Validation of the Model

We present validation of this model in three separate ways. In the first instance, Figure 8.8 gives a visual comparison of the moving average given in Fig. 8.3 and the generic model.

Clearness Index

Fig. 8.8 Comparison of the moving average and generic model for Adelaide data

These curves are not dissimilar, thus lending credence to the idea that the generic model is suitable. The second method of validation is given by the work of Jacovides et al. (2006) who compared 10 models in the literature for predict­ing diffuse fraction. One of these was a precursor to this generic model (Boland et al. 2001), wherein we used some data from the Geelong weather station to esti­mate parameters. The estimated ji0,p1 were similar to the present ones. Jacovides and his co-workers found that our model performed well for data from Cyprus in a reasonably exhaustive study.

In fact, two of the other models that were tested in that study, that of Reindl et al. (1990) and Karatasou et al. (2003) have been used here for a further val­idation of the generic model. The reason that the Reindl model has been cho­sen for this validation is that the study in question appears to have been quite well performed, with a list of 28 possible predictor variables being examined for their worth. We will refer to this study in Section 8 on adding more explanatory variables. The Karatasou model was one of the best performing models in the study – as it constructs the model with data from that region. Figure 8.9 gives a comparison between their one predictor models and the present model. The Karata – sou model is the lowest curve, the Reindl one is the piecewise linear and our model is the third. Additionally, we calculated the NRMSD and normalised mean bias dif­ference (NMBD) for all three models and obtained results given in Table 8.2. It can be seen from these results that the new model performs very well compared to these models. Further comparisons of this type will be reserved for the model developed with more predictors.

Fig. 8.9 The generic model and the Reindl and Karatasou models applied to the Adelaide data

Table 8.2 Comparison of statistical measures – the present model and the Reindl and Karatasou models

Boland and

Reindl et al.

Karatasou et al.

Ridley

NRMSD (%)

22.9

24.1

30.8

NMBD (%)

1.4

2.1

8.1

Updated: August 4, 2015 — 4:26 pm