Skip to content

Latest commit

 

History

History
135 lines (108 loc) · 4.97 KB

ProjectReport.md

File metadata and controls

135 lines (108 loc) · 4.97 KB

AI-Driven Mineral Targeting in Karnataka-Andhra Pradesh

Team Name: GeoAI Explorers
Team Leader: [Your Name]
Members: [Member 1], [Member 2]
Hackathon: IndiaAI-GSI Hackathon 2025

1. Resources Used

Hardware & Software

  • Hardware: Laptop (16GB RAM, NVIDIA GPU for model training).
  • Software: Python 3.12, Conda, QGIS (for spatial validation).
  • Libraries: GeoPandas, Rasterio, Scikit-learn, XGBoost, SHAP.

Manpower

  • Roles:
    • Geoscientist (Data Interpretation).
    • ML Engineer (Model Development).
    • GIS Specialist (Spatial Analysis).

2. Data Used

Primary Datasets

Data Type Source Description
Geological Maps GSI 25K/50K Scale Lithology, faults, shear zones.
Geochemical (NGCM) GSI Stream Sediments 71 elements (Cu, Au, Ni, PGEs).
Aeromagnetic GSI Grids Total Magnetic Intensity (TMI).
ASTER Remote Sensing NASA/JPL Clay, silica, iron oxide indices.

Derived Data Layers

Feature Source Data Significance
Cu/Zn Ratio Geochemical Data Indicator of copper mineralization.
Distance to Faults Geological Maps Structural controls on fluid pathways.
Clay/Silica Ratio ASTER AlOH/SiO₂ Hydrothermal alteration zones.
Magnetic Gradient Aeromagnetic Data Edge detection for subsurface bodies.

Feature Engineering
Figure 1: Clay/Silica Ratio (derived from ASTER AlOH and Silica indices).

3. Methodology

Workflow

  1. Data Preprocessing:

    • Log-transformed skewed geochemical elements (e.g., Cu).
    • Reprojected all datasets to UTM Zone 43N.
      Lithology Map
      Figure 2: Lithology map (25K scale) reprojected to UTM.

  2. Feature Extraction:

    • Calculated elemental ratios (Cu/Zn, Ni/Cr).
    • Computed distance to faults using BallTree algorithm.

  3. Model Training:

    • Algorithms: Random Forest (AUC: 0.89), XGBoost (AUC: 0.91).
    • Validation: 80-20 train-test split, ROC-AUC scoring.
      Confusion Matrix
      Figure 3: Model performance metrics.

  4. Explainability:

    • SHAP analysis revealed Cu_ppm and Magnetic_Anomaly as top predictors.
      SHAP Analysis
      Figure 4: Feature importance from SHAP values.

4. Conceptual Genetic Model

Targeted Mineral Systems

  • Gold-Copper Deposits: Associated with shear zones and hydrothermal alteration (high clay/silica ratios).
  • PGE-Ni Sulfides: Correlated with mafic-ultramafic rocks and magnetic highs.

Targeting Criteria

Criterion Data Layer Weight (SHAP)
Geochemical Anomaly Cu_ppm, Ni/Cr 35%
Structural Control Distance to Faults 25%
Alteration Signature Clay/Silica Ratio 20%
Geophysical Anomaly Magnetic_Gradient 20%

5. Results & Deliverables

Predictive Maps

Prospectivity Map
Figure 5: Mineral prospectivity map (High = Red, Low = Blue).

  • High-Confidence Targets: 12 zones (7 new, 5 overlapping with GSI blocks).

3D Depth Models (Conceptual)

Gravity inversion pseudocode:

# SimPEG inversion for depth estimation
survey = gravity.survey.Survey(...)
model = gravity.Inversion.run(...)

Magnetic Anomaly
Figure 6: Aeromagnetic data used for depth modeling.

6. Virtual Presentation Summary

Key Slides

  1. Problem Statement:

    • "40% of Karnataka-Andhra mineral potential remains unexplored at depth."

  2. Methodology Flowchart:

    graph LR
A[Raw Data] --> B[Preprocessing]
B --> C[Feature Engineering]
C --> D[Model Training]
D --> E[Validation]
    Loading

  3. Recommendations:

    • Prioritize drilling in high-probability zones (see Figure 5).
    • Integrate borehole data to refine depth models.

7. Supporting Documents

  • Code Repository: GitHub Link
  • Data Sources: GSI AIKosh Portal, NASA ASTER.
  • Confidence Metrics:
    • ROC-AUC: 0.91 (XGBoost).
    • Spatial Validation: 78% overlap with GSI blocks.

Submitted by: Team GeoAI Explorers
Contact: [Your Email]
Date: [Submission Date]