Solar TMY Data: Databases & Use in PV Design

Last updated: November 2026

Two weather files can produce different yield estimates even when the PV layout is unchanged. Before choosing the more favorable result, compare the weather sources, reference periods, timestamps and uncertainty assumptions. A finer output grid alone is not proof that one file better represents the site.

The TMY file you choose affects thermal modeling, production estimates and the financial case. Treat it as an engineering input with a documented source, not an invisible default. This guide explains the construction method, database choices and checks needed before importing a file into a PV simulation.

SurgePV publishes this post and sells solar design software.

How Solar TMY Data Is Built

A TMY is a synthetic representative weather year assembled from a longer record. It contains selected weather sequences rather than a smooth average for every hour. It is not an actual historical year or a forecast of next year.

The Sandia Method Behind TMY2 and TMY3

Wilcox and Marion’s 2008 Users Manual for TMY3 Data Sets, NREL/TP-581-43156 documents the historical TMY3 dataset and its interpretation. The report is available through OSTI; its historical laboratory name remains part of the report identifier.

Representative-month methods compare candidate months with the long-term distribution of weather variables. In outline, the workflow is:

1. Establish the historical period, available variables and data-quality rules.
2. Calculate the daily statistics used by the chosen method.
3. Compare each candidate month’s distribution with the long-term distribution for that calendar month.
4. Apply the method’s variable weights and selection criteria.
5. Select representative months and assemble the year.
6. Apply any prescribed boundary treatment and retain the selected-month metadata.

The Finkelstein-Schafer statistic is one way to quantify the difference between distributions. Selection weights and persistence criteria vary by method. Do not assume a provider uses the same procedure simply because its output is called TMY. Preserve the methodology and product version alongside the weather file.

ISO 15927-4 and Regional Variants

The European Commission’s PVGIS TMY documentation states that its described generator follows ISO 15927-4. It also identifies the source datasets and output metadata. That is a description of this implementation, not a claim that every TMY provider must use that standard.

Different variable weights can select different months from the same archive. Compare the resulting monthly irradiance and temperature profiles when changing methods. If the selection changes the simulated yield, explain the actual difference rather than applying a generic percentage adjustment.

Why 8,760 Hours?

A non-leap year has 365 × 24 = 8,760 hours. Hourly TMY files commonly use that structure, but check the file’s time step and leap-day treatment rather than assuming its row count. Sub-hourly products require a compatible simulation workflow.

The timestamp convention matters as much as the row count. Determine whether times mark the start, center or end of an interval and whether they use UTC or local standard time. PVGIS documents an irradiance time offset in its output header; apply the source’s guidance when calculating solar position.

A file can have the expected number of rows and still be misaligned by an hour. Plot irradiance against solar position and compare the first and last records before accepting an import. Avoid manually shifting timestamps until the source convention and the simulator’s expected convention are both clear.

What a Solar TMY File Contains

Not all TMY files contain the same parameters. Understanding which fields your simulation tool actually uses helps you choose the right source and avoid missing inputs.

The table lists possible weather and environmental inputs, not a mandatory set present in every export. Its parameter categories follow the Solargis documentation linked below. Check the units in the actual file: irradiance in W/m² and interval irradiation in Wh/m² are different quantities.

GHI, DNI and DHI describe the total horizontal, direct-normal and diffuse-horizontal components. For consistent simultaneous quantities, GHI = DNI × cos(zenith) + DHI. If GHI and DNI are already supplied, DHI can be obtained from that relationship subject to timestamp and quality checks; a decomposition model is needed when the available inputs do not specify the components. Read more about horizontal irradiance before converting it to the module plane.

Temperature and wind speed influence modeled module temperature. Use the selected module’s power temperature coefficient and the appropriate thermal model rather than a generic loss per degree. Ambient temperature is not the same as cell temperature.

Albedo describes ground reflectivity. If it is absent from the weather file, document the separate surface assumptions used for bifacial modeling. Ground cover, snow, row geometry and shading can change rear-side irradiance; a generic albedo number cannot establish the project’s bifacial gain.

Free vs. Commercial TMY Sources in 2026

Choose a data product by coverage, methodology, validation and project requirements. Price and grid spacing alone cannot determine whether a dataset is suitable for financing.

Sources: the linked provider documentation in the following sections, accessed November 2026. The PVGIS row describes version 5.3 specifically; inspect the version of any new download rather than assuming those dates cover every PVGIS product.

NSRDB

The National Laboratory of the Rockies GOES TMY API documentation specifies 4 km hourly data for this product. It is one NSRDB product, not a universal specification for the entire database. Historical TMY3 station files and current satellite-derived TMY products should not be treated as interchangeable versions of the same file.

Record the requested location, returned location and dataset identifier. Preserve the header so a later reviewer can identify the source and time convention. If a simulation interface downloads data automatically, inspect the resulting metadata before citing the database in a proposal.

PVGIS

The JRC PVGIS 5 user manual lists SARAH-3 at approximately 5 km spatial resolution, not 250 m. The selected radiation database depends on location and product version. Weather variables can also come from a different underlying source than the irradiance data.

Keep the weather-grid resolution separate from terrain or horizon resolution. A detailed terrain layer does not turn a coarser cloud field into equally detailed irradiance measurements. Check which horizon treatment is already included to avoid counting the same obstruction twice in a separate model.

Solargis and Solcast

Solargis describes time-series, TMY and probabilistic products with distinct uses. Its documentation explains representative-month selection and low-irradiation scenarios. Ask which uncertainty components belong to the delivered product before using it in a plant-level assessment.

Solcast describes its TMY offering as historical weather selected to match long-term conditions and provides several export and integration options. Its page distinguishes cloud-tracking resolution from downscaled parameters. Do not present the finest advertised output spacing as the native resolution of every input variable.

Neither provider name guarantees lender acceptance. Agree on the resource assessment and validation evidence with the independent engineer. A commercial license can provide useful data and support without eliminating site-specific uncertainty.

NASA POWER and Other Sources

NASA POWER’s methodology describes global solar and meteorological data drawn from satellite and model products. Check the source and resolution for each requested variable instead of assigning one grid size to the entire service.

Meteonorm and CAMS may also be considered during source selection, but their exact product specifications should be taken from the version being procured. Obtain a sample file and its documentation before comparing it with the options above. A provider comparison is useful only when the time basis, location and output variables are aligned.

TMY vs. Time Series vs. Long-Term Average

These three data products are often confused. Choosing the wrong one is a common source of error in solar analysis.

TMY data is a single synthetic year. It supports representative-year energy simulations and design comparisons. Its acceptability in a financing assessment depends on the complete methodology and evidence. It captures typical diurnal and seasonal patterns without requiring the user to process decades of data.

Time-series data preserves the full historical record. It is essential for understanding interannual variability, detecting trends, correlating production with energy prices, and supplying historical context for forecasting methods. If you need to know how often a site experiences an unusually cloudy year, use time series, not TMY.

Long-term averages (LTA) reduce everything to smooth monthly or daily mean curves. They are useful for site screening and high-level resource mapping, but they erase real weather variability. A simulation based on averages cannot recover the original weather-event sequence. Review how any synthetic profile represents variability before using it to assess clipping or storage operation.

The choice depends on the question:

P50, P90, and Probabilistic TMY

A representative weather file and a project’s exceedance-probability estimate are related but different. Weather selection alone does not account for all uncertainty in a PV plant’s delivered energy.

P50: The Central Estimate

P50 is a modeled median: the stated production has a 50% probability of being exceeded under the model. A representative TMY simulation may be used as the central estimate, but its suitability depends on the file’s construction and the project assumptions. Do not call every TMY result a validated P50 simply because the file is typical.

P90: The Downside Assessment

P90 means the stated quantity has a modeled 90% probability of exceedance for the specified period. A provider’s low-irradiation weather scenario is not automatically a plant P90. The project assessment must also consider modeling, equipment, loss assumptions and other relevant uncertainties without double counting them.

PVsyst’s probability documentation explains that the user must supply uncertainty parameters in addition to the simulation. Its described method assumes a normal distribution and combines uncertainty components in quadrature. That is a modeling choice requiring justified inputs, not a guarantee of future output.

Specify whether the assessment concerns one year or a multi-year average. Weather variability and systematic resource uncertainty do not necessarily diminish in the same way over longer periods. Agree on the period and method with the independent engineer before transferring the result into a financing model.

Beyond P90

Other exceedance levels, such as P75 or P99, describe different points on the modeled distribution. A more extreme percentile is especially sensitive to distribution assumptions and omitted risks. Avoid applying a universal percentage gap between P50 and P90 across climates or projects.

Record the central yield, uncertainty components, correlation assumptions and calculation method. If using a probabilistic weather file, identify what it already accounts for. Our probability-of-exceedance guide explains the distinction between a modeled probability and a production guarantee.

Accuracy, Uncertainty, and Common Traps

TMY data is a model of reality, not reality itself. Understanding its limits prevents costly design mistakes.

Read the Validation Basis

A useful validation report states the locations, reference instruments, comparison period and error metrics. Annual bias, hourly error and uncertainty in a long-term mean are different quantities. Do not apply one published percentage to every climate or turn an irradiance error directly into an identical energy-yield error.

Assess whether the validation sites resemble the project’s terrain, cloud regime and aerosol conditions. Record any site-adaptation method and the quality of the measurements used. The uncertainty budget should explain why its values are appropriate for this project rather than repeating a generic database claim.

The Microclimate Problem

A TMY file represents the average conditions within its spatial grid cell. A 4 km grid cell may include mountains, coastline, urban heat islands, and agricultural land. A site 2 km inland from a foggy coast can have meaningfully different irradiance than the cell average.

For high-value projects, consider a ground-measurement campaign and long-term correlation designed by the resource specialist. Duration, instrument maintenance, seasonal coverage and the reference dataset affect its usefulness. There is no guaranteed percentage reduction in uncertainty from a short measurement campaign.

Soiling Is Not in the TMY File

TMY files contain weather. They do not contain dust, pollen, bird droppings, or snow accumulation. Soiling losses must be added separately based on local experience, satellite-derived soiling models, or measured data. Document the assumed loss, seasonal pattern and cleaning plan separately. Weather data can inform that assessment without directly measuring the condition of the modules.

Climate Trend Risk

TMY datasets are built from historical records. They do not automatically account for shifting climate patterns. Rising temperatures, changing cloud regimes, and increasing aerosol loading can make a TMY based on 1990-2010 data less representative of 2026-2050 conditions. Some providers now offer trend-adjusted or recent-period TMY products to address this.

What Most Designers Get Wrong

The most common TMY mistakes are:

• Using a single global database for all projects regardless of region
• Ignoring spatial resolution and assuming the nearest grid point is representative
• Confusing GHI with plane-of-array irradiance
• Assuming GHI and DNI share the same validation error or uncertainty
• Applying P50 production estimates in financial models without a downside case

TMY Data in the Solar Design Workflow

A professional workflow treats TMY selection as a documented engineering step. Preserve enough information for another reviewer to reproduce the import and explain its limitations.

Step 1: Define the Project Requirements

Define the decision the model supports, the expected review standard and the uncertainty assessment required. A free dataset may be suitable when supported by evidence; a commercial dataset still needs review. Agree on measurement and validation needs before procuring data.

Step 2: Select the TMY Source

Compare available sources for the location and intended assessment. Record coverage, reference period, data version, variable definitions and quality flags. Keep the original file unchanged and save any conversion as a separate documented step.

Step 3: Validate the File Against Local Knowledge

Check the annual GHI and monthly profile against known local conditions. If the file shows a cloudy July for a site famous for summer drought, investigate. Cross-check with a second independent source when possible.

Step 4: Run the Energy Simulation

Import the TMY into a simulation tool that supports its format. The simulator will transpose GHI/DNI/DHI to plane-of-array irradiance, apply temperature and electrical models, and calculate hourly AC output. Make sure the tool models shading accurately; site and horizon shading need separate attention, including for open-field arrays.

Step 5: Apply Probabilistic Analysis

For projects requiring exceedance estimates, have the responsible analyst prepare the agreed uncertainty assessment. Identify the period and probability case in both technical reports and commercial documents. Document the weather source and vintage without implying that the TMY file alone supplies a complete downside analysis.

Step 6: Update After Commissioning

Once the system is operating, use actual-period weather and operating records to diagnose performance. A difference from a TMY forecast can reflect unusual weather as well as equipment or modeling issues. Update assumptions only after separating those causes and retaining the evidence.

How SurgePV Supports the Yield-Modeling Workflow

SurgePV supports roof modeling, array layout, shading analysis, energy yield, financial modeling and proposals. Those design steps need traceable inputs. The product scope does not justify assuming automatic weather-source selection, native P90 calculation or warnings about every dataset limitation.

Before committing to a workflow, ask which weather sources, file formats and outputs are supported for the project. Use a sample project to establish how source metadata and assumptions are retained. If a separate resource assessment is required, name its owner and specify the handoff into the yield and financial models.

Keep a concise assumptions register with the dataset version, historical period, location, horizon treatment, temperature model and loss assumptions. Link it to the exact layout revision. A later equipment or geometry change should not be confused with a weather-data change when comparing results.

For a practical walkthrough, book a SurgePV demo and bring a weather-data checklist alongside the proposed roof design. The useful outcome is a clear record of the supported workflow and any specialist analysis that must be completed separately.

Key Takeaways for TMY Selection

A useful TMY file matches the question, location and simulation workflow. It is one part of a credible yield assessment, alongside equipment definitions, shading, losses and uncertainty analysis.

• Match the data product to the project’s requirements and review its validation evidence.
• Distinguish representative weather from actual historical sequences and operational forecasts.
• Keep resolution, accuracy and uncertainty separate when comparing sources.
• Preserve timestamps, units, version and selected-month metadata through the import.
• Treat P50/P90 analysis as a documented project assessment rather than a weather-file label.
• Use actual-period weather to interpret operating performance after commissioning.

A clear record of these choices makes the next design review easier. It also prevents a change in dataset from appearing as an unexplained improvement in the proposal’s production estimate.

Frequently Asked Questions

What is solar TMY data?

Solar TMY data is a synthetic weather year selected to represent longer-term conditions at a location. It supplies time-stepped inputs for a PV simulation. It is neither an actual historical year nor a forecast for next year.

How many hours are in a TMY file?

A conventional hourly non-leap TMY contains 8,760 records. Check the product’s time step, leap-day handling and timestamp convention before importing it. Other temporal resolutions require a compatible simulation workflow.

Is PVGIS SARAH-3 weather data a 250 m grid?

The cited PVGIS 5 user manual lists SARAH-3 at approximately 5 km spatial resolution. Do not confuse weather-grid spacing with terrain or other downscaled layers. Record the specific PVGIS version and dataset used.

Does a TMY file provide a project P90 estimate?

A TMY file alone does not provide a complete project P90 estimate. A probabilistic weather product may address resource conditions, while plant yield also depends on other uncertainties. State the assessment period and avoid double counting uncertainty.

Which TMY source should I choose for solar design?

Choose a source whose coverage, validation, reference period and format suit the project. NSRDB, PVGIS and commercial products have different specifications and workflows. Agree on the evidence required by the project reviewer rather than choosing by price alone.

Can TMY data forecast tomorrow’s solar production?

TMY data cannot forecast tomorrow’s weather or solar output. Use an operational forecast with appropriate weather inputs for that task. TMY is intended for representative-condition simulations.

How should I compare TMY predictions with operating data?

Use actual-period weather and plant operating records to investigate differences from a TMY prediction. Unusual weather, outages, soiling and model assumptions can all affect the comparison. A raw difference from a typical-year estimate is not a diagnosis.