The rising investment in solar energy projects has necessitated the development of improved methods to quantify and assess the uncertainty of solar resource data. The practical challenges in this area stem from the instrument used to monitor the measurand, the model and input data used to predict and forecast the measurand, and their interactions. By maintaining the proper traceability, these sources of uncertainty amplify or compensate each other as they propagate from the reference data to prediction and forecast, for example. This propagation of uncertainty has significant impacts on the prediction and forecast data, which subsequently affects the project's financing, as well as the levelized cost of energy (LCOE) and decision making at various steps. This chapter discusses the uncertainties associated with various forms of solar resource data and how these data impact the predictions of physical or empirical models that use such data. For the purposes of this chapter, solar resource data can be classified into three different categories: experimental data, as those measured at ground stations; modeled data, estimated for past periods using physical, semiempirical, or other radiative models; and forecast data, which use current experimental data and models to estimate the future irradiance for a particular area, season, and time. The latter can be distinguished between short-term forecasting (intrahour, intraday, and days-ahead) and long-term predictions for the next decades.

Accurate measurement, prediction, or forecasting of the solar resource is complicated by the rapidity with which the solar irradiance can change, both in magnitude and spectral distribution, and the varied environmental conditions experienced during measurements.

In the case of predicted and forecasted datasets, it is essential to understand the factors that impact their accuracy relative to ground measurements because of, for example, error propagation. In parallel, the quality of these measured data is key for confidence in the determination of uncertainty in predicted and forecasted datasets. Additional factors can be considered, such as the interannual variability if using only a short dataset of, for example, 12 months or less. Thus, the overall uncertainty of a modeled dataset should include an estimate of the uncertainty of the modeled solar resource, of the ground measurements, and that resulting Chapter 10-2

This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.