulated output for a proposed site puts a production curve behind a development thesis before a met mast is ever financed.
• Citation-grade research - peer-reviewed model lineage (GSEE, vwf) and bias-corrected country calibration survive academic scrutiny; see citation-grade research.
• Portfolio comparability - because every country series shares the 0-to-1 scale and hourly grain, cross-border capacity-value comparisons become a groupby, not a data-engineering project.

Data scientists get a typed, regular hourly panel - timestamp, numeric target, no gaps to paper over - that drops straight into feature pipelines. Investors and quants use the 0-to-1 country factors to stress revenue models for renewable portfolios across decades of weather rather than one marketing-friendly year. Market researchers and consultants benchmark national output behavior across the EU-28 plus Norway and Switzerland without stitching twenty statistical offices together. Journalists and academics cite a model with named institutional provenance instead of a spreadsheet of uncertain origin. Developers and builders scope candidate sites with a defensible first-pass production curve before committing to measurement campaigns.