Solar Radiation Data

Our data is measured from approximately 5000 US ground stations and another 3500 international stations.  With few exceptions, our data has half the error of satellite and modeled data.  This difference is enough to make a solar plant appear profitable on paper when in fact it is not!  

The US daily data is collected every week from dozens of sources such as the USDA, National Forest Service, and universities.  This process takes between 40 and 50 hours.  After collecting the data, we convert all of the disparate formats into a single database with consistent units and perform multiple levels of quality control - including custom solar envelopes for each NOAA climate region and nearby station comparisons.  Finally, we generate the contour maps for our monthly atlas.  We have learned after two years that it is impossible to automate this process completely, so weekly maintenance is required to preserve the integrity of the data.
Data Accuracy
We believe we're the best and we're happy to back it up with real numbers that a layman can understand.  In most US locations, we have half the error of  our competitors.  Learn more.
We have very good coverage for the continental US as well as international stations in Europe, South America, Africa, Australia, and Asia.
View coverage maps.

Free Data
Free is data available from NREL and NASA. It's excellent for multi-year averages, but performs poorly for hourly and daily measurements.
Show me some free data.
Our Data Sources
We collect data from approximately 70 different networks, most of which are professionally maintained by state and government agencies. Here are a few of them.

Sample Data
Want to dive into the data?  We have contour maps and raw data files to look at here.
Data Pricing
Here is our latest price sheet for hourly and daily data extraction with quality control.

Data Accuracy

We recently consulted with a large, multi-national company who relied heavily on accurate solar radiation data for their agricultural research. They proudly described their real-time data feed with solar radiation data for every 5 km block of the US, updated every 15 minutes. We asked how the measurements were made. They didn’t know, but gave us the name of their supplier. We contacted the supplier. The supplier didn’t know how the measurements were made – they simply provided the surface modeling and gridding. We obtained the name of their supplier, and so on, only to find that there was not a single measurement of solar radiation anywhere in the chain. All of this mission-critical information on solar radiation was being estimated from cloud cover and humidity at airports.  Desktop software makes it easy to fit data with colorful contours and interpolate the surface down to very fine grids. It can give the user a false sense of precision and makes it easy to mask the key question:  How accurate is the data?

US Solar Radiation Datasets

Historical datasets of solar radiation are a key element in designing solar power systems and energy efficient buildings; however finding accurate multi-year data near the design site has always proved challenging. There are only 100-200 sites in the US providing research-quality observations of solar radiation, so this data is generally not available for engineering or architectural purposes.

Typical Meteorological Year

Perhaps the solar radiation dataset most widely used by US engineers and architects is the Typical Meteorological Year version 3 (TMY3) from the National Renewable Energy Laboratory (NREL). Each month in a TMY3 dataset contains historical observations. Twelve specific months from a 10-30 year history were selected as representative of the location and concatenated into a typical meteorological year. TMY3 data is intended only for relative comparisons of designs at one location or estimates of long-term solar radiation at a site, not for detailed engineering design or simulations (see user manual).

Satellite-Based Observations

The best known alternative to the TMY3 dataset is satellite-based observations of solar radiation. In fact, a significant portion of the NREL’s TMY3 incorporates SUNY (State University of New York) gridded satellite data. All satellite observations of solar radiation are modeled, since they must estimate ground radiation based on clouds and atmospheric conditions. The most widely used models in the US were developed by Perez, et al. Satellite datasets have the advantage of complete coverage, but the models have known inaccuracies due to persistent clouds, snow cover, and microclimates that can occur near mountains or large bodies of water. The National Solar Radiation Database provides free access to gridded satellite observations for 2000-2005 and recent SUNY satellite data is available from several commercial suppliers. NASA also provides free access to their satellite-based observations on their POWER (Prediction of World Energy Resource) website.

Medium-Quality Ground Stations

One significant but relatively untapped solar resource is data from ground-based stations equipped with medium quality solar sensors. In the US there are approximately 5000 of these sites from many different networks with daily, hourly, and sub-hourly observations for the past 5-25 years. The majority are professionally run and maintained by universities and government agencies for specific purposes such as agriculture, water management and environmental monitoring. Representative examples are the AgWeatherNet from Washington State University and the Oklahoma Mesonet. These networks overlap, so typically there are stations from three or four networks operating simultaneously in the same area. Most of these observations are available to the public via the internet for free or for a modest access fee. Wider use of this resource has been limited by a general lack of knowledge about the networks and how to access data. There have also been concerns about accuracy, quality control and difficulties in converting the data to a format usable for solar project simulations. One commercial supplier, the Solar Data Warehouse, has aggregated most of this data into a single database including quality control measures. The next section is a summary of a study comparing the observations from TMY3, NASA and SUNY datasets to this network of medium-quality stations.  It will show that they provide a significant and surprisingly accurate resource for solar radiation data.

Calculating the Accuracy of Solar Radiation Observations

The accuracy of a dataset is determined by comparing the observations to a highly accurate reference. Even small differences in location can affect the amount of solar radiation on the ground, especially for short time intervals, so comparisons should be done at exactly the same location. This can prove challenging for solar radiation observations where test sites are near, but not co-located, with the reference. In addition, extremely accurate reference measurements of solar radiation are not available. In 1989 the World Climate Research Program estimated that routine-operational ground solar radiation sites had end-to-end inaccuracies of 6-12%, with the highest quality research sites in the range of 3-6% inaccuracy (reference). These constraints make absolute comparisons between solar radiation datasets difficult, but it is still possible to estimate the relative accuracy if the same reference observations are used. For this study, research-quality observations from USCRN (US Climate Reference Network) were used as a reference. Most of these stations began operation between 2003 and 2005. The relative mean absolute error (rMAE) statistic was used to estimate the total error (bias plus precision) in the observations. None of the observations were co-located, so the total error includes effects due to physical separation, or in the case of satellite data, the grid size. The rMAE provides an easy way for practitioners to estimate how much error to expect in the data. The calculations only included comparisons where the observations of GHI (global horizontal irradiance) at the reference station were greater than 10 W/m2 so the statistics would not be skewed by low-light conditions.

Relative Accuracy of Various Solar Radiation Datasets

Seven of the earliest USCRN stations were operating in the southern half of the US during 2002-2005. Nearly 13,000 hourly measurements from 19 TMY3 months could be paired with those from USCRN sites at the same time and location. This overlap allowed direct comparison between the historical observations in various TMY3 months to high-quality ground observations. SUNY data, NASA data and observations from nearby medium-quality ground stations were also included in the comparison. Each of the medium-quality ground stations were within 20 miles and 500 feet elevation of the USCRN reference stations. The total errors (rMAE) in the observations of GHI are compared in Figure 5.

Figure 5 - Total errors in observations of global
horizontal irradiance from various sources.
Figure 6 - Bias errors in observations of global
horizontal irradiance from various sources.

The NASA observations had the highest daily total error (27%), TMY3 and SUNY had similar errors (19%) and the medium-quality ground measurements showed significantly lower errors (9%). The monthly total errors were similar for all datasets. The bias errors (rME) in the observations of GHI are shown in Figure 6. NASA observations had the highest daily bias error (16.7%), TMY3 and SUNY had similar daily bias (9.2% and 8.1%) and the medium-quality ground measurements showed significantly lower daily bias error (3.1%). The monthly bias errors were similar for all datasets. The similarity of errors between the TMY3 and SUNY datasets should not be surprising, since a significant portion of the TMY3 data comes from the SUNY gridded data in the National Solar Radiation Database. Leaving out TMY3, a more comprehensive comparison can be made between the NASA, SUNY and ground-based data. The next comparison used all of the data from 2002-2005 where there was overlap between the SUNY gridded data in the National Solar Radiation Database and seven USCRN stations in the southern half of the US (259 location- months of data). NASA satellite data for the locations was obtained from their POWER website. The Solar Data Warehouse provided corresponding ground-based measurements from one or more stations at each area. All ground stations were within 20 miles and 500 feet elevation of the USCRN reference stations. The same procedure was used to calculate the total and bias errors in the various datasets. Figures 7 and 8 show that daily observations from medium-quality ground stations had less than half the errors of the NASA and SUNY observations.

Figure 7 - Total errors in observations of global
horizontal irradiance from various sources.
Figure 8 - Bias errors in observations of global
horizontal irradiance from various sources.

These results are similar to other published comparisons. NASA estimates that their measurements of daily solar radiation have an RMS error of 35 W/m2 (roughly 20% total error see reference). Other researchers comparing NASA solar radiation data found 19% total error in the daily observations (reference).

Accuracy of the Reference Stations

Within the USCRN network there are several pairs of stations in close proximity. This provides an opportunity to see if the observation errors between two high-quality stations are due mainly to sensor accuracy or separation distance. This comparison used 2003-2005 data from the paired USCRN stations in Lincoln NE, Newton GA, Stillwater, OK and Asheville NC. The separation distances between these pairs were 18 miles, 6 miles, 1.5 miles and 6 miles respectively. The solar radiation sensors used by the USCRN stations are rated at less than 1% non-linearity and ±2% stability per year, however figure 9 shows much higher total and bias errors in the observations. The total and bias errors between two nearby high-quality stations were remarkably close to the total and bias errors between a medium-quality and a nearby high-quality station (Figure 9). This suggests that the errors in the ground-based observations may be more influenced by physical separation than by the difference between high-quality and medium-quality radiation sensors.

Figure 9 - Total and bias errors in global horizontal
irradiance observations from paired high-quality stations


Vignola, et al. suggest that there is a place for both satellite and ground-based measurements in forming a comprehensive solar radiation database for the entire US, with satellite data providing general coverage augmented by ground-based data for detailed engineering and scientific purposes. The findings of this study support this conclusion.

Our Technical Papers

The following papers were presented at the 2011 ASES conference in Raleigh:

Quality Analysis of Global Horizontal Irradiance Data from 3500 U.S. Ground-Based Weather Stations by James Hall
Links:  paper presentation
Note: We're up to 5000 stations now!

Forecasting Solar Radiation for the Los Angeles Basin - Phase II Report by James Hall
Links:  paper | presentation
Note: We have the most accurate published forecasting technology for the U.S. proven by out-of-sample tests.

2012 ASES conference:

An International Solar Irradiance Data Ingest System For Forecasting Solar Power And Agricultural Crop Yields by James Hall
Links:  paper presentation


International Stations
U.S. Stations        International Stations

Free Data

Do you think our data should be free to the general public?   Talk to NREL about it. Our data is better than theirs and costs a small fraction of what they're spending.  Anyway, here are some free data options:

NASA Agroclimate Data - It has high error compared to others, but is provided free as a public service.

NREL's TMY3 Data - This data can be very good for monthly averages but is terrible for hourly and daily data.  NREL says "The TMY should not be used to predict weather for a particular period of time, nor is it an appropriate basis for evaluating real-time energy production or efficiencies for building design applications or solar conversion systems." (TMY3 User Manual). 
NREL's SUNY Gridded Satellite Data (2000-2005) - This data is better than the TMY3 data but still not nearly as accurate as ground-based observations.  In fact, some of the TMY3 dataset comes from the SUNY satellite data.

We have a paper that explains the strengths and weaknesses of these data sources in detail.  It was presented at the 2011 ASES conference in Raleigh.  Check it out.

Ground Station Data Sources

Community Environmental Monitoring Program (funded by DOE)
Federal Aviation Administration (FAA)
National Oceanic and Atmospheric Administration (NOAA)
National Renewable Energy Laboratory (NREL)
National Weather Service (NWS)
Oklahoma Mesonet
Special projects funded by various state climate centers and universities
US Bureau of Land Management (BLM)
US Bureau of Reclamation (USBR)
US Department of Agriculture (USDA)
US Department of Commerce (DOC)
US Department of Defense (DOD)
US Department of Energy (DOE)
US Department of the Interior (DOI)
US Forestry Service
Other state and federal agencies

Sample Data


Sample Hourly Product - This is a 2 day sample of hourly data from several stations with a custom analysis of the sensor accuracy.  We believe it's a mistake to automate everything, so we like to get our hands in the data and make sure things look right. 

3 Station Comparison - This is an unformatted file with sample raw data from 3 stations near Palmdale, CA.

Single Station Daily Summary - This file contains raw daily data with averages for a station near Susanville, CA.

We're experienced database programmers, so let us know what kind of product we can generate for you!

Contour Maps

The first three maps were generated from 3600 sites.  We're now up to 5000.  For the latest data, check out the monthly atlas.

June 12, 2008
June 14, 2008
June 16, 2008

Monthly Atlas

Data Pricing

Here's our latest price sheet for data.  For solar forecasting information or custom quotes, please email:


Phillip Dorsett Jersey Mario Williams Jersey Jerick McKinnon Jersey Rob Gronkowski Jersey Max Unger Jersey Shane Vereen Jersey Nick Mangold Jersey Malcolm Smith Jersey