Uncertainty analysis of artificial neural network models and support vector machine in rainfall estimation

In this research, we tried to determine the input composition and model for estimation of precipitation in Shahrood. To achieve this objective, monthly weather data including evaporation, temperature, relative humidity, solar radiation, wind speed during the period of 1963 to 1915, and artificial neural network and support vector machines have been used. 75% of the data was used for calibration and 25% for validation of the models. In this research, an artificial neural network of laminated perceptron with a sigmoid tangent function and 1 to 30 neurons in the hidden layer was used and a support vector machine model with radial base kernel function was used to estimate rainfall in Shahrood district. The performance of each model was evaluated using the statistical mean square error and correlation coefficient. The uncertainty of the models was also determined for two parameters, d-factor and p-factor. Considering that both models have good performance in rainfall estimation, the support vector machine model with less error and uncertainty than artificial neural network model has better performance in predicting rainfall in Shahrood. Therefore, a support vector machine model can be used as a very suitable model for precipitation estimation.