Forecasting of monthly average lake levels of Lake Michigan with artificial neural networks

Abstract

Forecasting of water level at various time intervals using historical record series is important in water resource management and related engineering. Similarly, a reliable estimation of water level change is required in drought and flood hydrology studies. In this study, Lake Michigan between 1981-2020 was modeled with 3 different Artificial Neural Networks (ANNs) using monthly average water level data. These are Multilayer ANN, Radial Based ANN, and Generalized ANN models. Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Coefficient of Determination (R²) were used as comparison criteria. When the results were compared, the lowest error rate and the highest coefficient of determination were seen in the 12 inputs of the MANN model (MAE= 0.0342, RMSE= 0.0435, R²= 0.9906).