Weight combinations approach of models with inverse variance and least square regression methods for dew point temperature estimation

Dew point temperature estimation with proper method is useful in many fields such as agricultural planning including crop protection to the damages, meteorological, hydrological and ecological studies. In this study, the forecast combination approach of regression tree, Group Method of Data Handling(GMDH) and experimental method were applied to forecast dew point temperature in Rasht, Yazd and Urmia stations. The input variables of individual models were the maximum wind speed, mean wind speed, maximum, minimum and mean temperature, mean relative humidity, maximum and minimum relative humidity and saturation vapor pressure. The used methods of weight combination approach were inverse variance and least square regression. In the individual models, GMDH is more efficient than other models, so that the RMSE decreasing from empirical and regression tree to GMDH model in Rasht station is 66.66% and 59.45%, respectively. The combination approach is more accurate rather than the individual models and least square regression method has less error than the inverse variance with different error criteria, so that Nash-Sutcliff coefficient in Rash, Urmia and Yazd stations is 0.97,0.96 and 0.87, respectively. Also, the kind of error criteria and defined power in the inverse variance method is effective on forecasting values and the proper power basis on available data was proposed. In order to investigate the impact of climate diversity, Rasht station has the least error. The use of robust individual models will also increase the ability of forecast combination approach.