Creating more accurate solar datasets for Americas using GOES-R satellite data

The launch of GOES-R (operated by NASA and NOAA) has given solar modelers a great opportunity to use higher-quality satellite imagery. As an outcome, they can now achieve higher-quality solar irradiance data.

In this article, we are sharing some of the challenges that the integration of GOES-R meant for our satellite data team and how the model was adapted and improved.

Improving the solar irradiance model thanks to GOES-R

Since 2018, after reaching the geostationary orbit and location it was planned for (0º latitude and 75.2°W longitude), GOES-R satellite started providing stable data inputs for solar irradiance models.

A set of new instruments using the most advanced technology is installed on the GOES-R satellite. Of especially high importance for solar irradiance modelling is its Advanced Baseline Imager (ABI), capable of viewing the Earth across 16 spectral bands, including two visible channels, four near-infrared channels, and ten infrared channels.

This new ABI provides three times more spectral information, four times the spatial resolution, and more than five times faster coverage than the previous satellite technology.

After the first GOES-R data integration we did a few years ago, our satellite data team has been working on additional improvements across the solar irradiance model used for sites covered by this satellite. Having a longer period of data has helped us better understand some of the limitations of the original model and address them in the new methodology.

All this work has been materialized in the new Solargis model version 2.2.35 which is now available for any data delivery in the Americas (and Eastern Pacific).

The improvements cover such areas as:

• Satellite data pre-processing
• Terrain corrections
• Detection of clouds & snow
• Sun position

Satellite data pre-processing

The integration of inputs from a new satellite means adapting the whole data processing chain, starting from the beginning.

Raw satellite data requires processing from the moment we receive it. Particularly for GOES-R data, our team implemented a more advanced pre-processing algorithm. This helped overcome spatial “noise” and achieve smoother transitions after data re-projecting.

Terrain corrections

Geostationary satellites orbit the Earth at “fixed” position, approximately 36,000 km above the ground while getting information for a remarkably large area (entire hemisphere, or Earth disk) every ten minutes. However, areas further away from the center of the disk (e.g. northern latitudes) are “seen” by the satellite at high viewing angle, causing distortion of the signal, reduced spatial detail and in the case of mountainous regions, positional displacement as well.

To tackle this issue, our satellite data team implemented a reprojection algorithm that is able to correct terrain distortions and provide better positional accuracy. See the image below.

Clouds and snow detection

The irradiation model relies mainly on the visible channel of the satellite instrument to characterize the clouds.

Since clouds and snow are both bright in the visible spectrum, one of the biggest challenges for the algorithm is to successfully distinguish these two types of features and properly estimate cloud properties. We used auxiliary information from infrared channels and other data sources to improve the estimates in snowy conditions. Other high-albedo surfaces such as glaciers, bright deserts, and salt pans are treated in a similar manner.

Another challenge is modeling broken clouds situations, as the satellite pixel records mixture of cloud and surface properties. These situations in some regions are common, and so the Solargis team has paid special attention to it.

In the picture, you can see an example of cloud detection over snow-covered areas along the US-Canada border. In the old version of the model, the cloud representation has frequent holes and high-contrast patches due to over- and under-correction of cloud index over snow surface. In the new model, the clouds are more consistent, homogeneous, and accurate due to enhanced correction.

Old version of the model (daily GHI average on March 2, 2023):

New version of the model (daily GHI average on March 2, 2023):

Sun position

Several steps of the model use the calculations of the exact Sun position. Thanks to the implementation of more complex parameterization, we can obtain a more accurate position of the Sun. This allowed enhancing the results a bit more.

We have also incorporated the effect of eclipses. These may not be important for long-term calculations at the prospecting stage of a solar project but will be useful when looking at monitoring or forecasting applications.

What to expect from the new model

As a result of incorporating the GOES-R satellite and upgrading our solar data models (as well as new processing infrastructure to keep the delivery time short), we are now able to provide 10-min resolution datasets since 2019 (as opposed to the previous 15-minute resolution).

Overall model performance has improved as the validation statistics prove – see the next section for more info.

For some areas, the upgrade of the model has resulted in slightly different long-term values (see map below). The reason? New changes implemented in the processing chain of GOES-R (2018 onwards) and additional model fixes of the model for the previous period.

Validation with ground stations

Recalculation of validation statistics also reflect the improved results of the new Solargis model for the Americas.

After comparing Solagis irradiance value with ground-measured data on more than 60 locations, we came to several conclusions:

• RMSD values have been reduced. Especially for direct normal irradiance values (DNI), RMSD is now lower for most of the sites with available ground-measured data. RMSD of global horizontal irradiance (GHI) has been reduced as well, with lower intensity than DNI. A lower RMSD means a lower spread of error (check more about how to validate solar models in this white paper).
• General decrease of DNI bias. GHI bias remained stable. Small increases in the bias were observed in few of the locations (below the uncertainty of ground measuring instruments). This is expected after upgrading a solar model like ours; a model that is applied consistently and does not use per-site model tuning.
• Overall model performance has improved.

More information and validation results available here.

To keep the best quality of solar radiation data calculated from satellite observations, it is required a constant effort of adapting data processes and models. Although that can sometimes be challenging, we are excited to see how the achievements made by our satellite data team for the GOES region will help unlock more and more opportunities for solar energy with higher-quality solar resource data services.

Please contact us if you have any questions about how the new generation of satellites is helping improve the energy estimates of your solar projects.