While satellite data bias for a site can be confidently guaranteed only between +/-3%-6% of daytime mean irradiance (depending on climate and terrain), a documented strength of satellite models is their ability to capture relative interannual variability for a given site (Ineichen 2011). In other words, the models may exhibit a bias for a particular location as a result of, for example, the difficult nature of the terrain and low resolution of the aerosol data; however, this bias will tend to persist over the long term. Therefore, if the satellite model can be calibrated against a short-term measurement campaign, its long-term accu – racy—that is, its ability to predict irradiance before and after the measurement period—should be substantially improved. A calibration campaign of 6-12 mo will typically reduce the MBE confidence interval by half. This calibration process is often referred to as site adaptation or measure-correlate-predict.
Site adaptation methodologies range in complexity from a simple bias correction, such as applying the same calibration factor to all predicted data points, to more sophisticated and generally more effective techniques consisting of matching measured and modeled frequency distributions by reducing the KSI error metric, whereby a different correction factor is applied to the model depending on its value. In arid areas with sparse cloud cover, MBE is usually driven by problems with aerosol parameterization; therefore, methods based on adaptation of aerosols to local conditions may be very effective. Figure 2.17 is an example of site adaptation.