Artificial Neural Network models for estimating daily solar global UV, PAR and broadband radiant fluxes in an eastern Mediterranean site

In this paper, simple Artificial Neural Network (ANN) models for estimating daily solar global broadband as well as solar spectral global UV and PAR radiant fluxes have been established. The data used in this analysis are global ultraviolet UV (GUV), global photosynthetic photon flux density (PPFD-QP), broadband global radiant flux (Gh), extraterrestrial radiant flux (G0), air temperature (T), relative humidity (rh), sunshine duration (n), theoretical sunshine duration (N), precipitable water (w) and ozone column density (O3). By using different combinations of the above variables as inputs, numerous ANN-models have been developed. For each model, the output is the daily global GUV, QP and Gh solar radiant fluxes. Firstly, a set of 2 × 365 point (2 years) has been used for training each network-model, whereas a set of 365 point (1 year) has been engaged for testing and validating the ANN-models. It has been found that the ANN-modelsʹ accuracy depends on the parameters employed as well as spectral range considered. Comparisons between proposed ANN-models and conventional regression models revealed that the results of both methods are statistically significant. On closer examination of many error measures, though, it is clear that the ANN-models perform better overall. From this point of view, it turned out that the neural network technique is better suited further suggesting that the ANN methodology is a promising and a more accurate tool for estimating both broadband and spectral radiant fluxes.