LAUSR

Accurate prediction of dew point temperature is very important in decision making in many fields of water resources planning and management, agricultural engineering and climatology. This study investigates the ability of some data-driven models (DDMs) in predicting daily dew point temperature. These models include traditional group method of data handling (GMDH), improved GMDH models (GMDH1, GMDH2), and two hybrid GMDH-based models (GMDH-HS and GMDH-SCE) which were developed by combination of GMDH with two optimization algorithms, harmony search (HS) and shuffled complex evolution (SCE). 11 years of daily recorded weather variables at Kerman synoptic station including mean temperature (Ta), sunshine hours (S), soil temperature (Ts), mean relative humidity (Rh), and wind speed (Ws) were used to evaluate the proficiency of developed models. Sensitivity analysis revealed that Rh is the most influential input variable in predicting dew point temperature. Seven quantitative standard statistical indices including coefficient of efficiency (CE), correlation coefficient (CC), root mean square error (RMSE), mean square relative error (MSRE), mean absolute percentage error (MAPE), relative bias (RB) and threshold statistic (TSx) were employed to examine the performance of applied models. The results indicated the superiority of combinatorial models (GMDH-HS and GMDH-SCE) to the other developed models in predicting the dew point temperature (Tdp). In terms of threshold statistic, GMDH2-HS had the highest values of TSx (the best model) and GMDH2-SCE, GMDH1-HS, GMDH1-SCE, GMDH2 and GMDH1 got the next ranks, respectively. It was observed that GMDH2-HS could predict the Tdp (with CE = 0.979 and RMSE = 0.745) better than the other models (with CE = 0.958 and RMSE = 0.932, in average), indicating its high efficiency.