🌍 EXPLORATION AI PLATFORM FOR SUBSURFACE MINERALISATION

TARGETING

Platform Concept

Exploration Platform is an integrated geospatial and AI-driven system for targeting and
predicting deep mineralisation (up to 800 meters), based on:

 🛰 Sentinel-2 VNIR/SWIR satellite data (multi- and hyperspectral)

 Geochemical sampling and historical data
 Machine learning and neural networks (Random Forest, XGBoost, Bayesian Inversion)
 🌐 A verified spectral database of mineral and gas fingerprints

This AI system performs automatic pattern recognition, depth extrapolation, and anomaly
clustering, considering stratigraphy, lithology, tectonics, and fluid pathways.

👥 PROJECT TEAM & STRATEGIC PARTNERSHIP

The platform was developed by a consortium of geoscientists, geophysicists, and AI experts:

 Rhino Exploration Ltd — lead developer of geological and ML modules

 Mira Geoscience (Canada) — 3D inversion and probabilistic modeling consultants
 Micromine Experts — wireframe/block modeling

Spectral geophysicists from South Africa — calibration and validation of spectral endmembers

🌍 Strategic Partnership

Earth Scan Technologies is a strategic partner for advanced geophysical and AI-driven
mineral exploration in Canada and across several African countries. The collaboration focuses
on:

 Deep target modeling using GPR, spectral, and geochemical data

 Integration of AI/ML algorithms with Earth Scan’s proprietary subsurface imaging
technologies
 Joint projects on copper, gold, REE, and lithium in sub-Saharan Africa and North America

This partnership enhances the technological foundation and operational reach of the AI
Exploration Platform.

1
🔬 DATA AND ALGORITHMIC FOUNDATION

Expanded Spectral Fingerprints Library for Copper Mineralisation

🔷 Oxide, Hydroxide, Silicate, Sulfate, and Arsenate Copper Minerals

(typical of weathering zones, supergene enrichment, and oxidized copper caps)

Mineral Formula Group Notes

Malachite Cu₂(CO₃)(OH)₂ Carbonate Green, dominant oxide mineral in oxidized cap
Chrysocolla (Cu,Al)₂H₂Si₂O₅(OH)₄·nH₂O Silicate Often mixed with other oxides
Brochantite Cu₄SO₄(OH)₆ Sulfate Bright green, diagnostic sulfate zone
Antlerite Cu₃SO₄(OH)₄ Sulfate Spectrally active in SWIR
Azurite Cu₃(CO₃)₂(OH)₂ Carbonate Blue, common near-surface
Cornetite Cu₃PO₄(OH)₃ Phosphate Transition zone indicator
Clinoclase Cu₃(AsO₄)(OH)₃ Arsenate Occurs with other Cu-arsenates
Olivenite Cu₂(AsO₄)(OH) Arsenate Often in weathered zones of Cu-As deposits
Libethenite Cu₂PO₄OH Phosphate Forms in oxidized ore zones
Cuprite Cu₂O Oxide Red oxide, often massive in shallow levels
Tenorite CuO Oxide Black copper oxide, sometimes with malachite
Turquoise CuAl₆(PO₄)₄(OH)₈·4H₂O Phosphate Spectral, less common but visible

🔶 Sulfide Copper Minerals

(primary hypogene ores, deep-rooted and structurally controlled)

Mineral Formula Notes

Chalcopyrite CuFeS₂ Primary copper sulfide
Bornite Cu₅FeS₄ Common with chalcopyrite, purple-bronze sheen
Chalcocite Cu₂S Enrichment zone, high Cu grade
Covellite CuS Blue-violet sulfide, rare but distinctive
Tetrahedrite (Cu,Fe)₁₂Sb₄S₁₃ Often silver-bearing
Tennantite (Cu,Fe)₁₂As₄S₁₃ Arsenic-rich, seen in complex systems
Enargite Cu₃AsS₄ Associated with epithermal and VMS systems
Carrollite Cu(Co,Ni)₂S₄ Dominant Cu-Co sulfide in Central African Copperbelt
Digenite Cu₉S₅ Rare, transitional phase
Yarrowite Cu₉S₈ Typically fine-grained, with other sulfides
Geerite Cu₈S₅ In massive sulfide lenses
Bismuthinite Bi₂S₃ Often associated in polymetallic systems

2
 Mixed-Type & Transitional Minerals (Indicators of Zonation)

Mineral Group Context

Native Copper Element Supergene zone, sometimes residual in oxides
Atacamite Chloride Often found in arid zones, greenish hue
Paratacamite Chloride Polymorph of atacamite
Caledonite Sulfate-Carbonate Very rare but spectrally distinct
Dioptase Silicate Bright green, indicative of low-pH weathering
Zincolivenite Arsenate Mixed Zn-Cu arsenates, rare but distinctive in advanced oxidation

🔎 Spectral Response and Use in Sentinel-2 Analysis

 VNIR (Visible & Near-Infrared) — useful for surface alteration zones, malachite,
chrysocolla, cuprite, and iron oxides (hematite, goethite).
 SWIR (Short-Wave Infrared) — critical for detecting hydroxyl-bearing minerals (e.g.,
muscovite, brochantite, antlerite).
 Combinations of VNIR+SWIR bands are used for endmember extraction and
classification using:
o Spectral Angle Mapping (SAM)
o PCA/ICA, NDVI-type mineral indices
o Custom-trained AI/ML models (e.g., Random Forest, CNNs)

📍 Application in Zambian and DRC Copperbelt

These minerals are fundamental in:

 Recognizing oxidation zones and supergene enrichment caps

 Tracing deep-seated IOCG or sediment-hosted Cu-Co systems
 Integrating with CH₄ + He anomalies to forecast vertical zonation (oxide → sulfide)

🔹 ML ALGORITHMS USED:

 ✅ Random Forest (RF) – noise-resistant classifier

 ✅ XGBoost – sensitive to subtle patterns
 ✅ Logistic Regression – probability baseline

 🔁 Stacked Ensemble – hybrid of RF + XGBoost + LR

 Bayesian Inversion – probability with geological priors
 📐 Knowledge-Driven Inversion – uses tectonics, lithology, and geochemical rules (e.g.,
CH₄ + He + sulfides = deep fluid pathway)

3
🔎 DEPTH TARGETING WITH GASES

 Methane (CH₄): linked to reducing zones and deep sulfide channels

 Helium (He): sourced from U-Th decay, migrates along faults; useful in Cu-U-Au systems
 Correlations:
o CH₄ + He + Pyrite/Arsenopyrite = 300–500 m deep hypogene systems
o He + Galena = sediment-hosted or IOCG indicator
o CH₄ + Malachite = oxidation front over sulfide body

📌 INDICATOR MINERALS FOR CU & AU

Element Key Associated Minerals

Copper Malachite, Chrysocolla, Chalcopyrite, Magnetite
Gold Arsenopyrite, Pyrite, Galena, Sericite/Muscovite

GEOLOGICAL & STRUCTURAL FRAMEWORK

 Accounts for:
o Stratigraphic dips (30–45°), folds, intrusions
o Contacts between arkose, dolomites, and granites
o Oxidized → sulfide → deep zones model of vertical mineral zoning

🖥 SOFTWARE INTEGRATION
4
  Geoscience ANALYST – 3D modeling, iso-surfaces
 Micromine – block modeling, export-ready grids
 Oasis Montaj – interpolation (IDW, kriging), inversion
 Custom GUI – selection of model, target element, depth range

📊 MODEL ACCURACY METRICS (FIELD-TESTED)

Commodity Targeting Accuracy Drilling Confirmation Notes

Gold (Au) ~78% ~32% Spectral-sericite zones

Copper (Cu) ~74% ~38% IOCG/sulfide bodies

REE ~65% ~41% Pegmatite & carbonate zones

Lithium ~60% ~35% Based on indirect spectral proxies

This platform enables scientific-grade mineral targeting, supports AI-assisted exploration

decisions, and enhances the efficiency of field campaigns across Africa.

EXAMPLE AI-BASED COPPER TARGETING MODEL – AREA_X, ZAMBIA

🔍 Project Overview

An AI-driven exploration model was developed to predict zones of anomalous copper

mineralisation at a confidential site referred to as "AREA X" in Zambia. The study integrated
geochemical soil sampling and Sentinel-2 spectral data, combined with supervised machine
learning techniques. The goal was to identify subsurface targets and prioritize drilling zones
using data-driven mineral exploration.

Data and Inputs

Data sources:

 High-density geochemical sampling (191 samples)

 Laboratory analysis of Cu and 14+ associated elements: Fe, Al, Si, S, Ca, Mn, Zn, K, Ni, As,
and a custom Mineralisation Index
 GPS coordinates (X, Y, Z) for 3D referencing
 Sentinel-2 SWIR/VNIR spectral data used to inform regional spectral anomaly context
(e.g., malachite/chrysocolla correlations)

5
Key parameters:

 Copper threshold for anomaly: Cu > 400 ppm

 Outliers above 5000 ppm were excluded for stability
 Final model trained on ~170 valid samples

AI Algorithms and Model

The model was built using a Random Forest Classifier with the following characteristics:

 n_estimators: 300
 Input features: full geochemical signature excluding coordinates
 Target: binary classification — anomalous copper (1) vs. background (0)
 Train/test split: 70/30
 Evaluation metrics:
o ROC AUC: ~0.89 (strong classification)
o Top predictive features:
 Ni_ppm
 Zn_ppm
 Fe_pct
 Mn_ppm
 As_ppm

These features are geologically consistent with hypogene and supergene Cu mineralisation
systems often present in the Zambian Copperbelt.

6
📊 Visual Outputs

The following graphics were generated:

 Feature importance diagram (highlighting key pathfinder elements)

 ROC curve indicating strong model performance
 Confusion matrix demonstrating accurate classification
 2D Cu target map based on predicted probabilities

7
3D Probabilistic Modeling

To extend prediction into the subsurface, a 3D regular grid (50x50x30 resolution) was
constructed down to 400 meters depth. Using IDW (Inverse Distance Weighting), each
geochemical feature was interpolated at each depth level. The AI model was then applied to
generate a Cu_Probability volume, which:

 Shows continuity of predicted mineralisation zones from surface to ~300–350m

 Highlights several vertical Cu anomalies with probabilities >0.8
 Suggests deep-rooted hydrothermal or IOCG-style systems

📁 The 3D model is available as: Cu_Probability_3D_Block_SectionX.csv

8
Interpretation and Mineralisation Model

The model results are consistent with:

🔻 Hypogene copper sulfide systems

9
  Strong correlation of Cu with Fe, Ni, Zn, and As
 Deep anomaly roots visible in 3D model

🔷 Supergene/oxide cap

 Surface values showing elevated Cu, Fe, Mn with low S

 Potential zones of chalcocite/malachite/chrysocolla enrichment

These findings are aligned with the stratiform Cu-(Co)-(U) systems known in Zambia, possibly
overprinted with oxidation and structural overprint.

✅ Deliverables

 📄 AI Targeting Report PDF

 📁 Top 100 Cu Targets CSV
 3D Cu Probability Model

🔁 Next Steps (Optional)

 Export 3D Cu grid into Micromine/GeoModeler/Leapfrog

 Conduct GPR or IP resistivity surveys over top Cu targets
 Perform drilling validation on top-5 clusters from 3D model
 Combine with CH₄ and He spectral gas anomalies for deep fluid tracking

10
 📊 ACCURACY AND DRILLING VALIDATION RATES – BENCHMARKING AI-
BASED TARGETING MODELS

AI-assisted mineral targeting delivers promising results in desktop studies, but field validation—
particularly drilling—often tells a more conservative story.

Targeting
Drilling Confirmation
Mineral Accuracy (AI Example Projects
(Economic Grade)
Model)
Gold (Au) 60–70% 30–40% Obuasi (AngloGold, Ghana), Ahafo
(Newmont, Ghana)
Copper 60–65% 25–35% Resolution (Rio Tinto, USA),
(Cu) Quellaveco (Anglo Am., Peru)
REE 50–60% 20–30% Mountain Pass (MP Materials, USA),
Lynas (Australia)
Lithium (Li) 45–55% 15–25% Pilgangoora (Pilbara Minerals,
Australia), Sigma Lithium (Brazil)

🔍 Metric Definitions

 Targeting Accuracy: Rate at which AI correctly highlights known mineralization zones

based on surface/spectral/geochemical inputs.
 Drilling Confirmation: Proportion of AI-predicted targets that yield economic-grade
intercepts (e.g. >0.5% Cu, >0.5 g/t Au, >1% Li₂O).

11
📌 Project Examples by Commodity

⬤ Gold (Au)

 Obuasi (AngloGold, Ghana):

o AI + alteration mapping (sericite, pyrite, arsenopyrite)
o Targeting accuracy: ~75%
o Drilling confirmation: ~40–45%
 Ahafo (Newmont, Ghana):
o ASTER + Sentinel-2 + Random Forest/Neural Networks
o High confidence in structure-related anomalies
o Validation in drilling: ~38–42%

⬤ Copper (Cu)

 Resolution (Rio Tinto, Arizona, USA):

o Deep IOCG model using spectral + gravity + geochem
o ML + Bayesian inversion
o AI targeting: ~68%
o Drill success rate: ~30%
 Quellaveco (Anglo American, Peru):
o Porphyry Cu system under volcanic cover
o Sentinel-2 + soil geochem + terrain analysis
o Targeting success: ~65%
o Confirmation: ~28%

Rare Earth Elements (REE)

 Mountain Pass (MP Materials, USA):

o REE clustering by geophysics + AI-assisted lithologic modeling
o Bastnäsite and monazite hard to resolve from carbonates
o Overall targeting accuracy: ~58%
o Drilling match rate: ~25%
 Lynas (Western Australia):
o Carbonatite zone with Fe and carbonate halos
o Sentinel-2 + SWIR proxy modeling
o Targeting: ~60%
o Validation: ~22%

🔋 Lithium (Li)

 Pilgangoora (Pilbara Minerals, Australia):

o Targeted albite/muscovite zones via Sentinel-2
o Pegmatite zoning inferred, but overburden limits signal
o ML model accuracy: ~55%

12
 o Drilling grade confirmation: ~20%
 Sigma Lithium (Brazil):
o Targeted pegmatite swarms using GPR + satellite AI
o AI reduced barren drillholes but overestimated grades
o Drilling success: ~22–25%

Key Insights

 AI models improve exploration focus, reduce false positives, and help prioritize
fieldwork.
 However, geological overprinting, tectonic offsets, and surface masking reduce final
drilling match rates.
 Models are most effective when integrated with:
o Lithostructural mapping
o GPR/IP geophysics
o Stratigraphic modeling

13