Predict Saturated Thickness using TensorBoard Visualization

Abstract

Water plays a critical role in our living and manufacturing activities. The continuously growing exploitation of water over the aquifer poses a risk for over-extraction and pollution, leading to many negative effects on land irrigation. Therefore, predicting aquifer water level accurately is urgently important, which can help us prepare water demands ahead of time. In this study, we employ the Long-Short Term Memory (LSTM) model to predict the saturated thickness of an aquifer in the Southern High Plains Aquifer System in Texas, and exploit TensorBoard as a guide for model configurations. The Root Mean Squared Error of this study shows that the LSTM model can provide a good prediction capability using multiple data sources, and provides a good visualization tool to help us understand and evaluate the model configuration.