Aim of the study:
The objective of this study is to evaluate the impact of three kernel functions − Pearson VII, radial basis function (RBF), and polynomial − on the predictive performance of Support Vector Regression (SVR) and Gaussian Process Regression (GPR) models.

Material and methods:
Three machine learning models − Random Forest (RF), Support Vector Regression (SVR), and Gaussian Process Regression (GPR) − were applied to estimate monthly evaporation at Boukourdane Dam, Algeria. The dataset included 240 observations over 20 years, with the following inputs: max./min. air temperature, relative humidity, wind speed, and water temperature; the output being: evaporation.

Results and conclusions:
Model performance was evaluated via Correlation Coefficient (CC), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). RF outperformed GPR and SVR across kernels, achieving MAE = 1.01 mm, RMSE = 1.29 mm, and CC = 0.81 in testing. Moreover, the Pearson VII kernel delivered the highest accuracy within both the GP and SVM frameworks. Sensitivity analysis highlighted relative humidity as the most influential factor in evaporation forecasting.