mineral exploration as a complex global challenge with...
TRANSCRIPT
Mineral Exploration as a Complex Global Challenge
with Numerical Modelling Solutions
Jean-Marc LulinAzimut Exploration Inc.
7th International Megaprojects Workshop: Theory meets Practice
Artificial Intelligence and Megaprojects
June 13, 2019
Global Mineral
Exploration
1) A complex challenge with low success rates
2) Advanced data processing as a potential breakthrough
3) Successes and pitfalls: Simplexity versus complexity
Azimut in Quebec, Canada
• Core business since 2003: Big Data analytics applied to mineral exploration alongside partnership development
• 31 partnership agreements, including Rio Tinto (3), Goldcorp (2), IAMGOLD (2), Hecla Mining (2) and SOQUEM (2) for a total of~$140 million in work commitments
• Discovery of 400+ mineral prospects as a direct result of Azimut’sproprietary targeting methodology (AZtechMineTM)
• Holder of the largest exploration portfolio in the province
• Quebec: One of the top mining jurisdictions in the world
What is Mineral Exploration Today?
• Global activity
• US$10.1 billion budget in 2018 (non-ferrous exploration)
• 3,300 companies: major, intermediate, junior, parastatal Source: S&P Global
A Complex Global Challenge
1) New mineral deposits must be discovered to sustain a modern world
2) Discovery chances are decreasing in mature mining districts
3) Significant potential remains in known districts but at greater depths, and in
remote and/or politically risky regions of the world
A multiparameter task involving many interrelated factors
that are not easy to predict
➢ Mineral deposits: Small objects unevenly distributed within the Earth’s crust:
Size, grade, 3D location
➢ Mid- to long-term metal prices
➢ Technology-driven needs
➢ Political, legal, fiscal, societal, environmental factors in
multiple jurisdictions
A Complex Global Challenge
1) Universal paradigm: Use of geoscientific databases in exploration, one of the main drivers for discoveries
2) Explosive growth of data production and availability, but declining discovery rates
…A paradox!
Two possible combined explanations:
➢ Increasing global maturity
➢ Too much data producing too many targets, without theright selection tools: Quantity is the enemy of quality
Mineral Exploration: A Global Challenge
Exploration success rates
1,000
1,000100
100
10
10
1
1
Miracle method 1 target:1 discovery
Oil industry 1:10
Mining industry
1:1,000
Number of
targets
Number of
discoveries
Poor use of too much data
= too many targets
= poor success rates
_________________________
Advanced data processing
= fewer but higher quality targets
= dramatically improved success rates
All rights reserved
Advanced Data Processing: A New Exploration Paradigm
1) Accurate data processing can significantly improve the discovery rates
2) Selecting the right targets at the initial exploration stage reduces the technical and financial risks
Example of Quebec
➢ One of the best geoscientific digital databases worldwide, covering almost the entire province – coherent, reliable, accessible
➢ Azimut conducted systematic predictive modelling for selected mineral deposits using proprietary big data expert system AZtechMineTM
Advanced Data Processing applied to Mineral Exploration
Type of Processing
Innovative statistical approach linking regional-scale parameters to a database of mineral prospects and deposits to extract a reliable footprint for selected deposit types
Quebec-scale processing: 87.5 million pixels; cell size: 200 x 200 m; up to 70 parameters per pixel; 500 GB database
- Purely data-driven methodology
- Relies exclusively on measured numerical data over regular grids
- No patchy or local data
- No interpreted data
- No parameter weighting
- Automated procedures, but processing steps entirely controlled
Advanced Data Processing applied to Mineral Exploration
Supporting Concepts
1) Link regional geoscientific data to the mineral database to characterize the statistical footprint of specific deposit types / commodities
2) Convert the footprints of known deposits into discovery-probability maps that include unexplored but comparable footprints that may represent valuable new targets
3) To be relevant, predictive modelling must retain the smallest surface area while capturing the largest number of deposits to be characterized
Advanced Data Processing applied to Mineral Exploration
Scale of Analysis
Regional to country-scale analysis provides the best leverage to recognize the strongest and largest mineralized systems
A large geostatistical database allows the right footprints to be discriminated against marginal or second-order targets
Geoscientific Data
- Multi-element geochemistry: Lake-bottom sediments, stream, till, soil, rocks- Geophysics: Magnetism, gravity, electromagnetism, radioactivity- Geology- Drilling- Remote sensing- Digital topography
Rock sampling &geological observations
1,031,800 points
Rock samples(299,773 incl. 23,257 from AZM)
Geological observations(731,991)
Data: MERN, Azimut
Processing: Azimut
Geochemistry Surficial SedimentsGovernment Surveys
500,000 samplesover 1.5 million km2
Sampling Points
Data: MERN, Azimut
Processing: Azimut
Lake-bottom sediments (150,635)
+ Azimut samples (17,165)
Stream sediments (230,224)
Soils (75,845)
Tills (44,877)
500 km
Percentile
Data: MERN, Azimut
Processing: Azimut
Copper Content in
Lake-Bottom Sediments
Copper
1
50
100
Percentile
Data: MERN
Processing: Azimut
Geophysical Data
Magnetism
(total field)
Diamond Drill Holes
147,506 holes for a total of
24,280 km of core
(60% of Earth’s circumference)
Drilling Data
Data: MERN, Azimut
Processing: Azimut
500 km
Predictive Modelling that Works
Data processing (for a defined region) leads to:
➢ Footprints of already known mineral deposits and prospects
➢ Comparable footprints of unexplored/underexplored sectors = new potential targets
Field work leads to:
➢ Discoveries
Strong correlation rates between predictionand field validation
26 properties 10,157 claims 4,843 km2
PROPERTY PORTFOLIO IN QUEBEC
Major Results obtained by
Azimut and Partners since 2003
500 km
Copper
Gold
Uranium
PolymetallicUngava Bay uranium
province: Rössing-type
Rex Trend polymetallic
province: Au-Ag-Te-Bi-Cu-W-
Sn, IOCG & Intrusion-related
systems
Intrusion-related gold
mineralization in the
Eleonore mining camp
Prospects
Nantais Belt: Au, Ag, Cu, Zn
Surface area: 1,667,000 km2
Predictive Modelling that Works
Quebec-Scale Examples
Analyzed surf. area Results
Gold 1,169,473 km2 42.4% Au deposits captured within0.46% of the surface area
New targets also located within the 0.46%
Gold Potential Predictive Modelling
over the James Bay region: 224,430 km2
Dates: 2003, 2005, 2009, 2015 and 2016
Parameters: Lake-bottom sediment geochemistry
(30,060 samples), magnetism & gravity data
Azimut’s properties
Eleonore Mine
FootprintSlide 22
Munischiwan
FootprintSlide 23
500 km
Analyzed surface area
21
James Bay Predictive Modelling (2016)
Eleonore Mine Footprint
8 million oz of gold (250 t Au)
22
Munischiwan Property (Azimut – SOQUEM)
James-Bay Predictive Modelling (2016)
Discovery of the InSight Prospect (2018)
InSight Prospect
23
Predictive Modelling that Works
➢ Implementation of AZtechMineTM: Big data expert system applied to initial exploration targeting
➢ Extensive track record of field validation: Discovery of 400+ prospects, including district-scale gold, copper and uranium mineralized systems
➢ Well beyond the experimental stage!
➢ Designed by its users with fully understandable processing steps and results
➢ Geographically transferable: Can be applied whereverthe right database exists
Summary
1) Mineral exploration: Global challenge with partial solutions (regional to country-scale) to improve success rates
2) Large geoscientific databases: Complex and spatially clustered
3) Too many possible targets: Adequate predictive modelling must only retain the very best
Two main ways to proceed:
➢ Expert systems
➢ AI (machine learning, neural networks)
Simplexity versus Complexity
➢ Expert systems
- An expert defines the rules
- Reasoning and processing chains are entirely controlled
- Can be adjusted
- Final human interpretation and ranking
▪ White box
➢ AI (machine learning, neural networks)
- A computer creates the rules
- Gap between software engineer and end-user
- Internal processing steps not mastered, not known by the user
- Not always clear how the results have been produced
- Final human interpretation and ranking
▪ Black box
Simplexity versus Complexity
Common pitfalls in predictive modelling
➢ We learn something that we already know…No risk, but no upside
➢ We learn something that we cannot explain…Potential upside, but high risk
➢ Overfitting models with poor predictive capabilities
Expert systems appear more adapted to the exploration challenge with minimal conceptual bias
Overcome the complexity of large databases and the natural variability of mineral deposits by finding simpler generic solutions.
Simplexity: Berthoz, 2009
Munischiwan
Principaux projets
➢Elmer
➢Munischiwan
➢Pikwa
➢Eléonore Sud
Thank you !
Merci !ᓇᑯᕐᒦᒃ
ᑭᓇᓈᐢᑯᒥᑎᐣ