GeaSpirit Platform — Mineral Intelligence System

// PILOT RESULTS — ABLATION STUDY

Chuquicamata Performance

MODEL PERFORMANCE — ABLATION STUDY (spatial block CV)

Experiment	Bands	AUC	Precision	Recall	F1
Phase 3 baseline	19	0.6844	0.606	0.284	0.345
A: S2 only	5	0.7325	0.803	0.880	0.822
B: Full satellite	19	0.8530	0.933	0.764	0.833
C: Geology only	5	0.7356	0.751	0.738	0.685
D: S2 + geology	10	0.8094	0.850	0.897	0.862
E: Full fusion	24	0.8622	0.948	0.760	0.837

MODEL CALIBRATION

Metric	Before	After
Brier Score	0.1955	0.1711
ECE	0.1446	0.0000

Brier score: lower = better calibrated probabilities.
ECE (Expected Calibration Error): 0.0000 = perfectly calibrated.
When the model says "60% chance of mineral", it means exactly 60%.

TOP 10 FEATURE IMPORTANCE — WHAT FINDS MINERALS

#	Feature	Importance	Source
1	Terrain ruggedness	18.6%	DEM
2	Elevation	12.1%	DEM
3	SAR VH backscatter	10.9%	Sentinel-1
4	Ferrous iron index	7.4%	Sentinel-2
5	Thermal z-score	7.0%	Landsat
6	Thermal P90	5.9%	Landsat
7	Clay/hydroxyl	4.5%	Sentinel-2
8	SAR VV	3.9%	Sentinel-1
9	SAR texture	3.8%	Sentinel-1
10	Iron oxide	3.3%	Sentinel-2

TRAINING DATA — PHASE 3B (CURATED)

Curated deposits (Cu/Au/Ag)	43 (from 152 raw MRDS)
Positive pixels	33,428
Geology-aware negatives	14,483 (random + hard + matched)
Image resolution	1856 x 1857 px · 24 bands · 30m
Sensors	Sentinel-2 + SAR + DEM + thermal + Macrostrat geology
Area covered	~50 x 50 km
Validation	Spatial block CV (10km blocks, 5 folds)

RESEARCH — PHASE 3B KEY DISCOVERY

The #1 improvement factor was LABEL CURATION, not more data.

Phase 3 trained on 152 deposits including limestone, dolomite, silica, and boron — geological noise that confused the model. Phase 3B curated to 43 real Cu/Au/Ag metal deposits. Same satellite data, same algorithm. Result:

AUC: 0.6844 → 0.8530 (+0.1686)

"Clean labels matter more than fancy sensors."

WHAT EACH LAYER CONTRIBUTES:

Satellite only (19 bands):    AUC 0.8530
+ Geology (Macrostrat):      AUC 0.8622 (+0.009)

Geology adds real but marginal signal. Macrostrat is too coarse for Chile (1 lithology per 50x50km). Finer geological maps will add more.

COMPARISON WITH INDUSTRY:

Random guessing:          0.50 AUC
Phase 3 (noisy labels): 0.68 AUC
Phase 3B (curated):       0.86 AUC ← HERE
Goldspot ($50M+):         0.85-0.93 AUC
KoBold ($3B+):            not published

GeaSpirit now matches entry-level commercial systems using only public satellite data.

HONEST CAVEATS:
⚠ Class balance changed (1:3 → 2.3:1) — amplifies AUC delta
⚠ Only 43 curated deposits — small sample, may not generalize
⚠ Cross-zone validation incomplete (Pilbara/Zambia pending)
⚠ Macrostrat geology too coarse for Chile — need SERNAGEOMIN

ALL 24 FEATURES (PHASE 3B)

Sentinel-2 (5)	Iron Oxide · Clay/Hydroxyl · Ferrous Iron · Laterite · NDVI
Sentinel-1 SAR (5)	VV · VH · VV/VH ratio · GLCM variance · GLCM contrast
DEM (6)	Elevation · Slope · sin(Aspect) · cos(Aspect) · TPI · Ruggedness
Landsat thermal (3)	Median LST · P90 LST · Thermal z-score anomaly
Macrostrat geology (5)	Lithology code · Group · Geological age · Distance to contact · Availability

// TARGET DISCOVERY — PHASE 3B MODEL

Exploration Targets

DISCOVERY ENGINE — GEOLOGY-AWARE MODEL

50 exploration targets ranked above 0.6 probability threshold
using the Phase 3B geology-aware model (AUC 0.8622).

Higher-discrimination model finds more candidates than Phase 3.
All targets are outside a 5km exclusion buffer from known deposits.
Target discovery uses calibrated probabilities — when we say 60%, we mean exactly 60%.
Field validation required to confirm — many candidates may be false positives.

// NEXT OPERATION — PILBARA, AUSTRALIA

Next Target: Pilbara

WORLD-CLASS OPEN GEOPHYSICS — ALL CC-BY 4.0

■ 80m aeromagnetics ........... Geoscience Australia (TMI anomaly grid)
■ 100m radiometrics ........... Geoscience Australia (K, Th, U dose rate)
■ 800m gravity ................ Geoscience Australia (Bouguer anomaly)
■ 1:100K geological maps ...... GSWA / DMIRS (lithology + faults)
■ Known iron + gold deposits .. MRDS + Geoscience Australia databases

Pilbara has the richest open geophysics dataset in the world.
Adding airborne magnetics and radiometrics to the satellite stack could push AUC beyond 0.90 for Pilbara specifically.

// TECHNOLOGY STACK

Systems Online

■ Sentinel-2 L2A ............ Spectral mineralogy (5 indices)
■ Sentinel-1 SAR ............ Radar structure + texture
■ Copernicus DEM GLO-30 ..... Terrain ruggedness
■ Landsat 8/9 thermal ....... Temperature anomalies
■ Macrostrat geology ........ Lithology + rock age
■ Google Earth Engine ....... Cloud processing
■ USGS MRDS (curated) ....... 43 Cu/Au/Ag deposits
■ XGBoost ................... Gradient boosting classifier
■ Spatial Block CV .......... Honest 10km block validation
■ Isotonic calibration ...... Probability calibration
■ Ablation study ............ Per-layer contribution analysis
□ GA aeromagnetics .......... PENDING (Phase 4 — Pilbara)
□ GA radiometrics ........... PENDING (Phase 4 — Pilbara)
□ EMIT hyperspectral ........ PENDING (Phase 5)
□ SERNAGEOMIN geology ....... PENDING (finer Chile maps)
□ Cross-zone LOZO ........... PENDING (Phase 5)

// ROADMAP

Development Phases

PHASE 0 Setup + GEE auth + environment COMPLETE

PHASE 1 Chuquicamata pilot — S2 only AUC 0.73

PHASE 2 Multi-source fusion — 19 bands COMPLETE

PHASE 3 Honest spatial CV AUC 0.68

PHASE 3B Label curation + geology AUC 0.86

PHASE 4 Pilbara pilot + GA open geophysics NEXT

PHASE 5 EMIT hyperspectral + cross-zone LOZO PLANNED

PHASE 6 Subsurface probability inference PLANNED

// DETECTION CAPABILITIES

What We Can Detect

DETECTABLE NOW (surface)

✓ Iron oxides (hematite, goethite)

✓ Clay minerals (kaolinite, montmorillonite)

✓ Ferrous iron bearing rocks

✓ Laterites and gossans

✓ Vegetation presence/absence

✓ Terrain structural anomalies

✓ Thermal surface anomalies

✓ Lithology classification (coarse)

COMING SOON (Phase 4-5)

◌ Aeromagnetic anomalies (Pilbara)

◌ Radiometric alteration (K/Th/U)

◌ Precise mineral species (hyperspectral)

◌ Cross-zone transfer learning

◌ Vegetation stress from metals

FUTURE GOAL (Phase 6)

○ Subsurface mineral probability

○ "Where to dig" recommendations

○ Depth estimation via proxy fusion

○ Global mineral intelligence network

// DATA SOURCES — OPEN PROGRAMS

Credits & Sources

▸ Sentinel-2 .... European Space Agency (ESA) — Copernicus Programme
▸ Sentinel-1 .... European Space Agency (ESA) — Copernicus Programme
▸ Landsat 8/9 ... NASA / USGS — Landsat Programme
▸ Copernicus DEM European Space Agency (ESA) — GLO-30
▸ MRDS .......... U.S. Geological Survey — Mineral Resources Data System
▸ Macrostrat .... University of Wisconsin — Geological map API (CC-BY 4.0)
▸ Processing .... Google Earth Engine — Non-commercial Community Tier

Data from ESA Copernicus, NASA, USGS, and Macrostrat open programs.