Assessing solar radiation models using multiple variables over Turkey

Evrendilek F., ERTEKİN C.

CLIMATE DYNAMICS, vol.31, no.2-3, pp.131-149, 2008 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 31 Issue: 2-3
  • Publication Date: 2008
  • Doi Number: 10.1007/s00382-007-0338-6
  • Journal Name: CLIMATE DYNAMICS
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.131-149
  • Akdeniz University Affiliated: Yes


Solar radiation drives many environmental processes; however, needs to be estimated indirectly from more commonly measured meteorological variables since these data are not readily available from most climate stations. A geo-referenced dataset from 159 climate stations of 17 variables (maximum possible sunshine duration, mean, minimum and maximum air temperature, soil temperature, mean and maximum relative humidity, precipitation, cloudiness, evapotranspiration, extraterrestrial radiation, day length, declination angle, day of the year, latitude, longitude, and altitude) was used to model spatio-temporal dynamics of solar radiation over Turkey. A total of 78 empirical models of different mathematical functions with a different combination of 17 explanatory variables were compared based on the error statistics of the Jackknifing validation. The empirical models had adjusted coefficient of determination (R-adj(2)) values of 22.7-96.5% based on the parameterization dataset (P < 0.05). Models 46 and 22.2 provided the most robust performance and were identified as generic models for the estimation of monthly changes in solar radiation over topographically complex terrain of the entire Turkey as a function of maximum possible sunshine hours, extraterrestrial solar radiation, mean temperature, and precipitation. The quadratic and cubic models performed best in terms of the error statistics (P > 0.05), while the performance of the hybrid models was worse than that of the linear, quadratic and cubic models in terms of maximum relative percentage error (e) (P < 0.01). In comparing the interpolation methods of inverse distance weighting and universal co-kriging, anisotropic spherical semi-variogram model of universal co-kriging was found to provide the best description of spatial autocorrelation and variability latent in these data based on the spatial leave-one-out cross-validation.