Exploring groundwater patterns in Souss-Massa Mountainous Basin, Morocco: A fusion of fractal analysis and machine learning techniques on gravity data

Abstract

Groundwater potential in Morocco’s Souss-Massa mountainous basin (SMMB) is being identified using geospatial tools and geological data. We deployed four mathematical models, namely Data-Driven Multi-index Overlay (DMIO), Geometric Average (GA), Support Vector Machine (SVM), and Logistic Regression (LR), to establish data-driven patterns among the nine influencing factors, primarily drainage density, permeability, slope, distance to rivers, elevation, lineament density, distance to lineaments, intersection node density, and rainfall. Based on the Concentration-Area (C-A) fractal approach, the findings of the four models were developed and classified into five levels of potentiality ranging from very low to very high. The regions designated as having high and very high potentialities for the DMIO, GA, SVM, and LR models, respectively, account for 22.44 %, 9.80 %, 19.36 %, and 26.77 % of the overall basin. We validated the models by calculating each model's area under the ROC curve (AUC). The estimated AUC values are more than 70 %, suggesting the model performs well. The four models' performance was compared, revealing that the SVM model outperforms the others. Gravimetric data shows that possible groundwater zones closely coincide with gravimetric lineaments. The findings of this study can provide valuable insights to decision-makers, allowing them to improve decision-making processes and develop holistic groundwater resource management in the Souss-Massa mountainous basin (SMMB).

Keywords

• Groundwater
• Machine learning
• Fractal
• Souss-Massa mountainous basin