Predicting Groundwater Indicator Concentration Based on Long Short-Term Memory Neural Network: A Case Study.

Area of Science:

• Environmental Science
• Hydrogeology
• Data Science

Background:

• Groundwater quality prediction is crucial for sustainable water resource management.
• Existing research in the study area primarily addresses water distribution and utilization, with limited focus on quality prediction.
• There is a need for advanced models to forecast groundwater quality, especially considering seasonal variations.

Purpose of the Study:

• To introduce and evaluate a Long Short-Term Memory (LSTM) neural network model for predicting groundwater quality.
• To forecast concentrations of key groundwater quality indicators including TDS, fluoride, nitrate, phosphate, and metasilicate.
• To assess the model's accuracy in predicting groundwater quality for dry and rainy periods.

Main Methods:

• Utilized groundwater monitoring data spanning October 2000 to October 2014.
• Selected five key indicators: Total Dissolved Solids (TDS), fluoride, nitrate, phosphate, and metasilicate.
• Applied a Long Short-Term Memory (LSTM) neural network model to time series data, accounting for seasonality.

Main Results:

• The LSTM model demonstrated high accuracy in predicting groundwater quality.
• Achieved Mean Absolute Errors (MAEs) of 0.21, 0.20, 0.17, 0.17, and 0.20 for the selected parameters.
• Obtained Root Mean Square Errors (RMSEs) of 0.31, 0.29, 0.28, 0.27, and 0.31, respectively, indicating reliable predictions.

Conclusions:

• The developed LSTM model is effective for groundwater quality prediction.
• Findings provide a valuable reference for groundwater management and sustainable utilization in the study area.
• Offers a novel approach for similar hydrogeological conditions in coastal cities, enabling timely warnings and mitigation measures.