Lake water level modeling using a Long-Short-Term-Memory (LSTM) neural network

High lake water level-induced flooding could cause catastrophic property damage and loss of life. The frequency and severity of lake flooding has been increasing in recent years, likely due to climate change . To quantify the lake flooding risk, accurate modeling of lake water level is critical. However, simulation of lake water level is a challenging problem in the field of hydrology, due to the various hydrological and morphological characteristics of river-lake systems. To solve this challenge, Moody's RMS has developed a coupled physical based – Machine Learning model, using a Long-Short-Term-Memory (LSTM) neural network which incorporates both dynamical variables and static variables . This model is tested and validated with representative lakes in the Southeastern US, and compared with other models including linear, dense, decision tree regression, random forest, and convolution neural network, which demonstrates the reliability and superiority of LSTM in lake water level modeling