Application of linear genetic programming in conjunction with Kohonen's self-organizing map and ensemble empirical mode decomposition for improved streamflow prediction at lock and dam 10 along the Kentucky River

The accurate prediction of streamflow from precipitation and runoff information has been widely studied due to the complexities of the rainfall-runoff process. Popularized data-driven techniques like artificial neural networks (ANNs) and genetic programming (GP) have proven to be useful alternatives to more complicated conceptual and physically based models. Linear genetic programming (LGP), which is applied in this study, differs from GP in that it allows faster processing times and a greater abundance of solutions. The application of LGP on hydrologic information from the Kentucky River Basin resulted in slightly improved forecasting when compared to ANN models. To further explore the capability of LGP, hybrid models incorporating the data decomposition technique of ensemble empirical mode decomposition (EEMD) and the data clustering technique of a self-organizing map (SOM) were applied to the same study area. The EEMD-SOM-LGP hybrid model proved to significantly outperform the utilization of LGP on its own.