What to expect from your solar data provider: Reliability

Solargis’ Technical Director, Tomas Cebecauer, and Managing Director, Marcel Suri speak about the core ingredients for a reliable database, Solargis’ ongoing efforts to enhance and validate its data services, and how users of Solargis’ data can most effectively undertake their own validation using on-site measurements.

Why is a reliable solar database important for the continued global growth of the solar industry?

Marcel: As solar energy continues to deliver ever lower energy costs than traditional forms of power generation, so it has expanded around the world. We’ve seen increasing uptake of solar in countries where it had not previously been competitive, or where infrastructure is less developed.

It is vital that solar data keeps pace with the rapid expansion of the industry, supporting asset owners and operators as they optimise ever more complex projects and ensure correct valuations.

What are the core ingredients of a reliable database?

Tomas: A good model must be able to handle outputs from a range of different sources, and ensure that any conflict is handled in a stable manner to ensure data is representative - for example in overlapping satellite zones or when modern satellites replace old ones.

A special challenge was aligning the modern satellite data with the older archives via in-house developed geometric and radiometric pre-processing tools. These helped us to include valuable long-term history, while assuring its high positional and temporal stability and overall reliability. Today Solargis archives are complete and date back to years 1994, 1999 and 2006, respectively, depending on the satellite region.

With the expansion of solar energy globally, it is essential to accurately represent all climate zones and geographical peculiarities. The data should work just as well in tropical rainforests or high latitudes as in the extreme deserts of Atacama.

Good models should deliver reliable outputs for all primary parameters: Global Horizontal Irradiance (GHI), Direct Normal Irradiance (DNI) and Diffuse Horizontal Irradiance (DIF) – and these must be in equilibrium with each other.

Typically, ensuring the accuracy of DNI data is more demanding. However, without good DNI, calculation of Global Tilted Irradiance (GTI) – which is vital for trackers – can be subject to large errors. A database that boasts highly accurate GHI, but falls short on DNI, may not be fit for purpose in many contexts.

What we observe with some other data products is ‘overfitting’, where the model is forced to match on-site measurements perfectly, while disregarding imperfections and the resulting uncertainty in those on-site measurements. There must be low bias, but crucially, also low Root Mean Square Error (RMSE) at all points. Critical - for a good model - is to maintain also a harmonised representation at the level of sub-hourly data values in all energy ranges. Again, balance is important – no single measurement, parameter or metric should dominate. This is indeed tricky, as we are working with imperfect reference data - the ground measurements.

The model should also be able to differentiate different types of aerosols and clouds properly, including high and rapidly changing atmospheric turbidity, occurrence of snow, ice, fog and thin clouds. Without this, it is hard to attain reliable data for mountainous, densely populated, industrial and coastal areas.

Finally, data must be available at high spatial resolution to accurately reflect unique site conditions. A digital terrain model, such as the one we have harmonised from multiple sources and implemented within the Solargis Prospect app, makes it possible to calculate global data layers at a grid resolution of 90 metres.

What has Solargis done to ensure maximum reliability from day one?

Marcel: At Solargis, we have not only put years of R&D effort into ensuring our models are robust and accurate, we have also ensured that our solar datasets are the most extensively validated.

We decided to develop our own computational and delivery systems even before we delivered our first commercial data products. At the beginning Dr. Richard Perez (a leading academic and professor at the State University of New York) provided support with definition of the models, and later we developed our own solar-computing methodology, also published in several academic papers.

We started in Europe and extended the coverage globally in steps. It took us several years to acquire the necessary historical satellite images. There were sticking points, such as availability of data on Aerosol Optical Depth, but now there are several options.

Tomas: We knew that it was important to ensure fast delivery to make sure asset owners had the most up to date intelligence, down to only a few minutes globally.

For the optimisation of a PV design, the ability to deliver data in 10 and 15-minute resolution is needed. Hourly data is not sufficient, anymore. It became also important to ensure reliable distribution of sub-hourly values, especially high and low energy values for the correct evaluation of combined performance of PV modules and inverters. We are increasingly receiving requests for 1-minute data to enable the simulation of hybrid PV- battery solutions.

We use data from meteorological models and global data processing projects to further increase the accuracy of our energy simulation models and PV performance analysis. This data is well-validated and in case of parameters such as air temperature, and atmospheric pressure, also recomputed at higher resolution and better accuracy. We have also developed a new high-resolution global albedo database and calculation scheme based on a complex methodology combining the best features of several public databases.

Marcel: Extensive validation is a critical step in demonstrating the accuracy of satellite-based solar data used to make development, financing and operational decisions. To prove reliability requires rigorous validation of all solar components. We have validated the model at over 1000 sites across the world, far in excess of other players in the market.

Ultimately, this combination of technology and experience enables us to save our customers money. Every 1% reduced uncertainty increases return on equity by approximately 5% for the investor.

What measurement data should you provide to support validation?

Tomas: Our customers can validate our data themselves, via one-to-one comparison to their on-site meteorological measurements or to PV production data. This further solidifies trust in our processes, once people see for themselves how accurate our data and energy calculations are.

Unfortunately, solar measurements are prone to many errors and issues, therefore prior to any data comparison, it’s important to control the quality of the measured data – otherwise the comparison suffers from errors. Optimally the evaluation of comparison should take into account the uncertainty of ground measurements.

Typical errors we see in solar measurements are due to polluted sensors, shading, misaligned instruments and outdated calibration. In addition, there can be missing records due to issues with data loggers, power supply or incorrect data manipulation and postprocessing.

Even solar data coming from high-grade scientific measuring networks suffer from various errors and need to be rigorously quality-screened before further use.

Marcel: The good news for the industry is that there are already professional commercial measurement providers who offer reliable services, helping to ensure smooth continuity between historical data, real-time updates and forecasting.

We believe strongly in transparency – everybody can see key data features and long-term values of Solargis data parameters globally, within the free version of our Prospect app. Reliable data come with a guarantee of validation: Solargis time series are available for any location and can be validated if good quality local measurements exist.

How can better data deliver better results for solar asset owners?

Marcel:  Our data can be used at all stages of the project lifecycle – from prefeasibility assessments, throughout project development, engineering and financial calculations. After the power plant enters operation, our models continue supplying data for regular performance evaluation and solar power forecasting.

Through our consultancy services, we also help large-scale asset owners to re-evaluate initial assessments on the basis of the improved accuracy and reduced uncertainty via site-adaptation of the models with the use of high quality local measurements.

Tomas: Solargis data and services are constantly undergoing further refinement and long-term development – we are systematically working on making the calculation schemes more accurate, while continuing to develop new models which make better use of the wealth of data from more detailed modern observation systems.

This allows us to plan for delivery of new products and services such as the new tools for manipulation, visualization, quality control and analysis of large-volume time series data. We are also focusing on developing tools for more efficient combined use of measurements and modelled data for reduced uncertainty. Ultimately, our photovoltaic simulation software will benefit from the improved data sets and support tools.