| Abstract Submission No. | ABS-4.1-0123 |
| Title of Abstract | Translation of AEM resistivity into lithological model using Random Forest Classifier |
| Authors | PEELA MADHU*, PAYAL RANI, AKHIL MISHRA, VED P MAURYA, SUBASH CHANDRA |
| Organisation | CSIR-NATIONAL GEOPHYSICAL RESEARCH INSTITUTE, Hyderabad; AcSIR |
| Address | II/32, NGRI STAFF QUARTERS, NGRI COLONY, CSIR-NGRI, HYDERABAD Hyderabad, TELANGANA, India Pincode: 500007 E-mail: madhupeela1999@gmail.com |
| Country | India |
| Presentation | Oral |
| Abstract | There is an increasing trend of integrated airborne electromagnetic (AEM) surveys for high resolution aquifer mapping worldwide, including India over the past few decades. Translation of AEM resistivity data into lithological and hydrogeological parameters is an essential part of aquifer mapping, which is mostly carried out manually by comparing AEM resistivity with borehole data. The AEM surveys provide large scale resistivity data, whereas borehole data provide precise lithological details but at a point scale in a discrete manner. Besides the large size of the AEM data, the inherent issue of non-unique resistivity inversion and complex resistivity�lithology relationship, translation of resistivity into meaningful lithological and hydrogeological models is a challenging task. These difficulties can be addressed by a machine learning approach. In this study, we use machine learning techniques for translating AEM resistivity data into lithological and hydrogeological models. The approach involves pre-processing and statistical analysis of AEM resistivity data, followed by correlation with borehole lithology and finally supervised machine learning (Random Forest Classifier) based classification of lithology in terms of sand and clay probabilities. The term hydrofacies is used to represent strata having uniform hydrogeological properties. The Aquifer which has fine, medium, coarse sands and gravel comes under sand hydrofacies class, whereas clay and silt are considered as clay hydrofacies class. This study provides reliable hydrofacies prediction as Sand and Clay Probabilities even in data-sparse regions. Application of this methodology in the alluvial formation of the Middle Ganga Plain in Ganga-Yamuna Doab region, demonstrates its capability to delineate permeable and impermeable layers, aquifer boundaries, and provide vertical and lateral heterogeneities with higher accuracy than the conventional interpretation methods. This study highlights the role of machine learning techniques in providing an automated and accurate translation between geophysical measurements and hydrogeological parameters through practical case studies. |
| Keywords | Electromagnetic, Resistivity, lithology, Hydrofacies, Supervised machine Learning, Random Forest Classifier |