Baseline v0
Computed 1 November 2026 · reproducible from the public endpoint below
What it is
Baseline v0 is a density prior over a regular grid. It is the simplest ranking we could publish honestly, and it exists so that every later model has a public number to beat. It is not a prediction and not the deep-learning model, which is in development.
Data
Geological Survey of NSW mineral occurrence records, feature type erl:MineralOccurrenceView, fetched over WFS 2.0 from gs.geoscience.nsw.gov.au/geoserver/ows for the window 146.5–150.2°E, 31.5–35.0°S. 8,031 records; 3,914 carry a copper, gold, silver or molybdenum commodity, and 133 copper and 768 gold records are typed as deposit. Each record has a point location, a commodity list and a type, deposit or occurrence, assigned by the survey. Licence CC BY 4.0.
Score
w_type(r) × w_commodity(r)
w_type: deposit = 3 · occurrence = 1
w_commodity: copper = 1.0 · gold = 1.0 · molybdenum = 0.7 · silver = 0.5 (max over the record's commodities)
cells: 0.1° × 0.1° (1,295 in the window); scores normalised to the maximum
Records inside a cell never contribute to that cell's score, so a cell is ranked only on its surroundings. Records in neighbouring cells within 15 km still count, and deposits cluster, so this is a weak check rather than a hold-out. With that caveat we can ask whether highly ranked cells happen to contain a copper or gold deposit record.
Baseline check
| Measure | Value |
|---|---|
| Cells scored | 1,014 of 1,295 |
| Top-20 cells containing a Cu or Au deposit record | 9 of 20 |
| All cells containing a Cu or Au deposit | 19.5% |
This is a baseline check, not model performance. A density prior ranks the historical goldfields around West Wyalong and Temora first, because that is where the records are densest. It cannot find ground nobody has worked, which is the job of the model in development.
Limitations
- Occurrence records reflect where people looked, not where mineralisation is.
- Positional accuracy varies by record, from 50 m to 250 m and worse for old map-based entries.
- Commodity weights are hand-set and round; they are not fitted.
- Each record carries the commodities the survey lists for it; no record in the window lists both copper and gold, so copper-gold deposits are counted under one metal (Northparkes is recorded as copper).
- No geophysics, geology or imagery enters Baseline v0.
Reproduce it
- Request the feature type above with the window as a bounding box in EPSG:4326 and
outputFormat=application/json. - Keep records whose commodity list contains copper, gold, silver or molybdenum.
- Lay a 0.1° grid over the window and compute the score as defined.
- Rank cells and compare the top 20 against records typed as deposit with copper or gold.
Questions about the method: tuong@lanhaigeo.com