Performance of artificial intelligence model (LSTM model) for estimating and predicting water quality index for irrigation purposes in order to improve agricultural production.

Abstract

The primary goal of this study is to predict the current and future water quality index for irrigation (WQII) of the western Mitidja alluvial aquifer in northern Algeria. The modified WQII was used to evaluate groundwater suitability for irrigation through geographic information system (GIS) techniques. Additionally, a long short-term memory (LSTM) model was employed to calculate the WQII and map future groundwater quality, considering factors like overexploitation, anthropogenic pollution, and climate change. Two scenarios were analyzed for the year 2030. Results from applying the modified WQII model to 2020 data showed that about 83% of the study area has medium to high groundwater suitability for irrigation. The LSTM model exhibited strong predictive accuracy with determination coefficients (R2) of 0.992 and 0.987, and root mean square error (RMSE) values of 0.061 and 0.084 for the training and testing phases, respectively. For the first 2030 scenario, the area with low and medium groundwater suitability is expected to increase by 4% and 7% compared to the 2020 map. Conversely, under the second scenario, groundwater quality is predicted to improve, with a decrease of 14% and 11% in the low and medium suitability areas. The combination of the modified WQII and LSTM model proves to be an effective tool for estimating and predicting water quality indices in similar regions globally, offering valuable insights for water resource management and decision-making processes.