In this work, a new parameter of fracture media is introduced to the conventional DRASTIC model to establish a new model for the fractured aquifer vulnerability mapping. Therefore, Artificial Intelligence technique (AI) is employed to reduce potential uncertainties using unsupervised Multi-Frameworks Technique (MFT) (Standard DRASTIC-Fr Framework (SDF) and Fuzzy Membership Framework (FMF)) and supervised learning based on Multi-Models Approach (MMA) (Genetic Algorithm (GA) and Genetic Expression Programming (GEP)). This novel framework comprises two stages; Stage 1 building of two frameworks MFT and models MMA; Stage 2 developing four frameworks including MFT/MMA combination based on Support Vector Machine algorithm. The proposed framework is tested with the fractured aquifer of the Meknassy area (central Tunisia), where the rapid expansive agricultural habits have been converted into excessive usage of fertilizers. Outputs demonstrate that the vulnerability in the study area is very low VL =31%, low L = 21.5%, moderate M = 22.9%, high H = 15% and very high VH =9.6%.

The primary aims of this research paper involve the creation and verification of machine learning-based quality models that utilize Integrated Irrigation Water Quality Indices (IIGWQIs) through an integrated GIS approach. We utilize the Least-Squares Support Vector Machines (LS-SVM) and the Pearson Correlation Fuzzy Inference-based System (PC-FIS) to establish forecasts for groundwater quality in the Meknassy basin. This basin serves as a representative case of an irrigated region in a mining environment under arid climatic conditions in central Tunisia. The evaluated factors for groundwater quality encompass the Irrigation Water Quality Index (IWQIndex), Sodium Adsorption Ratio Index (SARIndex), Soluble Sodium Percentage Index (SSPIndex), Potential Salinity Index (PSIndex), Kelley Index (KIndex), and Residual Sodium Carbonate Index (RSCIndex). These factors were determined based on measurements from 53 groundwater wells, which included various physico-chemical parameters. The hydrogeochemical facies identified included Ca-Mg-SO₄, mixed Ca-Mg-Cl-SO₄, and Na-Cl facies, revealing processes such as carbonate weathering, carbonate dissolution, interactions between rocks and groundwater, and mixing ionic substitution. In terms of the irrigation suitability categories, the IWQIndex, SSPIndex, PSIndex, Kindex, and RSCIndex indicated no limitation or minimal limitation (77.36%), secure (92.45%), favorable to excellent (66.04%), favorable (81.13%), and average to secure (88.68%), respectively. However, only 15.09% were considered favorable, according to SARIndex. The evaluation of the predictive models revealed the effectiveness of both the PC-FIS model and the LS-SVM model in accurately forecasting the IIGWQIs.

Groundwater potentiality assessment in an arid zone using a statistical approach and multi-criteria evaluation, southwestern Tunisia

In the current work, an enhanced model was developed to map groundwater spring potential using parameters derived uniquely from digital elevation model (DEM) as inputs. The proposed method is based on combining Quantum Particle Swarm Optimization (QPSO) and the Credal Decision Tree (CDT) groups (QPSO/CDT model). The principle of the suggested algorithm is to establish a CDT tree realized according to Random Subspace model (RSS). Then, we integrated QPSO to improve the three indices (subspace size, number of CDT sub-groups, and the CDT highest-range-of-trees). To reach this goal, a case study area in northeast Tunisia (region of Mornag) was chosen and 10 parameters were derived from the DEM. The result shows high accuracy of the QPSO/CDT model outputs compared to other machine learning models. Across the ten parameters, the convergence index, topographic wetness, drainage density, and altitude are the most relevant parameters.