25 Mar State-of-the-art Methodologies

After joining forces, the Team has developed a new methodology for satellite retrieval of high frequency (15-min ≈ real-time) high spatial resolution (0.05 degree longitude x 0.05 degree latitude) DNI, GHI and DHI solar radiation spectra at the Earth’s surface that include the effects of clouds and aerosol such that the total DNI, GHI and DHI in each pixel provides a realistic measure of the solar energy.  In the first step, data from the EUMETCAST station of IAASARS/NOA was used. The system operationally retrieves, processes, archives and provides online, and in real-time, data retrieved from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) onboard the Meteosat Second Generation 3 (MSG3) satellite. This allows us to obtain pixel values for various cloud products and atmospheric parameters. Behind the maps lies deep science. In the second step, this data together with geolocation information is fed to a state-of-the-art neural network for converting cloud and aerosol data to solar radiation spectra at the surface.

“Solar energy modeling is now Big Data”

The neural network is trained using a giant (2.5 million record) look-up table (LUT) of radiative transfer simulations for both clear and cloudy skies generated by a bank of PCs running in parallel for over 3-months. Solar energy modeling is now Big Data. This approach is novel since:

  1. the effects of clouds and aerosol on the solar radiation spectrum at the surface are included directly, whilst the vast majority of solar energy estimates do not take these important factors into account,
  2. the satellite retrievals are of high temporal (15-min) and spatial resolution (0.05 degrees) enabling large-scale, high resolution real-time maps to be produced,
  3. the trained neural network is much faster than radiative-transfer simulations and enables real-time capability,
  4. at each pixel, the full DNI, GHI and GHI spectrum at high spectral resolution (0.5nm) is generated.

“Ten years ago, the prospect of being able to calculate your Vitamin D dose from a satellite weather map would have been science fiction”

For each pixel, we also extact the UV part of the surface solar radiation spectrum and multiply it by appropriate windowing functions to calculate important UV-dervied measures. In particular, on an operational basis, we provide the UV index (UVI) and Vitamin D effective dose (VDED) as nowcast products. These products can find direct application and will be of benefit to raising standards in health, tourism and outdoor leisure markets where people are exposed to sunlight on a daily basis during their outdoor activities. Ten years ago, the prospect of being able to calculate your Vitamin D dose from a satellite weather map would have been science fiction. Products like these will soon be available as apps on your mobile phone and “sentient” in the sense that such quantities will be calculated from your physical geolocation using GPS.


By combining a large-scale LUT spanning the broadest range of atmospheric conditions with a neural network, we have been able to dramatically speed-up the process of calculating solar radiation spectra in each pixel of the global domain. Our high-speed calculation system has been validated against 10,000 simulated results in over 0.5 million pixels (so far for Europe and the Mediterranean). Moreover, the system is continiously evolving and being fine-tuned with the outputs being monitored and validated to ensure their accuracy and quality.

No Comments

Sorry, the comment form is closed at this time.