Forecasting of Groundwater Quality by Using Deep Learning Time Series Techniques in an Arid Region

Abstract

Groundwater is regarded as the primary source of agricultural and drinking water in semi-arid and arid regions. However, toxic substances released from sources such as landfills, industries, insecticides, and fertilizers from the previous year exhibited extreme levels of groundwater contamination. As a result, it is crucial to assess the quality of the groundwater for agricultural and drinking activities, both its current use and its potential to become a reliable water supply for individuals. The quality of the groundwater is critical in Egypt’s Sohag region because it serves as a major alternative source of agricultural activities and residential supplies, in addition to providing drinking water, and residents there frequently have issues with the water’s suitability for human consumption. This research assesses groundwater quality and future forecasting using Deep Learning Time Series Techniques (DLTS) and long short-term memory (LSTM) in Sohag, Egypt. Ten groundwater quality parameters (pH, Sulfate, Nitrates, Magnesium, Chlorides, Iron, Total Coliform, TDS, Total Hardness, and Turbidity) at the seven pumping wells were used in the analysis to create the water quality index (WQI). The model was tested and trained using actual data over nine years from seven wells in Sohag, Egypt. The high quantities of iron and magnesium in the groundwater samples produced a high WQI. The proposed forecasting model provided good performances in terms of average mean-square error (MSE) and average root-mean-square error (RMSE) with values of 1.6091 × 10−7 and 4.0114 × 10−4, respectively. The WQI model’s findings demonstrated that it could assist managers and policymakers in better managing groundwater resources in arid areas.