Alex Waithera11 min readUpdated

    Wind and Solar Data for Hydrogen Projects: ERA5, Global Wind Atlas and PVGIS

    Isometric illustration of wind turbines, solar panels and an electrolyzer under a floating weather-data grid.

    The best wind and solar data for hydrogen projects is an hourly series for the exact site, checked against a long-term local mean: ERA5 gives the hourly shape, the Global Wind Atlas gives a finer-scale wind mean, and PVGIS gives ready-made hourly solar output. No single free dataset does all three jobs, and the choice matters: in the worked example below, a wind speed series that is 10% too low raises the modelled cost of hydrogen by about $0.81/kg.

    Short answer: use ERA5 (or PVGIS, which is built on satellite and ERA5 data) when you need 8,760 hourly values for electrolyzer dispatch. Use the Global Wind Atlas to sense-check the long-term wind speed at hub height, because a 31 km reanalysis grid cell cannot see local hills and ridges. Treat all three as screening data, and replace them with site measurements before a feasibility study.

    This is general pre-feasibility guidance, not engineering or investment advice.

    The real project question

    An electrolyzer runs hour by hour, so its output depends on when the wind blows and the sun shines, not just how much over a year. An engineer sizing a plant therefore needs a full year of hourly generation, and a developer screening sites needs to know whether that year is realistic. Our green hydrogen site selection checklist puts "test usable power hour by hour" second on its list. This article covers where that hourly data comes from, and how much a wrong choice costs.

    Wind and solar data for hydrogen: what each free dataset is

    All facts below were checked on the publishers' pages on 16 November 2026.

    ERA5 Global Wind Atlas 3 PVGIS 5.3
    Publisher ECMWF, via the Copernicus Climate Data Store DTU Wind Energy, funded by ESMAP (World Bank) European Commission Joint Research Centre
    What it is Global atmospheric reanalysis Downscaled long-term wind climate Solar radiation and PV yield tool
    Time step Hourly Long-term means (by year, month and hour of day) Hourly series, plus typical-year files
    Spatial detail 0.25° grid; about 31 km 250 m microscale grid, from 3 km mesoscale runs Depends on the radiation database chosen
    Period 1940 to present, updated daily 10 years of mesoscale simulation 2005–2023 for SARAH-3 and ERA5 data
    Key fields 100 m wind components; surface solar radiation Wind at 10, 50, 100, 150 and 200 m Hourly irradiance and, optionally, hourly PV output

    ERA5. The Copernicus Climate Data Store lists ERA5 as hourly data from 1940 to present on a 0.25° grid, updated daily with a latency of about five days. The ECMWF ERA5 documentation lists 100 m wind components (u and v) and surface solar radiation downwards, and gives the underlying resolution as 31 km. Its strength is the hourly timing of wind and sun over decades. Its weakness for wind is detail: one grid cell averages ridges, valleys and coastline together.

    Global Wind Atlas. Version 3.0, released by DTU Wind Energy, downscales 3 km mesoscale runs to a 250 m grid and reports wind at five heights from 10 to 200 m, up to 200 km offshore. It is built for long-term means and includes an energy yield calculator, rather than offering an hourly time series to download. That makes it the natural cross-check for the mean wind speed in an ERA5 series.

    PVGIS. The PVGIS 5.3 release notes list CMSAF SARAH-3 satellite data and ECMWF ERA5 radiation for 2005–2023, with ERA5-Land and ERA5 for temperature and wind. Its API returns hourly radiation and, if you supply peak power and losses, hourly PV production. Unless you choose one, it picks a database by location from PVGIS-SARAH (Europe, Africa and Asia), PVGIS-NSRDB (the Americas between 60°N and 20°S) and PVGIS-ERA5. It also builds typical meteorological year (TMY) files from at least 10 years of data.

    Three traps when building an hourly profile

    1. The hourly shape and the annual mean come from different places. A common approach is to take ERA5's hourly shape and scale it to the Global Wind Atlas mean at hub height. That is reasonable for screening, but not automatically better. A multi-country validation preprint by Gruber and co-authors found ERA5 outperformed MERRA-2 against historical wind generation, saw no major improvement from bias-correcting with Global Wind Atlas version 2, and found that correcting with version 3 reduced simulation quality. Check any correction against nearby measured output where you can.
    2. Time stamps. ECMWF documents that ERA5 radiation is accumulated over the hour ending at the time stamp, in J/m², and in UTC. Shift and convert it before pairing it with wind or with a local power price, or the solar peak will land an hour or more off.
    3. One year is not the climate. A single weather year can be unusually windy or calm. Run several years, or at least test how sensitive your result is to the year chosen. A typical meteorological year smooths out extremes, which is useful for energy yield but hides the storage and curtailment problems that bad weeks create.

    Worked example: what a 10% wind data error costs

    The example uses a wind-only plant, because wind output is far more sensitive to resource error than solar. All values are illustrative assumptions, not quotes, project data or H2Hub outputs. The arithmetic was run in Python.

    Stated assumptions:

    Then we scaled every hourly wind speed up or down, as if the dataset had been too high or too low.

    Results: wind speed error

    Wind speed vs base Mean (m/s) Wind full-load hours Electrolyzer full-load hours Hydrogen (t/yr) Cost ($/kg)
    −10% 7.5 3,134 4,138 7,523 $7.15
    −5% 7.9 3,391 4,425 8,046 $6.70
    Base 8.3 3,640 4,699 8,543 $6.33
    +5% 8.7 3,873 4,949 8,998 $6.03
    +10% 9.1 4,084 5,169 9,398 $5.79

    In the base case the cost splits into $2.61/kg for power, $3.40/kg for electrolyzer capital and fixed operating cost, and $0.33/kg for stacks. A series 10% too low cuts hydrogen output by about 12% and raises cost by $0.81/kg, or about 13%. A series 10% too high flatters the project by $0.55/kg. Because power curves are steep between cut-in and rated speed, a small error in wind speed becomes a bigger error in energy.

    Results: same mean, different distribution

    Annual mean wind speed is not enough on its own. Keeping the 8.3 m/s mean and changing only the Weibull shape factor:

    Weibull shape Wind full-load hours Hydrogen (t/yr) Cost ($/kg)
    1.8 (more spread) 3,528 8,204 $6.58
    2.0 (base) 3,640 8,543 $6.33
    2.4 (steadier) 3,805 9,083 $5.98

    Two sites with the same average wind speed differ by about $0.60/kg here. That is why the dispatch model needs the hourly series, not just the long-term mean from an atlas. For the solar side the effect is gentler: PV output scales roughly in line with irradiance, so a few percent of irradiance error gives a few percent of output error. Our LCOH sensitivity analysis shows how this compares with other inputs.

    What the example leaves out

    The series is synthetic, so it ignores real diurnal and seasonal patterns, inter-annual variation and how wind lines up with sun at your site. It holds efficiency constant at part load and uses one generic turbine. It also fixes the renewable ratio: with a different resource, the cheapest oversizing changes too, as our renewables-to-electrolyzer ratio guide shows. And if you add solar, the mix question in solar vs wind for green hydrogen comes back into play.

    A practical data checklist

    1. Pull at least one full ERA5 year for the nearest grid cell, with 100 m wind and surface solar radiation, in UTC.
    2. Compare the ERA5 mean at hub height with the Global Wind Atlas mean for the actual turbine positions. If they differ by more than a few percent, run both cases.
    3. For solar, take PVGIS hourly PV output for your tilt, tracking and losses, from the database it selects for that location.
    4. Run more than one weather year, and report the spread, not just the base case.
    5. Before feasibility, plan a measurement campaign or buy a site-adjusted dataset. Free data is for screening.

    Try it in H2Hub

    You can start a project in H2Hub for free, without an account or a card. H2Hub states that its resource data comes from NASA POWER and ERA5. Use this example as your baseline: a 100 MW electrolyzer at $2,200/kW and 55 kWh/kg with wind power at about $36/MWh. Then enter your own site's location and compare the result with the table above.

    Your first full project and its PDF report are free; downloading the PDF needs a free sign-in. Re-running with new inputs or starting a second project requires a subscription, so screen your strongest site first. The Renewable Resource Lab is a subscriber tool for looking at resource inputs in more depth; signed-out visitors see an introduction.

    A note on pre-feasibility

    Free reanalysis, atlas and satellite data are screening proxies, not bankable resource assessments. The figures above are calculations from stated assumptions and a synthetic profile. They do not replace on-site measurement, an independent energy yield assessment or financial advice.

    Frequently asked questions

    Is ERA5 good enough for wind data for a hydrogen project?

    For screening and for the hourly shape, often yes: it is hourly, global and runs from 1940. Its 31 km grid can miss local terrain, so cross-check the mean with the Global Wind Atlas and, before investment, with site measurements.

    Does the Global Wind Atlas give hourly wind data?

    Its released features are long-term means at 250 m resolution, shown by year, month and hour of day, plus an energy yield calculator. For an 8,760-hour dispatch model you still need a time series such as ERA5.

    Can PVGIS give hourly solar data for an electrolyzer model?

    Yes. The PVGIS API returns hourly radiation and, with peak power and losses supplied, hourly PV output. Its radiation data in version 5.3 runs to 2023.

    How much does wind data error change the cost of hydrogen?

    In our illustrative 100 MW wind-only example, a 10% lower wind speed raised cost from $6.33/kg to $7.15/kg. The size of the effect depends on the turbine, the site and the plant design.

    Primary sources and further reading

    1. European Commission Joint Research Centre — PVGIS 5.3 release. Datasets and coverage; checked 16 November 2026.
    2. European Commission Joint Research Centre — PVGIS API non-interactive service. Hourly series and TMY tools, databases by region.
    3. Copernicus Climate Data Store — ERA5 hourly data on single levels from 1940 to present.
    4. ECMWF — ERA5: data documentation. Variables, resolution and the radiation accumulation convention.
    5. DTU Wind Energy — Global Wind Atlas 3.0 released, October 2019.
    6. Gruber, Regner, Wehrle, Zeyringer and Schmidt — Towards a global dynamic wind atlas, arXiv preprint, December 2020 (not peer-reviewed in this version).
    7. International Energy Agency — Global Hydrogen Review 2026: Assumptions Annex. Electrolysis cost and operating assumptions; onshore wind cost and full-load hours.
    8. U.S. Department of Energy — Technical Targets for Proton Exchange Membrane Electrolysis, page updated 29 May 2026.

    Editorial verification: 16 November 2026. Worked examples are calculations from stated assumptions, not forecasts, market prices or H2Hub-generated results.

    Sources

    Explore project tools

    Related articles

    Test your own project assumptions

    H2Hub models power, water and hydrogen costs together — every figure shows its source, date and confidence.

    Screen your site's resource in H2Hub — start free

    H2Hub produces preliminary project-development analysis. It is not a substitute for licensed engineering, FEED, due diligence or investment advice.