NoraResearchLab/mineral-prospectivity-grid-na

North America Structural Favorability Grid

A continental-scale, validated mineral prospectivity layer derived from the DS-424 GeoParquet dataset. Each grid point carries a structural favorability score based on proximity to mapped faults and geologic contacts — a classic targeting signal for structurally-controlled mineralization.

Quick Links

• 📦 Dataset on Hugging Face
• 🧑‍💻 Source code & pipeline on GitHub
• 🗺️ Source dataset: DS-424 GeoParquet
• 🏢 NORA Research Lab on Hugging Face

Methodology

For each point on a 10km grid across the mapped extent of North America:

1. Distance to the nearest mapped fault and nearest geologic contact is computed
2. A composite structural favorability score is derived via inverse-distance decay (exp(-distance / 5km)), min-max normalized to [0, 1]
3. The score is validated against ~real, independent data: USGS Mineral Resources Data System (MRDS) known occurrence locations

Validation Results

Known mineral occurrences were snapped to the nearest grid cell and binned by that cell's score decile (grid-defined boundaries, not deposit-defined — avoiding circularity):

Score Decile (9 = highest)	Share of Known Deposits
9	40.3%
8	23.7%
7	17.1%
6	9.2%
5	5.4%
4	2.5%
3	1.3%
2	0.4%
1	0.03%

40.3% of known deposits fall within the top-scoring 10% of the grid — roughly a 4x lift over the 10% expected under no relationship, with a clean monotonic decay across deciles. This confirms structural proximity alone carries real, measurable predictive signal for known North American mineral occurrences at 10km resolution.

Files

• structural_favorability_grid.parquet — full scored grid (point geometry, distance to fault/contact, structural score, underlying geologic unit attributes)
• validation_capture_rate.csv — decile-level validation results shown above

Usage

import geopandas as gpd

grid = gpd.read_parquet("structural_favorability_grid.parquet")
top_targets = grid[grid["structural_score"] > grid["structural_score"].quantile(0.9)]

Limitations

• Structural proximity alone — does not yet incorporate lithology, tectonic setting, or geochemistry, which would likely sharpen the signal further
• 10km grid resolution — coarser than typical field-scale exploration targeting, intended as a regional screening layer, not a drill-target generator
• MRDS occurrence data has not been systematically updated since 2011 (per USGS); recent discoveries are not reflected

License & Provenance

Derived from public-domain USGS data: DS-424 (Geologic Map of North America) and MRDS (Mineral Resources Data System). This derived product is released under CC0.