When it comes to forecasting solar irradiance a complex model is not necessarily the best

Clear-sky models are commonly used for solar forecasting. A clear sky model provides an estimate of solar radiation under a cloud-free atmosphere, which is often used as a foundation for further modeling.

Dazhi Yang compared three clear-sky models to investigate whether a complex model is better than a simple model. The three models, in increasing complexity, were the Ineichen-Perez clear-sky model, known for its simplicity and availability in common programming languages Python and R; McClear, a more complex model but with wider accessibility; and REST2, the most comprehensive model of the group.

Yang compared the models based on accessibility, forecast performance and statistical properties. Though previous studies suggested that more complex clear sky models represent the radiation in clear skies more accurately, he found that the root-mean-square errors of forecasts using different clear-sky models are similar.

“Generally, irradiance that was detrended using the clear sky index should be stationary, however, even the best clear-sky models today are unable to obtain a stationary clear-sky index time series. On this point, the more intricate clear-sky models do not possess any advantage in solar forecasting. Therefore, the accessibility becomes the most important concern when opting a clear-sky model during forecasting,” said Yang.

Yang found all three models to perform comparably in terms of overall irradiance forecast accuracy, and identified the McClear model as the best choice due to its accessibility.

“The conclusion of this paper allows solar forecasters to select clear-sky models in their works freely,” said Yang. “More importantly, this paper allows authors to rebut any criticism on the choice of clear-sky model.”

Source: “Choice of clear-sky model in solar forecasting,” by Dazhi Yang, Journal of Renewable and Sustainable Energy (2020). The article can be accessed at https://doi.org/10.1063/5.0003495 .