Lake level dynamics exploration using deep learning, artificial neural network, and multiple linear regression techniques

Abstract

Estimating the lake level dynamics accurately on a daily or finer timescale is important for a better understanding of ecosystems, especially the lakes in Badain Jaran Desert, China. In this study, lake level dynamics of Sumu Barun Jaran are simulated and predicted on a 2-h timescale using the deep learning (DL) model, which is structured for the first time in this area by considering critical environmental factors. Two machine learning methods, namely multiple linear regression (MLR) and the three-layered back-propagation artificial neural network (ANN), are also adopted for the prediction purpose. The performances of these models are evaluated by comparing the values of average relative error, the mean squared error, and the coefficient of determination. The result shows that the DL model performs better than MLR and ANN on these three criteria, and this DL model is beneficial for exploring the mechanism of lake level dynamics in Badain Jaran Desert.