ED044-0001
Affinity analysis of mineral co-occurrence: Predicting unknown mineral occurrences with machine learning

Tuesday, 15 December 2020
Poster
Shaunna M Morrison1, Anirudh Prabhu2, Ahmed Eleish2, Shweta Narkar2, Peter Arthur Fox2, Joshua J Golden3, Robert T Downs4, Samuel Perry5, Peter C Burns6, Jolyon Ralph7 and Robert M Hazen1, (1)Carnegie Institution for Science, Earth and Planets Laboratory, Washington, DC, United States, (2)Rensselaer Polytechnic Institute, Tetherless World Constellation, Troy, NY, United States, (3)University of Arizona, Tucson, AZ, United States, (4)University of Arizona, Department of Geosciences, Tucson, AZ, United States, (5)University of Notre Dame, Notre Dame, United States, (6)University of Notre Dame, Civil & Environmental Engineering & Earth Sciences, South Bend, IN, United States, (7)mindat.org, Surray, United Kingdom
Abstract:
The expansion of mineralogical data resources facilitates the development of predictive methods, such as affinity analysis [1], that identify the location of previously unknown mineral occurrences, deposits or geologic environments, as well as the probabilistic prediction of the likely mineral inventory at any given locality on Earth’s surface or, where suitable data is available, other planetary bodies.

Mineral Affinity Analysis [2-3] can be used to answer many questions of scientific interest. The most basic application of mineral association rules is to identify the most likely location to find a new occurrence of a specific mineral species, with various probability metrics. This can be expanded to identify the most likely location to find a mineral assemblage - an assemblage that could correspond to a certain geologic setting, planetary environment, or deposit type. This will allow researchers interested in locating planetary analogy sites, exploring and assessing resources, or even simply collecting mineral specimens to identify locations that are not currently known to have the mineral or mineral assemblage of interest, but are likely to. This method goes well beyond querying a database to find a match to a list of minerals - it predicts previously unknown information. Furthermore, researchers can use this recommender system to predict what minerals are likely to occur at a specific location of interest. This has a broad range of applicability, from predicting which rare mineral species, indicative of certain planetary conditions or history, are likely to occur on the surface of Mars based on the broad mineralogy detected by remote sensing to predicting the full mineral inventory for mineral collectors who focus on particular localities.

[1] Brin S, Motwani R, Silverstein C (1997) Beyond Market Baskets, ACM SIGMOD Record.

[2] Prabhu et al. (2019) Predicting unknown mineral localities based on mineral associations, AGU, Abstract EP23D-2286

[3] Morrison SM, Prabhu A, Eleish A, Narkar S, Fox P, Golden JJ, Downs RT, Perry S, Burns PC, Ralph J & Hazen RM (2020) Mineral Affinity Analysis: Predicting Unknown Mineral Occurrences with Machine Learning, Goldschmidt annual meeting (virtual)

Previous Abstract | Next Abstract >>