In this study, daily river stage–discharge relationship was predicted using different modeling scenarios. Ensemble empirical mode decomposition (EEMD) algorithm and wavelet transform (WT) were used as hybrid pre-processing approach. In the WT-EEMD approach, first temporal features were decomposed using WT. Furthermore, the decomposed sub-series were further broken down into intrinsic mode functions via EEMD to obtain features with higher stationary properties. Mutual information was used to select dominant sub-series and determine efficient input dataset. Relevance vector machine (RVM) was applied to forecast river discharge. Three scenarios were developed to predict river stage–discharge process. First, a successive-station form of forecasting was proposed by incorporating geomorphological features into the modeling process. Subsequently, an integrated RVM (I-RVM) was trained based on the concept of the cascade of reservoirs and the meta-learning approach. The proposed I-RVM had the semi-distributed characteristics of the river discharge model. Finally, a multivariate RVM was trained to predict discharge for different points of the river. For this reason Westhope station's features were used as input to predict discharge at downstream of the river. Results were compared with rating curve and capability of proposed models were approved in prediction of short-term river stage–discharge.

You do not currently have access to this content.