Neurocomputing intelligence models for lakes water level forecasting: a comprehensive review

Abstract

Hydrological processes forecasting is an essential step for better water management and sustainability. Among several hydrological processes, lake water level (LWL) forecasting is one of the significant processes within a particular catchment. The complexity of the LWL fluctuation is owing to the diversity of the influential parameters including climate, hydrology and some other morphology. In this study, several versions of neurocomputing intelligence models are developed for LWL fluctuation forecasting at five great lakes Lake Superior, Lake Michigan, Lake Huron, Lake Erie, and Lake Ontario, located at the north of USA. The applied models are including M5-Tree, multivariate adaptive regression spline (MARS) and least square support vector regression (LSSVR). The models are developed using several input combinations that are configured based on the correlated lags in addition to the periodicity of time series. The sequential influence of the lakes order is considered in the modeling development. Also, cross-station modeling where lag time series of upstream lakes are used to forecast downstream LWL. Results are assessed using several statistical metrics and graphical visualization. Overall, the results indicated that the applied forecasting models efficient and trustworthy. The component of the periodicity time series enhances the forecasting performance. Cross-station modeling revealed an optimistic modeling strategy for learning transfer modeling of using information of nearby site.