Comparison of supervised learning methods for prediction of monthly average flow

Long-term planning of water engineering systems requires knowledge of long-term availability of water, most often in the form of monthly average flow information. Knowledge from stochastic hydrology is most often applied, although possible scenarios also involve generation of synthetic flow. The use of climatic models imposes the possibility of modelling based on future scenarios, and it is assumed in the paper that supervised learning can be applied for this purpose. The paper analyses accuracy of three supervised learning models in three approaches and the autoregressive model in the first approach, for predicting monthly average flow as related to the length of a historic dataset