Temperature-based model for monthly average hourly global solar radiation for the Caribbean island of Trinidad

Abstract

Hourly global solar radiation data are required to optimize various solar energy based designs. In this study, a new empirical temperature-based model for determining monthly average hourly global solar radiation on a horizontal surface for the Caribbean island of Trinidad was developed and validated using calibration and validation datasets of hourly global solar radiation and temperature from 2001 to 2005 and 2006 to 2010, respectively. The new model was evaluated alongside five existing monthly average hourly radiation ratio models due to Whillier/Liu and Jordan, Collares-Pereira and Rabl, Gueymard, Newell, and Jain. For all six models, measured and calculated monthly average hourly global solar radiation over the five-year validation period 2006–2010 were compared statistically using mean bias error, root mean square error, correlation coefficient, and Nash-Sutcliffe efficiency. For an average year, the temperature-based model was the best performing model for two months, the Collares-Pereira and Rabl and Gueymard models for ten months, and the Whillier/Liu and Jordan model for one month. Over an average year, the temperature-based and Gueymard models yielded root mean square errors of 0.12 MJ m−2 h−1 and 0.10 MJ m−2 h−1, respectively. Over the single years from 2006 to 2010, the models were evaluated for performance and robustness toward inter-annual fluctuations in global solar radiation and temperature. Some re-distribution of the same four models occurred among the months over the years, but overall, they were in agreement with those obtained for an average year. The Newell model was the most robust. Over these five years, the temperature-based and Gueymard models yielded root mean square errors of 0.17 MJ m−2 h−1 and 0.14 MJ m−2 h−1, respectively. For modeling of monthly average hourly global solar radiation on both yearly and average year bases, the temperature-based and Gueymard models were most suitable with root mean square errors of 0.15 MJ m−2 h−1 and 0.12 MJ m−2 h−1, respectively. Advantageously, the temperature-based model is independent of global solar radiation input unlike the other models.