With the dramatic reduction in solar panel costs over the past 15 years, we have come to view solar project economics through the cost lens. Unfortunately this downward cost trend has reversed (ref Lazards LCOE studies) and companies are having to think harder about the revenue line: volume and price.
In our experience, clients consistently overlook the importance of yield in their “cost per kWh” or “Levelised Cost of Energy” (LCOE) calculations because it’s a black box. Corrie has spent years researching the UK solar resource because the commercial viability of any tracker is critically dependent on the additional yield of the tracker compared to its cost. Over the next few weeks we will be writing a blog series about solar yield prediction. We’ll explain the major sources of error in solar yield predictions; quantify the UK solar resource; and help you understand how different solar technologies perform.
Lessons in solar yield prediction
In 2018-2020, the International Energy Agency (IEA) produced a comprehensive summary of the uncertainties in long term yield prediction, the cornerstone of any solar project. The most important factor is site-specific irradiation (in kWp/kWdc capacity installed) but the real value of the work was to show how susceptible this critical input is to the software, user experience and the assumptions used. In a great experiment, it gave the same data to seven skilled specialists and compared their results in two benchmarking exercises. Of the differences based on personal choice, the largest impacts were:
• The irradiance database selection and site adaptation (especially for mountainous terrain)
• Degradation assumption
• Total modelling uncertainty values (as seen in the P50 and P90 ranges)
• Soiling and far/near shading
The figure below shows the standard deviation of each software model step, normalised against the average annual global horizontal irradiance (Source: IEA 2020). The immediate conclusion is that more than half your error lies in choosing the GHI; the transpositional model (converting it into the right plane); and assessing the horizon shading. Has your consultant done that correctly? These figures are for the best case, using experienced analysts with no commercial interest. We’ve seen tender bids with up to 30% differences in base yield.
For daily and hourly irradiation, the errors are much higher. The root mean square error (RMSE) for the best performing yield models increases from 3-6%, to 19-23% for monthly/hourly irradiation estimates respectively. So take particular care when matching your onsite half hourly demand profile – run a few scenarios. Personally, I don’t like the use of “typical meteorological year” or TMY data. For robust analysis, TMY-based (P50) estimates should be complemented with multi-year time-series data to understand the potential range of outcomes and derive probabilistic yield forecasts (P75, P90, etc.).
The right UK tracker can generate up to 30% more energy per panel and generate longer through the day. But the differences in performance between tracking and fixed panels can only be accurately assessed using data on at least an hourly frequency. That is why in the next post we’re explain why we went back to basics and quantified the accuracy of UK solar yield predictions.
