Choice of model type in stochastic river hydrology

Stochastic models are widely used in hydrology, mainly for forecasting and data generation purposes. There are a number of commonly used models for these purposes such as the Seasonal Autoregressive Integrated Moving Average (SARIMA), deseasonalized Autoregressive Moving Average (ARMA), Periodic Autoregressive (PAR), Transfer Function-Noise (TFN) and Periodic Transfer Function-Noise (PTFN) models. We compare the relative performance of each model type by fitting the above five models to the monthly flows of the Ganges river. For the TFN and PTFN models, monthly rainfall data of India are also used. The performance matrices evaluated are the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE) of validation forecasts. The results demonstrate that the choice of model type, especially between periodic and non-periodic, has important role in stochastic river hydrology.