Spatial and temporal model for WQI prediction based on back-propagation neural network, application on EL MERK region (Algerian southeast)

Abstract

For the aims of evaluate the groundwater suitability for human consumption, and modelling the water quality index, temporally and spatially, using Artificial Neural Network. 37 samples were harvested from eight different wells in the region during the period from 2016 to 2019 in pre-monsoon and post-monsoon to conduct the necessary analyses of pH, mineralisation, TH, TA, Ca2+, Mg2+, K+, Na+, Cl−, HCO3−, SO42−, NO2− and NO3−. Based on WQI results, the groundwater of MioPliociene is classified as good and poor water in 34.5% and 65.5% of the samples, respectively. The seasonal variations of the chemical composition of water indicate a presence of NO3− and NO2− in an ascending way. It is also appeared that the aquifer was very sensitive to weathering. PCA illustrates the corelations. The WQI was always posed in a homogeneous group with TH and mineralization, which indicate that TH and mineralisation are main factors in controlling water quality. ANN method is used to create a model to predict WQI using mineralisation, TH, NO3− and NO2− as inputs. The model shows a performance of 1.64 as MSE. The suitability of model confirmed using Bland-Altman test. The test shows that the model is suitable for predicting WQI with an error rate of 9.3%.

Keywords

• Artificial neural network
• Water quality index
• Groundwater
• Modelling
• ILLIZI
• EL MERK