Simulation of dew point temperature in different time scales based on grasshopper algorithm optimized extreme gradient boosting

• Published in: Journal of hydrology (Amsterdam) Vol. 606; p. 127452
• Main Authors: Dong, Jianhua, Zeng, Wenzhi, Lei, Guoqing, Wu, Lifeng, Chen, Haorui, Wu, Jingwei, Huang, Jiesheng, Gaiser, Thomas, Srivastava, Amit Kumar
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2022
• Subjects: Cross-validation; Dew point temperature; Extreme gradient boosting; Grasshopper optimization algorithm; Time scale; Time scale; Grasshopper optimization algorithm; Extreme gradient boosting; Dew point temperature; Cross-validation
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2022.127452

•The GOA-XGBoost model was the best model for estimating Tdew.•The most important meteorological input was actual vapor pressure.•The hybrid modelling effects for Tdew under daily and hourly scales were assessed. Dew point temperature (Tdew) plays an important role in hydrology, meteorology, and other related research. This study evaluated the ability of a new machine learning model (hybrid extreme gradient boosting with grasshopper optimization algorithm (GOA-XGBoost)) to estimate Tdew and compared it with two other tree-based models (XGBoost and random forest (RF)). We collected meteorological data namely actual vapor pressure (ea), maximum air temperature (Tmax), minimum air temperature (Tmin), maximum relative humidity (RHmax), minimum relative humidity (RHmin), atmospheric pressure (Pa), 2 m high wind speed (Ud), during 2016–2019 on daily and hourly time scales from the Sijiqinglin station in China to train, test, and validate each model. The results showed that the GOA-XGBoost model performed best, and the RF model had severe over-fitting problems during the validation phase at daily time scale. The models showed the best accuracy and stability when the input was ea (on average R2 = 1.000, RMSE = 0.296℃, MBE = 0.001℃, MAE = 0.167℃, and KGE = 0.991). The models had more significant errors when the inputs were Tmax, Tmin (on average R2 = 0.721, RMSE = 6.756℃, MBE = -0.101℃, MAE = 5.071℃, and KGE = 0.771). The estimation loss exhibited by the models were similar for the hourly and daily scale patterns. T and RH were the most basic meteorological factors and adding extraneous factors would affect the estimation accuracy of the model. The variability of meteorological data varied less on an hourly scale than on a daily scale. Therefore, the accuracy of the models was higher, but the data set and the volume of operations became larger. This led to a possible reduction in model stability, but the hourly scales are better suited for assessing the effects of simulations in extreme situations. Taking accuracy and stability into account, the GOA-XGBoost model was the best model and the most practical input for both time scales was ea. Therefore, in subsequent studies, the GOA-XGBoost model can be combined with the input ea to estimate Tdew accurately.