Evaluating Capabilities of Gradient Boosted Tree and Optimized Random Forest Models in Estimating Daily Dew Point Temperature

Dew point temperature is very important in various fields including meteorology for weather forecasts. Therefore, it is necessary to provide suitable models to accurately predict the value of this meteorological variable for the practical use of agricultural engineers and nearby stations where it is not possible to measure this temperature. In the present study, we investigated the ability of four data-driven models, including gradient reinforcement tree, M5P tree model, random forest, and random forest optimized with genetic algorithm, in estimating daily dew point temperature. For this purpose, the daily meteorological data of two stations in Ardabil and Parsabad were used in the period of 2014 to 2019. The used meteorological parameters include minimum, maximum, and average temperature, relative humidity, sunshine hour, and wind speed, which were considered input variables for each of the mentioned models in 10 different combinations. The comparison of the results obtained for both stations showed that the M5P-8 model with a root mean square error of 0.54°C and a Wilmot coefficient equal to 0.998 in the Ardabil station and the M5P-6 model with a root mean square error of 0.29°C and Wilmot coefficient equal to 1.00 was introduced as the best models in Parsabad station.