Improving Solar Production Risk Assessment with Climate Trending and Stochastic Simulation

To evaluate current and future sites for solar energy production, the solar energy industry has generally relied on hypothetical power generation estimates calculated from sample weather datasets – either Typical Meteorological Years (TMYs) or historical time series data. In this work, we evaluate the shortcomings of these weather datasets, propose a method to overcome these shortcomings, and analyze the ramifications of the results for solar production risk assessment and decision making.

We start with an analysis of current methods. In using TMYs, one creates a “typical” year by concatenating 12 calendar months – with each calendar month represented by the most “typical” example of that month from historical data – and then calculates hypothetical power generation on this year. Historical time series analysis, on the other hand, evaluates power generation on the entire historical dataset, resulting in estimates of the mean, median, and variability in annual power generation.

Neither of these approaches takes climate trends into account–in fact, by calculating typical or mean power generation from historical data, the resulting estimates are affected by climate trends in exactly the wrong direction, with discrepancies only growing over time. For example, a site that is becoming cloudier over time will have an estimate based on sunnier–not cloudier–weather than the present day. Further, as irradiance data is reliant on satellite data and thus only available for a small number of years, any conclusions necessarily extrapolate from small datasets – leading to considerable errors in estimates of both typical and extreme annual production estimates.

To address these shortcomings, Sunairio calculates local climate trends, and uncertainty of those trends, explicitly and then creates stochastic weather simulations incorporating this information. By relying on simulation, Sunairio is also able to create arbitrary amounts of data, thus mitigating issues of data paucity.

To evaluate these methods, we considered a portfolio of 100 representative utility scale solar sites. We discovered an overall negative GHI trend of -0.225 W/m² per year, with 83% of the sites having negative GHI trends; we also found TMY estimates to overpredict with respect to historical time series analysis by 1%, likely due to selection of “typical” months from skewed GHI distributions. Creating 1000 Sunairio simulation weather sets for each solar site, we found site-dependent local-climate-trend production adjustments to range from -6.16% to 2.77% of simulated production in 2022–adjustments that grew in magnitude to -13.48% to 4.63% when extrapolated out to 2034. Over all 100 sites, these adjustments indicated production losses of -2.43% in 2022 and -4.98% in 2034 with respect to TMY estimates. Finally, the degree of uncertainty in production estimates was at least four times lower in the Sunairio simulation data compared to production estimates using historical time series.