Temporal and Spatial Flow discharge prediction using integrated artificial intelligence and pre and post-processing time series methods

Forecasting of river discharge is a important aspect of efficient water resources planning and management. In this study, time series pre and post-processing methods along with support vector machine (SVM) and Gaussian process regression (GPR) kernel based approaches were used to estimate flow discharge of two natural river in the United States with two consecutive hydrometric stations. The first river contained about 2 years of data and in the second river 4 years of daily discharge data was used. Different models were defined based on hydraulic characteristics and the capability of integrated pre and post-processing methods in two states of inter-station and between-stations was investigated. For data pre-processing, the Discrete Wavelet Transform (DWT) method was first used. Then, the high-frequency sub-series were selected and re-decomposed using the Ensemble Empirical Mode Decomposition (EEMD). Finally, sub-series with higher energy were imposed as inputs for kernel-based models. Non-linear neural average (NNA) model was also used for data post-processing. The obtained results from the defined models showed the high accuracy of the integrated methods used in the research in estimating flow discharge. At both stations, the error percentage was reduced by approximately 20 to 25% using the integrated pre-post-processing methods compared to the intelligent kernel based models. It was observed that in the case of river flow prediction based on the station's own data, the RMSE error value of the model decreased from approximately 0.3 to 0.26 and in the case of using the previous station data decreased from 0.44 to 0.33. Due to the high capability and accuracy of the pre-processing methods used in this study, similar studies are recommended in other rivers of the country.