Evaluating the Dibdibba Aquifer Productivity at the Karbala–Najaf Plateau (Central Iraq) Using GIS-Based Tree Machine Learning Algorithms

• Published in: Natural resources research (New York, N.Y.) Vol. 29; no. 3; pp. 1989 - 2009
• Main Authors: Al-Abadi, Alaa M., Handhal, Amna M., Al-Ginamy, Maithm A.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2020; Springer Nature B.V
• Subjects: Algorithms; Aquifers; Chemistry and Earth Sciences; Classifiers; Collinearity; Computer Science; Decision trees; Earth and Environmental Science; Earth Sciences; Elevation; Fossil Fuels (incl. Carbon Capture); Geographic information systems; Geographical locations; Geography; Groundwater; Learning algorithms; Machine learning; Mathematical Modeling and Industrial Mathematics; Mineral Resources; Original Paper; Physics; Productivity; Regression analysis; Specific yield; Statistical analysis; Statistics for Engineering; Sustainable Development; Transmissivity; Water wells; Water yield; Well capacity; Iraq; An Najaf Iraq; Geospatial modeling; Hydrogeology; Iraq; Groundwater; Machine learning
• ISSN: 1520-7439; 1573-8981
• DOI: 10.1007/s11053-019-09561-x

This study assessed the groundwater productivity of the Dibdibba aquifer on the Karbala–Najaf Plateau, central Iraq, using three GIS-based tree machine learning classifiers, namely classification and regression trees (CART), rotation forest (rF), and random forest (RF). The geographical locations of groundwater wells with a specific well capacity and eight groundwater productivity conditioning factors were used to build the three models. The factors used involved four related topographical factors (elevation, slope, curvature, and aspect), two aquifer hydraulic characteristics (transmissivity and specific yield), and two proximity factors (distance to Abu Jir Fault and distance to faults). The multi-collinearity preliminary test indicated that there was no multi-collinearity issue among the factors used. Investigating the powerful factors of the building models using the information gain ratio showed that five factors (distance to Abu Jir Fault, specific yield, transmissivity, elevation, and distance to faults) played a major role in controlling aquifer productivity. The five most important factors were used to build the models. The performance of the models was compared in the training and testing stages using three evaluation metrics: accuracy, kappa, and area under the relative operating characteristic curve. Applying the models showed that RF was the best classifier, followed by rF and then CART. The probability predictions of the models were interpolated and classified into three categories of aquifer productivity: low, moderate, and high; these classes encompassed approximately 23%, 8%, and 69% of the study area, respectively. The total sum of the moderate and high productivity classes was 75% of the study area; therefore, the study area was promising in terms of its water yield. The maps derived from this study could be used by hydrogeologists and decision makers as a guide to develop the studied aquifer by setting rules for a pumping scheme, particularly in areas where the productivity is low.