Benchmarking CPU vs. GPU on LSTM Model Using Groundwater Dataset

Surveying groundwater level is important to keep track of water availability in any region. Historical groundwater level data can provide key insights into the region's agriculture, irrigation, and livelihood of living things. Studying historical time series data on the groundwater level can be a key to understanding how a region might be affected under a drought or flood. Now imagine there is a tool that can read this time series data and forecast the future trend on the groundwater level. Long short term memory (LSTM) is such a kind of tool, which heavily depends on computational power because of the large data size and parameters that affect the groundwater level. This study benchmarks the performance of the LSTM model in predicting groundwater level using CPU and GPU. Groundwater level data collected by the United States Geological Survey (USGS) has been used to train the LSTM models. In addition to benchmarking, this study also reflects on the complexity of the dataset and potential optimization for groundwater level forecasting.