Orogenic Gold Mineral Prospectivity Mapping (MPM) of the Geraldton Area, Ontario

Orogenic Gold Mineral Prospectivity Mapping (MPM) of the Geraldton Area, Ontario | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Orogenic Gold Mineral Prospectivity Mapping (MPM) of the Geraldton Area, Ontario J. R. Harris, B. Lafrance, E. Grunsky, P. Behnia, M. Nagizedah, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5384377/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 04 Apr, 2025 Read the published version in Earth Science Informatics → Version 1 posted 16 You are reading this latest preprint version Abstract In this paper, we employ Random Forests (RF) (Breiman, 2001 ) to generate several Mineral Prospectivity Maps (MPMs) for orogenic gold in the Geraldton area, located within the Abitibi Tectonic Subprovince of Ontario, Canada. We address various issues pertinent to the Mineral Prospectivity Mapping (MPM) modeling process and propose solutions to these key challenges. Additionally, we utilize multiple methods to analyze text-based geoscientific information derived from geological maps, including a novel application of natural language processing to delineate the sources and traps of gold mineral systems. Mineral Prospectivity mapping gold exploration greenstone belts tectonics machine learning random forests Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 Figure 19 Figure 20 Figure 21 Figure 22 Introduction Machine learning, a significant aspect of the Artificial Intelligence (AI) phenomenon, has become increasingly common for creating Mineral Prospectivity Maps (MPMs). Mineral Potential Modeling (MPM) involves inputting geoscience data—represented as predictor (evidence) maps that illustrate geological, geophysical, and geochemical vectors related to mineralization—into a machine learning algorithm. This process also incorporates training data, which consists of the locations of known deposits, to generate a map highlighting areas with greater potential for mineral exploration for specific commodities (e.g., Au, Cu, Pb, Zn) Mineral Prospectivity Mapping can be approached through two primary methods: data-driven and knowledge-driven modeling. Data-driven methods require a set of known mineral deposits (occurrences) to train a machine learning algorithm (e.g., random forest, neural network) for producing prospectivity maps based on the predictor maps. In contrast, knowledge-driven approaches do not depend on known deposits; instead, each predictor map is assigned an importance weight by the geologist, and these maps are combined using either additive or multiplicative algorithms (e.g., Boolean and index overlay). Both methodologies aim to generate a map that accurately predicts known mineral occurrences while also identifying prospective areas without known mineralization. Validation techniques—such as ROC and classification efficiency curves, as well as cross-validation—are crucial to the MPM process and should always be conducted. Numerous studies have produced MPMs utilizing data-driven methods (Bonham-Carter et al., 1988 , 1989 ; Bonham-Carter, 1994 ; Harris et al., 2001 ; Harris at al, (2006,2008) ; Carranza, 2009 ; Harris et al., 2015a , b ; Rodrigues et al. (2014, 2015), Carranza and Laborte ( 2015a , c ), and Harris et al., ( 2022 ) have identified random forests as a highly effective modeling technique. Consequently, this paper focuses on employing random forests as a data-driven method to generate orogenic gold MPMs. Comprehensive summaries of machine learning algorithms, including random forests, have been provided by various authors (Carranza and Laborte, 2015a , b ; Doan and Foody, 2007 ; Harris et al., 2022 ; Ford, 2020 ; Wang et al., 2020 ; Zhang et al., 2018 ; Zhang et al., 2021 ; Zuo, 2020 ; Zuo and Carranza, 2011 ), so a detailed review will not be included here. The objectives of this paper are as follows: Discussion of Key Issues : A variety of issues relevant to the MPM modeling process are addressed throughout this paper, and solutions to these salient challenges are proposed. These include a novel application of natural language processing to generate predictor maps from geological data, methods for defining non-deposit training data for input into the RF algorithm, a weighting method for known gold mines, and a demonstration of combining MPMs in an ensemble approach. Application of Random Forests : Random Forest (RF) (Breiman, 2001 ) is utilized as a machine learning algorithm to produce several MPMs for orogenic gold in the Geraldton area, part of the Abitibi Tectonic Subprovince of Ontario, Canada. Study Area Regional Geology The granite-greenstone Wabigoon subprovince of the Archean Superior Province is bounded to the south by the metasedimentary Quetico subprovince and to the north by the metasedimentary English River subprovince and the gneissic-granitoid Winnipeg River subprovince (Beakhouse, 1991; Blackburn et al., 1991). The eastern Wabigoon, where the study area is located, extends over more than 180 km east of Lake Nipigon. It consists of two greenstone belts, the Onaman-Tashota greenstone belt to the north and the Beardmore-Geraldton Belt to the south, separated by a major structure called the Paint Lake Fault ( Fig. 1 ). INSERT FIG. 1 The Beardmore-Geraldton Belt is a transitional terrane at the boundary between the Wabigoon and Quetico subprovinces. The belt is East-trending and consists of three metasedimentary rock panels (southern, central, northern sedimentary units), imbricated and in fault contact with three metavolcanic rock panels (southern, central, and northern volcanic units). The southern and northern volcanic units are composed of massive and pillowed basaltic and andesitic flows, whereas the central volcanic unit is dominated by andesitic and dacitic pyroclastic rocks and flows (Shanks, 1993; Tomlinson et al., 1996). They yielded identical crystallization ages of 2724.9 ± 1.2 Ma obtained from samples of a massive felsic flow in the central volcanic unit and synvolcanic feldspar–quartz porphyry dike in the northern volcanic unit (Hart et al., 2002). The metasedimentary rocks panels were deposited above the older metavolcanic rocks as fluvial to alluvial fan deposits (northern sedimentary unit), deltaic to subaqueous fan deposits (central sedimentary unit), and deeper water turbidite deposits (southern sedimentary unit) (Devaney and Williams, 1989). The northern sedimentary unit consists of polymictic conglomerate and minor sandstone (Mackasey 1975, 1976; Mackasey et al. 1976). The southern sedimentary unit is dominated by turbiditic sandstone with minor horizons of polymictic conglomerate and banded iron formation, and the central sedimentary unit is transitional between the northern and southern units ( ibid ). The age of the three metasedimentary units is constrained between ca. 2700 Ma, the age of the youngest detrital zircon population, and ca. 2694 Ma, the age of a cross-cutting feldspar-quartz porphyry intrusion (Tóth et al., 2022). The Onaman-Tashota greenstone belt comprises metavolcanic and metasedimentary assemblages of both Mesoarchean and Neoarchean ages. Their ages and descriptions are from Stott et al. (2002a). Along the north side of the greenstone belt, where East-trending metavolcanic rocks border the Ombabika batholith to the south, the Mesoarchean Toronto assemblage (ca. 2922 Ma) occurs as enclaves within the batholith and as narrow slivers of massive to pillowed basaltic flows overlain by quartz porphyritic intermediate to felsic pyroclastic rocks and flows. In the centre of the belt, where the metavolcanic rocks trend roughly North-South in-between the Ombabika batholith and Elbow Lake pluton to the West and the Onaman pluton to the East ( Fig. 1 ), the Mesoarchean Tashota assemblage (ca. 2975 Ma – ca. 2968 Ma) flanks the Elbow Lake Pluton as a sequence of dacitic felsic tuffs intercalated with pillowed basaltic flows and massive basaltic sills or flows. On the East side of this North-South trending segment, the Onaman pluton is bordered by massive and pillowed basaltic flow and banded iron formation of the Neoarchean Onaman assemblage (ca. 2770 Ma – ca. 2780 Ma). Younger Neoarchean metavolcanic assemblages overlie those older metavolcanic assemblages. In the centre of the belt, pillowed tholeiitic basaltic flows of the Willet assemblage (ca. 2740 Ma) unconformably overlie or are in fault contact with the Tashota assemblage (Mark, 2024). Intermediate to felsic assemblages of similar age occur along the southern margin (Elmhirst-Rickaby) and northern margin (Marshall) of the belt. The Elmhirst-Rickaby assemblage consists of calc-alkaline basaltic to andesitic flows overlain by dacitic to rhyolitic flows and pyroclastic rocks, whereas the Marshall assemblage consists mainly of calc-alkalic dacitic flows and pyroclastic rocks. Volcanism in the belt continued with the deposition of dacitic to rhyolitic flows and pyroclastic rocks of the Metcalfe-Venus assemblage (ca. 2722 Ma – ca. 2734 Ma) and quartz porphyritic dacite of the Humboldt assemblage (<2713 Ma), and was capped by the deposition of turbiditic sandstone and conglomerate similar in age (<2710 Ma) to those in the Beardmore-Geraldton Belt. Deformation of the Beardmore-Geraldton Belt and Onaman-Tashota Belt began in response to south-directed D 1 accretion along the southern margin of the Wabigoon subprovince during the ca. 2690 Ma Shebandowanian Orogeny (Percival et al., 2012). This resulted in the interleaving of metasedimentary and metavolcanic panels of the Beardmore-Geraldton Belt (Devaney and Williams, 1989; Lafrance et al., 2004; Tóth et al., 2023), and in the infolding of metavolcanic rocks between the Ombabika batholith and Onaman pluton in the Onaman-Tashota Belt (Mark, 2024). The latter was coeval with the formation of a foliation (S 1 ) and high strain zones at intrusion-metavolcanic rock contacts. Renewed North-South shortening during a second D 2 deformation event (< ca. 2690 Ma) strongly affected the Beardmore-Geraldton Belt and the southern and northern sides of the Onaman-Tashota Belt and produced regional East-West trending F 2 folds with a prominent axial planar S 2 foliation. The intensification of this foliation formed East- to Southeast-trending deformation zones with a sinistral transcurrent component, such as the Humboldt Bay deformation zone (Fig. 1) in the Onaman-Tashota Belt (Culshaw et al., 2006) and the Paint Lake Fault (Fig. 1) and the Tombill-Bankfield deformation zone in the Beardmore-Geraldton Belt (Tóth et al., 2023). Later D 3 dextral transpression reactivated these deformation zones and East-trending lithological contacts as dextral transcurrent faults and was more pronounced in the Beardmore-Geraldton Belt where it produced multiple outcrop- and map-scale Z-shaped F 3 folds and a regional S 3 cleavage oriented anticlockwise to bedding and S 2 foliation (Devaney and Williams, 1989; Lafrance et al., 2004; Tóth et al., 2023). Regional greenschist facies metamorphism affected the Beardmore-Geraldton Belt and was higher in the Onaman-Tashota Belt, where most metasedimentary and metavolcanic rocks were metamorphosed to the greenschist-amphibolite facies transition and amphibolite facies (Haataja, in prep.). As all metasedimentary and metavolcanic rocks are metamorphosed, the prefix “meta” is dropped for brevity. Felsic magmatism in the Onaman-Tashota Belt range in age from ca. 2922 Ma to ca. 2680 Ma (Stott et al., 2002a). It occurred during volcanism with the emplacement of the syn-volcanic ca. 2777 Ma Onaman tonalite in the Onaman assemblage and the ca. 2731 Ma – ca. 2738 Elmhirst pluton in the Elmhirst-Rickaby assemblage. Syn to late tectonic, ca. 2700 Ma – ca. 2680 Ma, calc-alkalic to sanukitoid plutons, cut across the S 1 foliation in the Onaman-Tashota Belt and Beardmore-Geraldton Belt and provide minimum and maximum ages for the D 1 and D 2 events, respectively. Proterozoic olivine diabase dikes and sills (ca. 1109 Ma) associated with the mid-continental rift are abundant close to Lake Nipigon. Gold Mineralization Gold deposits in the Beardmore-Geraldton Belt are located within two main areas: the eastern Geraldton camp near the town of Geraldton and the western Beardmore camp immediately east of Lake Nipigon. The Bankfield-Tombill deformation zone hosts most of the deposits in the Geraldton camp including, from west to east, the Tombill, Bankfield, Consolidated Magnet, Consolidated Mosher Longlac, MacLeod-Cockshutt and Hard Rock deposits. The Hard Rock and McLeod-Cockshutt deposits were two of the three most prolific mines along the Bankfield-Tombill deformation. They produced 269,081 and 1,366,404 ounces of gold, respectively, from 1938 to 1966 (Mason and McConnell, 1983; Mason and White, 1986), and are being redeveloped as an open pit mine that will produce 5 million ounces of gold over the next 14 years (https://www.equinoxgold.com). The Bankfield-Tombill deformation zone is within the top northern section of the southern sedimentary unit, where intercalated gabbro, mafic volcanic rocks, feldspar-quartz porphyry, turbiditic sandstone, conglomerate and banded iron formation, are folded by tight isoclinal F 2 folds and transposed parallel to the strong east-southeast-striking foliation (S 2 ) that defines the Bankfield-Tombill deformation zone. Gold mineralization is present within all rock types and is preferentially associated with quartz-carbonate veins along sheared east-southeast-striking contacts between feldspar-quartz porphyry and sandstone, with quartz-carbonate veins and sulphide (pyrite, arsenopyrite) replacement lenses in banded iron formation, and with quartz-carbonate veins in massive sandstone and feldspar-quartz porphyry (Horwood and Pye, 1955; Pye, 1952; Anglin, 1987; Macdonald, 1988; Lavigne, 2009; Tóth, 2019). The quartz-carbonate veins were deposited during the D 1 event and folded during the D 2 event (Tóth, 2019). They are surrounded by sericite-carbonate-pyrite ± albite–rutile alteration halos. Other quartz-carbonate veins with similar alteration haloes and tourmaline-rich veins with carbonate-tourmaline-pyrite alteration halos were emplaced during the D 2 event (Tóth, 2019). The deposition of these veins collectively produced a large, up to 250 m wide, sericite-carbonate alteration envelope surrounding the deposit and characterized by elevated S, Te, As, W, and Bi. Leitch, Sand River, and Northern Empire are the main deposits and past-producing mines in the Beardmore camp. The Leitch and Sand River mines collectively produced 897,356 ounces of gold over 29 years of production from 1936 to 1965 (Ferguson, 1967). Gold is present in quartz-carbonate veins with wall-parallel, sericitic and chloritic, internal laminae and wallrock selvedges. The veins are in sandstone and minor iron formation of the southern sedimentary unit in the hinge area of a regional Z-shaped F 3 fold (Lafrance et al., 2004). The latter is defined by the folded contact between the southern sedimentary unit and volcaniclastic rocks of the central volcanic unit (Lafrance et al., 2004). The veins are grouped in two sets based on their orientation (Ferguson, 1967). One set consists of straight to curviplanar, continuous veins, which occupy eastnortheast-striking D 3 dextral shear zones that are parallel to the axial plane of the F 3 fold (Lafrance et al., 2004). The second set consists of strongly folded veins oriented roughly perpendicular to the axial plane of the F 3 fold. 20 km to the east, the Brookbank zone, which has an indicated resource of 600,000 ounces of gold (https://www.equinoxgold.com), was emplaced along similar dextral shear zones at the contact between polymictic conglomerate of the northern sedimentary unit and basalt of the northern volcanic unit (DeWolfe et al., 2007). Little information is available on the Northern Empire mine, which is located 1 km eastnortheast of the town of Beardmore. The mine produced 149,492 ounces of gold from steeply dipping, eastnortheast-striking quartz veins in sheared mafic pillowed volcanic rocks of the southern volcanic unit (Mason and White, 1986). Most gold occurrences in the Onaman-Tashota Belt are located in the Onaman assemblage (Tashota-Nipigon mine) and Elmhirst-Rickaby assemblage (Brenbar mine, Sturgeon River mine). The Tashota-Nipigon mine produced 12,356 ounces of Au, 14,527 ounces of Ag, and 575,000 lbs of Cu from eastnorteast-striking quartz-pyrrhotite-chalcopyrite veins and ore zones plunging 55°-60° to the northwest (Thurston, 1980). The Brenbar mine is hosted within intermediate to felsic volcaniclastic rocks of the Elmhirst-Rickaby assemblage and produced 134 ounces of Au from ENE-striking and steeply-dipping laminated quartz veins surrounded by alteration haloes of sericite, carbonate, and pyrite (Mackasey, 1975). The Sturgeon River Mine, which is located 2 km east of the Brenbar mine, produced 73,738 ounces of Au and 15,922 ounces of Ag from NNE-striking laminated quartz veins surrounded by alteration haloes of sericite and chlorite (Mackasey, 1975). Other smaller occurrences and artisanal mines from the early 1900’s are hosted by the roughly north-striking Tashota deformation zone in the centre of the Onaman-Tashota Belt. The Tashota deformation zone straddles the contact between the Ombabika batholith and volcanic rocks of the Tashota and Willet assemblages. Two of these occurrences, the Adair and Wascanna prospects, consists of quartz-chlorite ± muscovite ±ankerite veins, which are surrounded by alteration halos of chlorite, sericite and ankerite and are isoclinally folded parallel to the strong S 1 foliation along the deformation zone (Mark, 2024). In summary, gold mineralization is preferentially located in zones of structural complexity, such as the hinge of F 2 and F 3 folds, and lithological complexity due to shearing along lithological contacts. It is also located in east-northeast- to east-southeast shear zones and deformation zones in association with quartz-carbonate veins of similar orientation. The majority of Au mines, developed prospects and occurrences are found in the Beardmore- Geraldton greenstone belt ( Fig. 2 ) INSERT FIG. 2 Methodology and Experiments (objective 1) Experiment 1 – weighting of gold mines Having a sufficient number of training points (e.g., Au mines) for input to machine learning algorithms is an important issue discussed by many different authors (Carranza and Laborte, and references therein, 2015c; Zuo et al, 2015). Following the methodology presented by Harris et al (2006, 2008) we spatially weight the three most prominent gold mines in the study area by the number of ounces of Au produced before input to the modelling exercises. By adding additional points surrounding (within 200 m) each mine. We have also increased the number of training points from 12 to 21 thus providing a more robust training data set for modeling purposes. We then compare and contrast how well these MPMs perform using efficiency of validation curves (Chung and Fabri, 2003; Harris et al, 2006, 2008). We examine how well each MPM predict weighted Au mines, developed prospects and occurrences. Experiment 2 – production of random non-deposit sites for RF modelling We also investigate a number of different methods for generating non-deposits required for RF modeling. The methods we propose in this paper include the following: Select a random number of non-deposit points equal to the number of Au mines\prospects\occurrences (total of 189) without geographic restrictions. Select a random number of points equal to the number of Au mines\prospects\occurrences (total of 189) with geographic restrictions (> 2 km from a known Au mine). Select a random number of Au mines\prospects\occurrences (total of 189) but geographically restricted to lithologies not known to contain Au occurrences (low potential lithologies). Experiment 3 – Ensemble classification In keeping with the theory of ensemble classification (RF is an ensemble classifier by nature), we conduct a 5-fold repletion of our RF modeling using a separate set of randomly selected non-deposit points (total of 189 matching the total number of weighted Au mines, prospects and occurrences), greater than 2 km from a known mine for each repetition. This enables us to bracket the variability in a 5-fold repetition of the RF modelling. The variability refers to the oob and overall accuracies as well the best predictors. We then take both the average and weighted average of the 5 MPMs generated from the 5-fold repetition and compare, using overall accuracies and efficiency of classification curves, how well they predict the known Au mines, developed prospects and occurrences. Experiment 4 – Natural Language processing We also introduce the concept of Natural Language Processing (NLP) to extract and apply unstructured and semi-structured text data for the purposes of RF modelling. Bedrock geology maps represent an important source of geoscientific information, including rock types, geological ages, mineralogy, textures, and cross-cutting relationships that are essential for mapping the different component of gold-bearing mineral systems. These unstructured forms of geoscientific text data, which commonly occur in large volumes, are essential to the map-building process because they contain the concepts and observations underpinning the preferred map representation (Brodaric et al., 2004; Pavlis et al., 2010; Mantovani et al., 2020). For the Geraldton map area, text data is contained within attributes that are linked to geology polygons in GIS (i.e., "UNITNAME_P", "ROCKTYPE_P", "SUPEREON_P", "EON_P", "ERA_P", "PERIOD_P", "EPOCH_P", "RegionClas", "SubClass", "RAge"). Text data from each of these attributes was concatenated prior to applying the NLP methodology described in Lawley et al. (2022): (1) text data were converted to word tokens using the “tidytext” (Silge and Robinson, 2016) and “tokenizer” (Lincoln, et al., 2018) packages in R; (2) uninformative word tokens were removed using the tidytext list of English “stop words” (n = 1149) and a list of North American place names; (3) plural terms were relaced using the Harman (1991) method to focus text analysis on root words; and (4) numbers and word tokens with two or fewer characters were removed from further analysis. Each of the processed word tokens were then joined with a geoscience Global Vectors for Word Representation (GloVe) model (Lawley et al., 2022). This model was re-trained on public geoscientific documents from Canada and a subset of international and peer-reviewed publications, and has been shown to outperform more complex language models on geoscience analogy, clustering, relatedness, and nearest neighbour tasks (Lawley et al. 2022). Individual word vectors were then averaged to generate one word embedding for each map polygon. The cosine similarity between each map polygon embedding and the terms “igneous”, “metamorphic”, and “sedimentary” from the geoscience GloVe model were used as input for RF modelling. Unlike most previous studies that use text as a form of categorical data or apply one-hot encodings to create binary predictor maps, cosine similarities represent a continuous variable (-1 to 1) and are based on all of the available geoscientific information. Cosine similarities represent a form of semantic search and can be applied to map the most likely sources and traps of mineral systems. Data processing for producing MPMs (objective 2) Random Forest Algorithm Random forest (Breiman, 2001) is based on producing a forest of individual decision trees and data (predictor maps) are passed through the trees to create a final mineral prospectivity map (MPM) through majority vote for either the presence or absence of a deposit. Two aspects of randomness are introduced to the algorithm one through bootstrapping and the other through permutation of the predictor maps passing through each tree. The user defines the number of trees to build as well as the percentage of training samples to use for the construction of each tree. Typically for each tree 2/3 of the training data are selected with replacement for prediction and 1/3 are left for validation ( boot strapping ). This process allows for an oob ( out-of-bag-accuracy ) to be calculated. Furthermore, the geologist chooses the number of predictor maps that are used to pass through each tree in the validation and information gain calculation process that is used to rank the importance of the variables. Typically, the square root of the number of predictor maps are passed down each tree. The trees are kept short although the trees are not pruned. This characteristic, as well as bootstrapping, helps prevent overfitting the model. Furthermore, RF provides a measure of importance for each predictor map. In this study we used a technique called permutation importance whereby a baseline classification accuracy is established by passing out-of-bag (oob) samples through the random forest. The column of a single predictor map is permuted and then the samples are run through the random forest. The importance of a predictor map is determined by noting the difference between the baseline accuracy and the drop in accuracy by permuting the column. This procedure is computationally more complex but the results are mor reliable. Finally, an MPM is produced by calculating the probability of an Au deposit at a given location (pixel) by dividing the number of votes for Au by the total number of trees. For example, if a pixel location had 1 vote for Au and 10 trees the probability would be 1/10 which equal 0.01. Conversely, if the pixel had 10 votes that probability would be 10/10 which equals 1. Generation of training data for input to RF Gold was divided into 3 categories: occurrences (total of 166), developed prospects (11) and mines (12). However, following the methodology proposed by Harris et. (2006, 2008) we have weighted the mines that have produced more than 500,000 ounces for gold. These 3 mines are the Leitch mine (1,366.404 oz Au), the Beardmore mine (861,982 oz Au) and the Geraldton mine (605,449 oz Au). These mines developed prospects and occurrences are shown in Figure 2. Each mine was weighted by adding an additional point within 200m of the actual location of the mine as illustrated in Figure 3 . Thus, Leitch received a weight of 4 additional points, Beardmore, 3 points and Geraldton, 2 points. In addition to weighting these mines by importance, addition of these extra points increased the number of points for Au mines from 12 to 21, a number felt adequate for training and RF modelling. In addition, RF requires a set of non-deposit points equal in number to the number of Au mines, prospects and occurrences. The different methods of generating these randomly selected points are discussed below. INSERT FIG. 3 Generation of Predictor maps for input to RF Firstly, predictor maps (total of 31) for predicting orogenic Au occurrences are prepared for modelling. Table 1 provides a summary of the predictor maps used in this paper. INSERT Table 1 Lithology Natural language processing was used to extract textual information from the existing lithological map and descriptive legend, resulting in 5 predictor maps: (1) lithologic units modelled as categorical data (n = 13); (2) lithologic sub-units modeled as categorical data (n = 13); and (3) the cosine similarity between the word embeddings of each map polygon and “sedimentary”, “metamorphic” and “igneous” terms included within the geoscience GloVe model. Cosine similarities were added to the RF modelling as continuous variables. These five predictor maps are shown in Figure 4a,b and c . Overall, the NLP methodology is based on 8447 words, corresponding to approximately 23 tokens for each map polygon. Archean (n = 686), rock (n = 568), mafic (n = 447), volcanic (n = 428), and Precambrian (n = 361) represent the most frequently used words. Gold occurrences tend to be associated with mafic volcanic rocks and/or carbonaceous sedimentary rocks, which is expected given that gold is often transported as sulphide complexes that become de-stabilized during interaction with iron- or carbon-rich lithologies. Iron formation-hosted gold deposits in the southern Beardmore greenstone belt represent an important example of the relationship between gold mineralization and suitable trap rocks. Thus, iron formations were extracted from the lithology map and buffered to 1 km representing a zone of influence on alteration/mineralization determined from geological field observations ( Fig. 5a ). The same is true for lithologic contacts; they were extracted from the lithologic map and buffered to 1 km ( Fig. 5b ), representing possible channel ways for mineralized Au-bearing fluids. INSERT FIG.4 and 5. Geochemistry Rock-sample analysis data of 652 samples was extracted from the Metal Earth database. Major elements (Si, Al, Fe, Mn, Mg, Ca, Na, K, Ti, P, S) and trace elements (Ag, Au, Cu, Pb, Zn, Ni, Co, Cr, Hg, Sr, Ba, Sn, Sb, Mo, Te, Bi, W, and As) were selected based on their importance in gold mineralization and availability of data. Figure 6 shows the spatial distribution of the lithogeochemical data. The data is geographically restricted to the central/east portion of the study area. This spatial distribution will have a negative effect on the RF modelling which will be discussed later. Oxides first were transformed to element content and then all data were transformed to ppm. The log-ratio Expectation-Maximization method (IrEM function, implemented in zCompositions R package) was used to impute the missing and below detection limit (BDL) values (Martín-Fernández, et al., 2003; Palarea-Albaladejo et al., 2014). IrEM algorithm replaces BDL values with a value between 0 and the lower detection limit of each analysis for elements with 40% missing values. To deal with the closure problem of compositional data a centered log-ratio (clr) transformation (Aitchison, 1986) was applied on the imputed data. INSERT FIG. 6 Principal component analysis (PCA) was applied to the logcentred transformed imputed data (621 analyses) using measures of correlation and covariance. The principal component screeplot of Figure 7 indicates a steep decay of values from principal component 1 through 5. This reveals that most of the significant correlations and variability of the elements occurs within the first five components. The lesser eigenvalues likely represent under-sampled processes or random artefacts in the data (Grunsky, 2010). INSERT FIG. 7 Figure 8 shows a biplots of the PC1 vs. PC2. The principal component loadings (elements) are coloured according to a generalized Goldschmidt classification. The biplot shows a contrast between intrusive rocks along the negative portion of PC1 axis and volcanic rocks along the positive PC1 axis. The negative PC1-PC2 quadrant reflects relative enrichment of Al, S, Si, Fe, K, Cr, Ba, reflecting a mixture of both felsic and mafic intrusive rocks. The positive PC1, negative PC2 quadrant reflects relative enrichment in Mg. The positive PC2 axis shows relative enrichment in chalcophile elements, with the exception of Cu, suggesting that the fewer principal component scores that plot above the zero value of PC2 as possibly associated with alteration systems associated with Au or massive sulphide mineralization. INSERT FIG. 8 Figure 9 shows a biplot of PC2 vs. PC3 where there is a chalcophile trend along the positive PC2 axis, however, there is an inverse association between Sn-Sb with As-Zn-Bi-W-Pb along the PC3 axis, suggesting two different environments. INSERT FIG. 9 Figure 10 shows a biplot of PC3 vs.PC4. An association of Bi-Pb-W-Na-K-As-Ti-Zn-Sr and Cr occurs along the negative PC3 axis. This represents an association of chalcophile elements with potential association with mineralization with the Na-K association of felsic (granitoid rocks). INSERT FIG. 10 Figure 11 shows a map of the scores for PC1, derived from the correlation-based PCA, where the colours of the symbols reflect a contrast between intrusive rocks (negative PC1) with volcanic rocks (positive PC1). INSERT FIG. 11 Figure 12 shows a map of PC2 scores, derived from the correlation-based PCA, where positive values reflect relative increases in the chalcophile elements W-Zn-Sb-Bi-Sb-As-Pb-Te. The map also shows the status of the mineral inventory for the area. Developed prospects, past and present producing mines are shown as distinct symbols. There is a clear association of positive PC2 values associated with the Ishkoday [Au] past producer, Paulpic [Au], Headway-Onaman [Zn, Pb], Lynx-Dejour-Reynolds North and South, Jacobus [Cu, Ni], Brookbank [Au], Nortoba-Tyson (No.3 vein) [Mo] prospects. Other elevated values of PC2 are also shown without any close geospatial association with known mineral occurrences. INSERT FIG. 12 Figure 13 shows a map of PC3 scores [correlation-based PCA] where negative scores show a relative increase in values of Sn-Sb and positive scores show a relative increase in values of W-Bi-Zn-As-Pb INSERT FIG. 13 Figure 14 shows a map of PC4 where there is a contrast between relative enrichment of Sn-Sb-Zn-Cu (negative scores) with Te-Bi-Pb-W (positive scores). INSERT FIG. 14 Based on the above analysis we selected components 2,3,4 and 5 to be used as input to the RF algorithm. However, because the data was not equally spread throughout the study area, we chose not to interpolate the data but rather buffer each sample point (shown in Fig. 6) to 1 km acting as a zone of influence. This approach has been previously used by Harris et al. (2022). The anomalous samples ( Fig.15 ) were selected greater than 3 standard deviations above the population mean for each component. As seen in Figure 15 , all the anomalous PCA geochemical sample points fell in the central-west portion of the study area and missing in the east portion of the study area. Based on the above analysis we selected components 2,3,4 and 5 to be used as input to the RF algorithm. However, because the data was not equally spread throughout the study area, we chose not to interpolate the data but rather buffer each sample point (shown in Fig. 6) to 1 km acting as a zone of influence. This approach has been previously used by Harris et al. (2022). The anomalous samples ( Fig.15 ) were selected greater than 3 standard deviations above the population mean for each component. As seen in Figure 15 , all the anomalous PCA geochemical sample points fell in the central-west portion of the study area and missing in the east portion of the study area. This will have negative effects on how useful the lithogeochemical data will have on the modelling procedure using the RF algorithm (incomplete coverage). INSERT FIG. 15 Geophysics The geophysical data analytic was carried out on 3D density, susceptibility, and resistivity models. The 3D density model was inverted from the gravity Bouguer anomaly (GSC, 2020) on a horizontal grid of 1x1 km. The 3D susceptibility model was generated by inverting the total magnetic anomaly (GSC, 2021) on a horizontal grid of 250x250 m. The Magneto-Telluric (MT) data acquired as part of the Metal Earth project was inverted to generate the 3D resistivity model on a horizontal grid of 1.5x1.5 km. The vertical grid of the 3D density, susceptibility, and resistivity models was exponentially increased toward the deeper parts of the model. All of the geophysical inversion tasks were carried out by imposing model smoothness constraints. Next, all three of the 3D models were re-sampled/interpolated into a regular grid of 500x500x500 m to generate co-located volumes of density, susceptibility, and resistivity suitable for analysis Figure 16 shows example of 3 geophysical images used in the RF modelling: (a) density from the surface to 750m deep, (b) susceptibility – 4250m deep and (c) resistivity – 2250m deep. Not all the geophysical images (Table 1) used in RF modelling are shown to save on space. INSERT FIG. 16 Faults Two sets of faults were used: mapped faults and interpreted faults from the airborne magnetic data. These faults were separated into cardinal strike directions and buffered using a 1 km zone of influence based on field observations with respect to the distance to which alteration effects could be mapped. Figure 17 shows the mapped and interpreted faults buffered to a distance of 1 km. as a zone of influence based on field observations of significant alteration. Although they are much younger (Proterozoic) buffered dikes (not shown) were used as were the faults as possible channels for Au-bearing fluids. The prominent Paint Lake Fault ( Fig 1 ) is also labelled on Figure 17 . INSERT FIG. 17 RESULTS Experiment 1 – MPMs based on Au mines, developed prospects and occurrences Figure 18 presents the 3 MPMs generated from Randon Forests; (a) based on weighted mines (21), developed prospects (11) and all Au mines, prospects and occurrences (166). The correlations between these maps are low ( Table 2 ) indicting overall spatial dissimilarities. The highest correlation is between the MPM based on the occurrences vs. all occurrences (0.81). This is to be expected due to the greater number of occurrences when comparing to all the Au points. INSERT FIG. 18 It is apparent that the Beardmore greenstone belt is more prospective for gold than the Geraldton belt when assessing the MPM based on mines. The out-of-bag ( oob ) error rates and accuracy of classification are listed in Table 3 . The MPM based on developed prospects shows poor oob accuracy but perfect classification accuracy. This is in part due to an insufficient number of training samples (11) and, therefore, the DVP MPM is not robust. INSERT TABLE 2 INSERT TABLE 3 Figure 19 presents efficiency of classification curves for the various combinations of gold and derived MPMs. The best performing models in order of prediction rate, as one might expect, involves the MPM based on the Au mines predicting the mines, followed by the developed prospects (DVPs) MPM predicting the DVPs and then the MPM based on the Au occurrences predicting the occurrences. All have areas under the curves more than 0.9 which represent good prediction results. With respect to mines approximately 92% of the mines are predicted in 2% on the most prospective areas whereas the prediction results for the DVP at 92% of the deposits is 12% of the most prospective areas and the occurrences are 14% of the area. INSERT FIG. 19 However, when considering whether the MPMs based on the mines DVPs and occurrence; they are not highly predictive of the Au points not involved in producing the MPM (Fig. 19) For example, the MPM based on the mines do not predict the DVPs nor occurrences very well. The same result (low prediction rate) based on the DCP MPM versus the Au occurrences, can be seen in Figure 19. These prediction rates are all characterized by less than 0.7 area under the efficiency of classification curve. The prediction rate for the above MPMs is not much better than random. The prediction rate for the MPM based on the occurrences versus the mines and DVPs is moderately better yielding areas under the curve of .887 and .873, respectively. The prediction rate for the DVP MPM versus mines is only moderate at best. Experiment 2 – Non-deposit point selection – MPMs based on 3 different methods for calculating non-deposit Au points for RF modelling Figure 20 shows the efficiency of classification curves for the 3 different methods of defining random non-deposit points discussed above. The curves are based on an MPM produced using all Au training data, DVPs and mines (total of 189) produced using a 5-fold repetition of the RF modelling as mentioned above. This was in part done to address how variable the results are from individual runs of the RF modelling and also to produce 5 MPMs that could be combined in an ensemble fashion (discussed in the next session). INSERT FIG. 20 The curves show that the MPM (not shown) based on all the Au data when predicting the mines are very similar and quite strong from a predictive point of view with areas under the curve greater than 0.9. It is interesting to note that best MPM in terms of prediction of all the Au data was based on the selection of non-deposit points restricted so that none fell within 2 km of an Au point. This was followed by non-deposit points restricted to lithologies not expected to contain orogenic Au, followed by non-deposit points selected randomly without any spatial restrictions. In fact, the MPM based on all the Au data was a strong predictor of mines, DVPs and all the Au data. With respect to the variability in results from the 5-fold repetition of the RF modelling, Table 4 shows the variability in the out-of-bag (oob) error, overall classification accuracy and strongest predictors (predictor maps) over 5 repetitions. INSERT TABLE 4 With respect to out-of-bag and overall classification accuracies, they are very similar and stable over the 5-fold repetition, as would be expected given the RF is based on random selection of training areas (boosting) and predictor maps (bagging) for each run of RF. However, there is some variability in the best predictors over the 5-fold repetition. The geological-sub units and density at the 750m and 7250m levels are important predictors in all runs of RF. The important predictor maps will be further discussed below. Experiment 3 – Ensemble Methods Following the 5-fold repetition,5 MPMs were generated using all the Au training points (189). Two ensemble methods were used to create combined MPMs: Method 1 – average = (mines + developed prospects + occurrences)/3…… eqn1 Method 2 – weighted average = ( (mines * 3) + developed prospects + occurrences)/3 ….. eqn2 These MPMs are shown in Figure 21 along with their associated efficiency of classification curves. In both maps the MPMs based on the mines showed the greatest prediction rates. The MPM based on the average of the 5 MPMs predicts 84% of the mines in 2% of the most prospective areas and 100% of the mines in 25% of the area. These prediction results improve for the weighed average MPM (as would be expected) resulting in again, 84% of the mines predicted in the top 2% of the area of the MPM but 100% of the mines in only 10% of the area. The results are less strong for the Au developed prospects and Au occurrences as well as all the 189 Au training points. INSERT FIG. 21 Experiment 4 – NLL The natural language processing was successful as the predictor maps resulting from this processing were among the most important predictors (geological units, geological sub-units – (see Table 1 and 3a for listing of predictor maps). One can see from Table 3a that the NLL has produced important predictor maps of orogenic gold especially geologic sub-units. Best Predictors There is some variability in the best predictors, as would be expected due to the random nature of the RF modelling algorithm, between all experiments. This has already been seen in Table 4 discussed above . In addition to Table 4 , Table 5 presents the 6 best predictors for the MPMs generated from the mines and developed prospects and all Au training data (see Fig 18 for MPMs). INSERT TABLE 5 Firstly, the top predictors for each MPM are quite different (suggesting a difference in deposit model – iron formation vs. orogenic, structurally controlled models). The best predictors for the mines which are considered the most important of the RF models are the geophysical data (susceptibility especially) followed by iron formations and NLL produced geologic sub-units. Overall, the best predictors, regardless of the MPM, are geological sub-units, susceptibility at the 4250m depth level and density at the depth 750m level. DISCUSSION Considering the MPMs based on the Au mines, developed prospects and occurrences ( Fig. 18 ), the most predictive map, through evaluation of efficiency of classification curves ( Fig. 19 ), oob and classification accuracies ( Table 3 ), is the MPM generated from the Au mines followed by all the Au training data and developed prospects. However, the MPM generated from the developed Au prospects is not robust, nor stable, due to too few training points. When considering a cross-comparison between all MPMs (e.g., mines vs. developed prospects, occurrences vs. prospects etc.) the predictive results are very much weaker (Fig. 19). This is somewhat contrary to the results achieved by Harris et al. ( 2006 , 2008) in a study of the Red Lake greenstone belt in Ontario where the cross-comparisons were much more predictive. This may be due, in part, to differences in the Au deposit models. This study has shown that the selection of training points is critical as the results from the RF modelling can be significantly different depending on the points chosen (e.g., mines, developed prospects, occurrences). In this study we consider the MPM based on the Au mines as being the most robust and predictive. It can be noted that the MPMs are somewhat blocky in appearance. This is due to incomplete coverage of the entire study area by the 3D geophysical data. However, this does not affect the RF modelling other than highlighting the boundaries of the geophysical data. Figure 22 is a map that combines the most prospective areas derived from the MPMs shown in Fig. 1 8. The most prospective areas were derived by taking greater and equal to 3 standard deviations above the mean probability from each MPM and creating a ternary combination whereby a green colour represents an area of 1 overlap (e.g., present on only one of the MPMs), and a purple colour represents areas where 2 MPMs overlap. There are no areas where all 3 MPMs overlap. Table 6 presents a geological summary of the each of the most prospective zones shown on Fig. 22 . Areas C and D are of interest as they comprise iron formations. Area B is associated with the Paint Lake Fault (see Figs. 1 and 17 for the location). Area A is of interest as it predicts 2 major mines; the MacLeod-Cockshutt and Geraldton mines. INSERT TABLE 6 INSERT FIG. 22 The majority of the most promising areas for Au exploration are found in the Beardmore-Geraldton greenstone belt, located in the southern part of the study area. Considering the method of generating non-deposit points our study indicates that a random selection of points equal to the number of training points but restricted to areas greater than 2 km from a training point (e.g., Au mine) produces the most predictive results ( Fig. 19 ). This method is recommended over a selection of random points without spatial restrictions. The weighted average ensemble of all MPMs (mines, developed prospects, occurrences - Fig. 21 ) provides strong predictions of the Au mines which is slightly better than the MPM generated from only the mines. This is not unexpected as the mines were more heavily weighted by a factor of 3 (see Eq. 2 ). However, the average and weighted average ensemble maps ( Fig. 21 ) do not predict very well the MPMs generated from the developed prospects and occurrences alone (Fig. 18). The most variable of the MPM parameters between different runs or especially when using different training areas, is the resulting most important predictor (predictor) maps. The strongest predictors for all variations of the MPM modelling process included the lithological data, obtained through NLL and the 3D geophysical data obtained at different depths. It is somewhat disappointing that the lithogeochemical data, processed using PCA, were not strong predictors due to relatively low number of sample points that were not distributed evenly across the study area. CONCLUSIONS This study has produced a number of Au MPMs of the Geraldton area in Ontario, Canada. Various experiments have been conducted that have relevance to the MPM RF modelling process. With respect to these experiments a number of conclusions can be summarized: The training data should be divided by importance (e.g., Au mines, developed prospects, occurrences). The MPMs generated by this procedure will be different as reflected by derived variations in measures of validation and visually with respect to prospective areas emphasized. These differences my shed light on slight differences in Au deposit models or tectonic environments in which the Au was deposited. As expected, the MPMs generated using the same training data you are trying to predict offer the best results. When cross-comparing MPMs based on different training data, the predictive results are lower as might be expected. One solution to this problem is to calculate an ensemble weighted average of the MPMs derived from different Au training datasets. In this study we recommend firstly weighting the more important Au deposits (e.g., by tonnage) and secondly providing more weight to the Au mines in generating the weighted ensemble MPM. Natural Language processing of textual geologic maps and legends provides valuable information for producing predictor maps for input to the MPM RF modelling process. We have identified a number of prospective zones in the study area for Au exploration follow-up. The majority of these areas occur in the Beardmore-Geraldton greenstone belt which appears more fertile than the Onaman-Tashota Belt. The lithogeochemical data were not strong predictors of the Au data due to an incomplete and sporadic coverage over the study area. Although PC5 which is heavily weighted by As and Zn and PC2 heavily weighted by chalcophile elements (W-Zn-Sb-Bi-Sb-As-Pb-Te) do show a weak correlation with Au mines as 2 of the 12 (non-weighted) Au mines fall directly on the buffered PC5 and PC2 anomalies (see Fig. 15 ). Declarations Author Contribution "J.H ., B.L. , C.L. , M.N., M.P., E.G., P.B., prepared the manuscript; J.H., E.G., B.P prepared the figures; J.H . , B.L., E.G., C.L., conceived the idea of the research" ACKNOWLEDGEMENTS The authors would like to thank Metal Earth (Laurentian University, Sudbury, Ontario, Canada) for supporting this research. The Metal Earth publication number for this paper is MERC-ME-2024-32. We would also like to thank the anonymous reviewers whose comments resulted in a much clearer rendition of our work. Funding Declaration This work was funded by the Metal Earth Project, Laurentian University, Sudbury, Ontario, Canada. Grant # CFREF- 2015-00005. References Aitchison, J., 1986. The Statistical Analysis of Compositional Data. Chapman and Hall, London, 416 p. Anglin, C.D., 1987. Geology, structure and geochemistry of gold mineralization in the Geraldton area, Northwestern Ontario. Unpublished Master’s Thesis. Memorial University of Newfoundland, St. John’s, Newfoundland. 283 p. Beakhouse, G.P., 1991. Winnipeg River Subprovince, in: Thurston, P.C., Williams, H.R., Sutcliffe, R.H., Stott, G.M. (Eds.), Geology of Ontario. Ontario Geological Survey, Special Volume 4. Part 1. pp. 279–301. Blackburn, C.E., Johns, G.W., Ayer, J., Davis, D.W., 1991. Wabigoon Subprovince, in: Thurston, P.C., Williams, H.R., Sutcliffe, R.H., Stott, G.M. (Eds.), Geology of Ontario. Ontario Geological Survey, Special Volume 4. Part 1., pp. 303–381. Bonham-Carter, G. F., Agterberg, F. P., Wright, D. F. 1989,. Weights of evidence modelling: a new approach to mapping mineral potential. In F. P. Bonham-Carter, G.F. (1994). Geographic Information Systems for Geoscientists: Modeling with GIS . Pergamon, Elsevier Science Inc., New York. Bonham-Carter, G.F., Agterberg, F.P., Wright, D.F. (1988). Integration of geological datasets for gold exploration in Nova Scotia., Photogrammetric Engineering and Remote sensing. 54(11), pp. 1585-1592. Breiman, L. 2001. Random forests. Machine Learning, 45(1), pp. 5-32. Brodaric, B., Gahegan, M., & Harrap, R. (2004). The art and science of mapping: Computing geological categories from field data. Computers & Geosciences, 30 (7), 719–740. Carranza, E.J.M. (2009). Controls on mineral deposit occurrence inferred from analysis of their spatial pattern and spatial association with geological features. Ore Geology Reviews, 35, 383-400. Carranza, E.J.M. and Laborte, A.G. 2015a. Data-driven predictive modeling of mineral prospectivity using random forests: a case study in Catanduanes Island (Philippines), Nat Resource Res,25: pp. 35-50. Carranza, E.J.M., Laborte, A.G. (2015c). Random forest predictive modeling of mineral prospectivity with small number of prospects and data with missing values in Abra (Philippines). Computers & Geosciences 74, 60-70. Carranza, E.J.M.; Laborte , A.G., 2015b Data-driven predictive mapping of gold prospectivity, Baguio district, Philippines: Application of Random Forests algorithm . Ore Geol. Rev., 71, 777–787. Chung, C.J., and Fabri, A., 2003.Validation of Spatial Prediction for Landslide Hazard Mapping, Natural Hazards, 30(3); pp 451-472, DOI: 10.1023/B:NHAZ.0000007172.62651.2b Culshaw, N., Purves, M., Reynolds, P., and Stott, G. 2006. Post-collisional upper crustal faulting and deep crustal flow in the eastern Wabigoon subprovince of the Superior Province, Ontario: Evidence from structural and 40 Ar/ 39 Ar data from the Humboldt Bay High Strain Zone; Precambrian Research, v.145, pp.272-288. Devaney, J.R., Williams, H.R., 1989. Evolution of an Archean subprovince boundary: a sedimentological and structural study of part of the Wabigoon–Quetico boundary in northern Ontario. Can. J. Earth Sci. 26, 1013–1026. DeWolfe, J.C., Lafrance, B., Stott, G.M., 2007. Geology of the shear-hosted Brookbank gold prospect in the Beardmore–Geraldton belt, Wabigoon Subprovince, Ontario. Can. J. Earth Sci. 44, 925–946. doi:10.1139/E06-118 Doan, H.T., Foody, G.M., 2007. Increasing soft classification accuracy through the use of an ensemble classifier. Int. J. Remote Sens. 28, pp. 4609–4623. Ferguson, S.A. 1967. Leitch Gold Mines Limited, Surface plan of eastern part of property, Parts of Eva and Summers Townships, District of Thunder Bay. Ontario Department of Mines, Preliminary Geological Map P.484. Ford, A. (2020). Practical implementation of random forest-based mineral potential mapping for porphyry Cu-Au mineralization in the Eastern Lachlan Orogen, NSW, Australia. Natural Resources Research , 29, 267-283. doi: 10.1007/s11053-019-09598-y Grunsky, E.C., 2010. The interpretation of geochemical survey data; Geochemistry, Exploration, Environment Analysis 10(1), p. 27-74, doi: 10.1144/1467-7873/09-210. GSC (Geological Survey of Canada), 2020. Geophysical Data. http://gdr.agg.nrcan.gc.ca/gdrdap/dap/search-eng.php. Accessed Feb 3, 2023. Harman, D. (1991). How effective is suffixing? Journal of the American Society for Information Science, 42 (1), 7–15. Harris, J., Grunsky, E., Behnia, P., Corrigan, D. (2015a). Data-and knowledge-driven mineral prospectivity maps for Canada's North. Ore Geology Reviews 71, 788-803. Harris, J., and Grunsky, E.C. (2015b). Predictive lithological mapping of Canada's North using Random Forest classification applied to geophysical and geochemical data. Computers & Geosciences 80, 9-25. Harris, J.R., M. Naghizadeh, P. Behnia, L., Mathieu, 2022. Data-driven gold potential maps for the Chibougamau area, Abitibi greenstone belt, Canada, Ore Geology Reviews, 150, 105176. Harris, J.R., Sanborn-Barrie, M., 2008. Mineral potential mapping: examples from the Red Lake greenstone belt, northwest Ontario. in : Harris, J.R. (Ed.), GIS for the Earth Sciences. Geological Association of Canada Special Paper 44, pp. 1–2. Harris, J.R., Sanborn-Barrie, M., Panagapko, D.A., Skulski, T., Parker, J.R., 2006. Gold prospectivity maps of the Red Lake greenstone belt: application of GIS technology. Can. J. Earth Sci. 43, 865–89 Harris, J.R.; Wilkinson, L.; Heather, K.; Fumerton, S.; Bernier, M.A.; Ayer, J.; Dahn, R. 2001. Application of GIS Processing Techniques for Producing Mineral Prospectivity Maps—A Case Study: Mesothermal Au in the Swayze Greenstone Belt, Ontario, Canada. Nat. Resour. Res 10 , 91–124. https://doi.org/10.1023/a:1011548709573. Hart, T.R., terMeer, M. and Jolette, C. 2002. Precambrian geology of Kitto, Eva, Summers, Dorothea andSandra townships, northwestern Ontario: PhoenixBedrock Mapping Project; Ontario Geological Survey, Open File Report 6095, 206 p. Horwood, H.C., Pye, E.G., 1955. Geology of Ashmore Township. In: 60 th Annual Report of the Ontario Department of Mines, Part 5. 139 p. Lafrance, B., DeWolfe, J.C., Stott, G.M., 2004. A structural reappraisal of the Beardmore-Geraldton Belt at the southern boundary of the Wabigoon subprovince, Ontario, and implications for gold mineralization. Can. J. Earth Sci. 41, 217–235. doi:10.1139/E03-090 Lavigne, M.J. 2009. Distribution of gold with respect to lithologies, metamorphic facies and strain state in the Beardmore–Geraldton greenstone belt; Ontario Geological Survey, Open File Report 6241, 88p. Lawley et. al., 2022, https://www.sciencedirect.com/science/article/pii/S2590197422000064 Lemkow, D.R., Sanborn-Barrie, M., Stott, G.M., Percival, J.A, Stone, D., Tomlinson, K.Y., Skulksi, T., McNicoll, V., Davis, D.W., Whalen, J.B. and Hollings, P. 2005. GIS compilation of the geology and tectonostratigraphic assemblages, Wabigoon–Winnipeg River–Marmion transect, western Superior Province, Ontario; Geological Survey of Canada, Open File 4971, Ontario Geological Survey, Miscellaneous Release—Data 187, Scale 1:250 000 Lincoln, L. A., Benoit, K., Keyes, O., Selivanov, D., & Arnold, J. (2018). Fast, consistent tokenization of natural language text. Journal of Open Source Software, 3 (23), 655. Macdonald, A.J., 1988. The Geraldton Gold Camp: The Role of Banded Iron Formation; Ontario Geological Survey, Open File Report 5694, 173 p. Mackasey, W.O., 1975. Geology of Dorothea, Sandra, and Irwin Townships, District of Thunder Bay. Ontario Division of Mines, Geological Report 122, 83 p. Mackasey, W.O., 1976. Geology of Walters and Leduc Townships, District of Thunder Bay. Ontario Division of Mines, Geoscience Report 149. 58 p. Mackasey, W.O., Edwards, G.R., and Cape, D.F. 1976. Legault Township, District of Thunder Bay. Ontario Division of Mines, Preliminary Map P-1191. Mantovani, A., Piana, F., & Lombardo, V. (2020). Ontology-driven representation of knowledge for geological maps. Computers & Geosciences, 139 , 104446. Mark, B., 2024. Geology and Structural Evolution of the Tashota Deformation Zone and Implications for Orogenic Gold Mineralization. Unpublished MSc thesis, Laurentian University, Sudbury, Canada, 191 pp. Martin-Fernandez, J.A., Barceló-Vida l, C. & Pawlowsky-Glahn, V. 2003. Dealing with zeros and missing values in compositional data sets using non-parametric imputation. Mathematical Geology, 35, 253–278.tern Fujian Province, China. Ore Geology Reviews , 71, pp. 502-515 Mason, J.K., McConnell, C.D., 1983. Gold mineralization in the Beardmore-Geraldton area, In: Colvine, A.C. (Ed.), The Geology of Gold in Ontario, Ontario Geological Survey, Misc. Paper 110. pp. 84–97. Mason, J.K., White, G., 1986. Gold occurrences, prospects and deposits of the Beardmore-Geraldton area, District of Thunder Bay and Cochrane. Ontario Geological Survey, Open File Report 5630. 680 p. Palarea-Albaladejo J, Martín-Fernández JA, Buccianti A (2014) Compositional methods for estimating elemental concentrations below the limit of detection in practice using R. J Geochem Explor. ISSN 0375–6742. https://doi.org/10.1016/j.gexplo.2013.09.003. Pavlis, T. L., Langford, R., Hurtado, J., & Serpa, L. (2010). Computer-based data acquisition and visualization systems in field geology: Results from 12 years of experimentation and future potential. Geosphere, 6 (3), 275–294. Percival, J.A., Skulski, T., Sanborn-Barrie, M., Stott, G.M., Leclair, A.D., Corkery, M.T., Boily, M., 2012. Geology and tectonic evolution of the Superior Province, Canada, in: Percival, A.J., Cook, F.A., Clowes, R.M. (Eds.), Tectonic Styles in Canada. Geological Association of Canada, pp. 321–378. Pye, E.G., 1952. Geology of Errington Township, Little Long Lac Area, in: 60th Annual Report of the Ontario Department of Mines. Ontario Department of Mines, 140 p. Rodriguez-Galiano, V., Sanchez-Castillo, M., Chica-Olmo, M., Chica-Rivas, M. (2015). Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines. Ore Geology Reviews 71, 804-818. Rodriguez-Galiano, V.F., Chica-Olmo, M., Chica-Rivas, M. (2014). Predictive modelling of gold potential with the integration of multisource information based on random forest: a case study on the Rodalquilar area, Southern Spain. International Journal of Geographical Information Science 28, 1336-1354. Shanks, W.S. 1993. Geology of Eva and Summers townships, District of Thunder Bay; Ontario Geological Survey, Open File Report 5821, 93p. Silge, J., & Robinson, D. (2016). tidytext: Text mining and analysis using tidy data principles in R. Journal of Open Source Software, 1 (3), 37. Stott, G.M., Davis, D. W., Parker, J.R., Straub, K.J., and Tomlinson K. Y., 2002a. Geology and tectonostratigraphic assemblages,eastern Wabigoon Subprovince, Ontario, Ontario Geological Survey Map P3449, scale 1:250,000, Geological Survey of Canada, Open file 4285 Stott, G.M., Davis, D.W., Parker, J.R., Straub, K.H., Tomlinson, K.Y., 2002b. Geology and Tectonostratigraphic Assemblages, eastern Wabigoon Subprovince, Ontario. Scale 1:250 000. Ontario Geological Survey, Preliminary Map P. 3449. 1 sheet Thurston, P.C. 1980. Geology of the Northern Onaman Lake Area, District of Thunder Bay; Ontario Geological Survey, Report 208, 81p. Accompanied by Map 2411, scale 1:31 680, and Chart A. Tomlinson, K.Y., Hall, R.P., Hughes, D.J., Thurston, P.C., 1996. Geochemistry and assemblage accretion of metavolcanic rocks in the Beardmore–Geraldton greenstone belt, Superior Province. Can. J. Earth Sci. 33, 1520–1533. Tóth, Z., 2019. The geology of the Beardmore-Geraldton belt, Ontario, Canada: geochronology, tectonic evolution and gold mineralization. Unpublished Ph.D. thesis, Laurentian University, Sudbury, Ontario. Tóth, Z., Lafrance, B., Dubé, B., 2023. Oblique lateral extrusion during dextral transpression along the Beardmore-Geraldton belt, Canada. Journal of Structural Geology, 169,104834. Tóth, Z., McNicoll, v., Lafrance, B., Dubé, B., 2022. Early depositional and magmatic history of the Beardmore-Geraldton belt: Formation of a transitional accretionary belt along the Wabigoon-Quetico subprovince boundary in the Archean Superior craton, Canada. Precambrian Res. 371. https://doi.org/10.1016/j.precamres.2022.106579 Wang, J., Zuo, R. and Xiong, Y. (2020). Mapping Mineral Prospectivity via Semi-supervised Random Forest. Natural Resources Research, 29, 189–202. https://doi.org/10.1007/s11053-019-09510-8 Zhang, S., Carranza, E.J.M., Xiao, K. et al. (2021) Mineral prospectivity mapping based on isolation forest and random forest: Implication for the existence of spatial signature of mineralization in outliers. Natural Resources Research , 31, 1981–1999. https://doi.org/10.1007/s11053-021-09872-y Zhang, S., Xiao, K., Carranza, E. J. M., & Yang, F. (2018). Maximum entropy and random forest modeling of mineral potential: analysis of gold prospectivity in the Hezuo-MeiwuDistrict, West Qinling Orogen China. Natural Resources Research , 28(3), 645–664 Zuo, R. (2020). Geodata Science-Based Mineral Prospectivity Mapping: A Review. Natural Resources Research 29, 3415-3424. Zuo, R., Carranza, E.J.M., 2011. Support vector machine: a tool for mapping mineral prospectivity. Comput. Geosci. 37, 1967–1975 Zuo, R., Zhang, Z., Zhang, D., Carranza, E.J.M. and Wang, H. (2015). Evaluation of uncertainty in mineral prospectivity mapping due to missing evidence: a case study with skarn-type Fe deposits in Southwestern Fujian Province, China, Ore Geology Reviews, 71, pp. 502-515 Tables Table 1 – Summary of predictor maps (vectors to mineralization) used for modelling Predictor map Comments Threshold Lithogeochemistry 1 PCA2 Buffered to 1 km >3 stdev 2 PCA3 Buffered to 1 km >3 stdev 3 PCA4 Buffered to 1 km >3 stdev 4 PCA5 Buffered to 1 km >3 stdev Lithology- possible sources of orogenic Au 5 Iron Formations Au known to be associated with Iron formations Buffered to 1 km - binary 6 Lithologic units Au source? multi class 7 Lithological sub-units Au source? multi-class 8 Sedimentary units Au source? multi-class 9 Metamorphic units Au source? multi-class 10 Igneous units Au source? multi-class Structure – possible zones of weakness for the migration of Au bearing fluids 11 Interpreted faults (mag) 60-100° strike direction buffered at 1 km - binary 12 Interpreted faults (mag) 40 – 60° strike direction buffered at 1 km - binary 13 Interpreted faults (mag) 100 – 260° strike direction buffered at 1 km - binary 14 Interpreted faults (mag) 0 – 40° strike direction buffered at 1 km - binary 15 Mapped faults 60 - 100°strike direction buffered at 1 km - binary 16 Mapped faults 40 – 60° strike direction buffered at 1 km - binary 17 Mapped faults 260 – 360° strike direction buffered at 1 km - binary 18 Mapped faults 100 – 260° strike direction buffered at 1 km - binary 19 Dykes All directions buffered at 1 km - binary 20 Lithological contacts All directions buffered at 1 km - binary Geophysics – to define lithology at different depths 21 Density 750m depth Continuous surface 22 Density 4250m depth Continuous surface 23 Density 7250m depth Continuous surface 24 Susceptibility 750m depth Continuous surface 25 Susceptibility 2250m depth Continuous surface 26 Susceptibility 4250m depth Continuous surface 27 Susceptibility 7250m depth Continuous surface 28 Resistivity 750m depth Continuous surface 29 Resistivity 2250m depth Continuous surface 30 Resistivity 4250m depth Continuous surface 31 Resistivity 7250m depth Continuous surface Table 2 – Correlation between MPMs generated from Au mines, developed prospects, occurrences Mines Developed prospects Occurrences All Mines 1 0.4 0.2 .01 Developed prospects 1 .53 .56 Occurrences 1 .81 All 1 Table 3 – Out-of-bag and overall classification accuracies from MPMs generated from mines, developed prospects, occurrences Mines Developed prospects (DVP) Occurrences Out-of-Bag (oob) accuracy 80% 59% 80% Overall classification accuracy 97.4% 100% 90.3% Table 4 – Results from 5-fold repetition of RF modelling (a) variability in best predictors, (b) variability in out-of-bag error, (c) variability in overall classification accuracy Rank Original Repeat 2 Repeat 3 Repeat 4 Repeat 5 1 3 3 3 3 2 2 25 26 26 25 25 3 2 25 25 2 3 4 26 2 2 26 26 5 13 13 13 24 13 6 24 24 8 5 5 a) 3 = geol_sub_unit; 25 = den_7250; 2 = geol_unit; 26 = den_4250; 13 = res_4250; 24 = den_750; 8 = sus_7250; 5 = new_met (see Table 1 for evidence maps) 82.4 81.8 81.7 82.2 82.5 82.1 ave b) Out-of-bag error for 5-fold repetition of RF modelling 94.7 95.2 94.6 94.7 94.7 94.2 94.6 ave c) Overall classification accuracies for the 5-fold repetition of RF modelling Table 5 – top 6 predictors (evidence maps) for MPMs generated from Au mines, developed prospects, and all Au training data Rank Au Mines Au Developed Prospects Au Occurrences 1 7 6 25 2 1 8 2 3 10 4 3 4 11 3 26 5 9 9 24 6 3 25 13 1 = iron formations; 2 = Geological units; 3 = Geological sub-units, 4 = geological sedimentary units; 6 = Geological igneous units; 7 = Susceptibility 750m depth; 8 = Susceptibility 7250m depth; 9 = Susceptibility 4250m depth; 10= Susceptibility 2250m depth; 11= Resistivity 7250m depth; 13 = Resistivity 4250m depth; 24 = dikes; 25 = density 750m depth; 26 = density 7250m depth Table 6 – Summary of Au high prospectivity areas shown on Fig. 22 Location Lithological / Structural Comments Predictions A Mostly metasediments and smaller E-W trending linear volcanic belt Predicts many 7 producing mines (including MacLeod-Cockshutt and Geraldton mines), 2 developed prospects and 14 Au occurrences B Mostly felsic and intermediate volcanic rocks Predicts many Au occurrences along the Paint Lake Fault C E-W striking metasedimentary rocks, strong association with banded iron formations Predicts many Au occurrences and 3 mines (including Leitch mine) D Mainly E-W striking mafic and intermediate volcanic rocks, strong association with banded iron formations Predict many Au occurrences and 1 mine (Northern Empire mine) E Area cored by folded diorite-monzodiorite-granodiorite suite Surrounded by mafic\intermediate volcanic and metasedimentary rocks, associated with shear zones Predict 6 Au occurrences F1 Same geology as E No known Au F2 Mafic/intermediate volcanics Predicts 3 Au occurrences G E-W trending mafic/intermediate volcanic rocks with small belts of felsic volcanics Predicts 3 Au occurrences and 1 developed prospect H Mostly mafic\intermediate volcanic rocks No known Au I Mostly mafic\intermediate volcanic rocks Predicts 3 developed prospects and 1 Au occurrence J Mostly mafic\intermediate volcanic rocks Predict 7 Au occurrences and 1 mine (Consolidated Louanna mine) Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 04 Apr, 2025 Read the published version in Earth Science Informatics → Version 1 posted Editorial decision: Revision requested 02 Dec, 2024 Reviews received at journal 01 Dec, 2024 Reviews received at journal 28 Nov, 2024 Reviewers agreed at journal 27 Nov, 2024 Reviews received at journal 26 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers agreed at journal 25 Nov, 2024 Reviewers invited by journal 25 Nov, 2024 Editor assigned by journal 25 Nov, 2024 Submission checks completed at journal 18 Nov, 2024 First submitted to journal 03 Nov, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5384377","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":385236158,"identity":"65e195b9-3cb9-4dfd-a378-eb30112600cc","order_by":0,"name":"J. R. Harris","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIiWNgGAWjYBACfmb2A8YgBhsD8wEgJSFDUItke08CVAtbAkgLD0EtBj0HDJghTB4DMElYi0RCQnFBzb18Pv4zn1/dqLHgYWA/fHQDPi3mEokHjGccK7Zsk8jdZp1zDOgwnrS0G/i0WM5ISDDmYUswYJPg3WacwwbUIsFjhleLwY0EA2Oef0At/GeeGef8I0bLmQMGxrxtQC0MOcyPc9uI0AIOZN4+kMPSzJhz+yR42Aj5BRiVx4x5viUYyPcffvw551udHD/74WN4tQABmwGMIQEmCSgHAeYHMMYHIlSPglEwCkbBCAQAwMBAoGB9/pQAAAAASUVORK5CYII=","orcid":"","institution":"* - Metal Earth – Lawrentian University","correspondingAuthor":true,"prefix":"","firstName":"J.","middleName":"R.","lastName":"Harris","suffix":""},{"id":385236160,"identity":"c1ac9034-dc76-4eed-98ae-6d53085dff60","order_by":1,"name":"B. Lafrance","email":"","orcid":"","institution":"* - Metal Earth – Lawrentian University","correspondingAuthor":false,"prefix":"","firstName":"B.","middleName":"","lastName":"Lafrance","suffix":""},{"id":385236162,"identity":"8c0bdbeb-3018-4078-b317-e01819bd1ec2","order_by":2,"name":"E. Grunsky","email":"","orcid":"","institution":"University of Waterloo","correspondingAuthor":false,"prefix":"","firstName":"E.","middleName":"","lastName":"Grunsky","suffix":""},{"id":385236164,"identity":"12d5c567-49c8-4ad6-ac95-7b0df3b4e645","order_by":3,"name":"P. Behnia","email":"","orcid":"","institution":"* - Metal Earth – Lawrentian University","correspondingAuthor":false,"prefix":"","firstName":"P.","middleName":"","lastName":"Behnia","suffix":""},{"id":385236166,"identity":"589f19fd-b810-4cc1-9154-ac9a3aebd97d","order_by":4,"name":"M. Nagizedah","email":"","orcid":"","institution":"* - Metal Earth – Lawrentian University","correspondingAuthor":false,"prefix":"","firstName":"M.","middleName":"","lastName":"Nagizedah","suffix":""},{"id":385236168,"identity":"7d4e497b-580d-4d7a-b48c-a2ac7e40f3b1","order_by":5,"name":"C. Lawley","email":"","orcid":"","institution":"Geological Survey of Canada","correspondingAuthor":false,"prefix":"","firstName":"C.","middleName":"","lastName":"Lawley","suffix":""},{"id":385236169,"identity":"81fab42d-4e62-4214-b931-c63127b957d5","order_by":6,"name":"M. Parsa","email":"","orcid":"","institution":"Geological Survey of Canada","correspondingAuthor":false,"prefix":"","firstName":"M.","middleName":"","lastName":"Parsa","suffix":""}],"badges":[],"createdAt":"2024-11-04 03:08:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5384377/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5384377/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s12145-025-01831-y","type":"published","date":"2025-04-04T15:57:39+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":71730188,"identity":"e4f8a3e6-6ccf-4519-ba5f-63b83e8b74e2","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":684989,"visible":true,"origin":"","legend":"\u003cp\u003eStudy area – generalized lithology, structures and mineral deposits\u003c/p\u003e","description":"","filename":"Slide1.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/8127f4130cc4fd2f083b22ab.png"},{"id":71732719,"identity":"7b1e05d0-45a9-4e86-9280-50e15ab87f33","added_by":"auto","created_at":"2024-12-18 07:07:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":380045,"visible":true,"origin":"","legend":"\u003cp\u003eGold mines, developed prospects and occurrences within study area\u003c/p\u003e","description":"","filename":"Slide2.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/d437872d4704fb4784c43d22.png"},{"id":71732147,"identity":"77e4e6a6-6d06-403f-9682-bbae43701356","added_by":"auto","created_at":"2024-12-18 06:59:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":26747,"visible":true,"origin":"","legend":"\u003cp\u003eWeighting system for major Au mines\u003c/p\u003e","description":"","filename":"Slide3.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/4771f675a7ce7d7e855cb1bf.png"},{"id":71730920,"identity":"470dbd13-7b2c-4718-a283-01700a6f0206","added_by":"auto","created_at":"2024-12-18 06:51:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1096116,"visible":true,"origin":"","legend":"\u003cp\u003eLithology descriptors derived from Natural Language Processing (NLP) (a) lithology sub-units, (b) lithology units, (c) sedimentary, metamorphic and igneous units (see text for description)\u003c/p\u003e","description":"","filename":"Slide4.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/3bcad08fae56188a0ec530ee.png"},{"id":71730184,"identity":"7e3ba171-4329-4c01-a04f-d917f0cda90b","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":317010,"visible":true,"origin":"","legend":"\u003cp\u003e(a) buffered (1 km) iron formations, (b) buffered (1 km) contacts\u003c/p\u003e","description":"","filename":"Slide7.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/1379949d2e7332b9f1fa8c05.png"},{"id":71730189,"identity":"c6731503-de84-4627-b7cd-616b56e9f5ab","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":246702,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of the lithogeochemical data\u003c/p\u003e","description":"","filename":"Slide8.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/a4d72e858142d481beee738b.png"},{"id":71730186,"identity":"13e282d1-f431-44ad-8d5d-1a608738b683","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":91225,"visible":true,"origin":"","legend":"\u003cp\u003ePCA scree plot. The plot displays a steep decay of eigenvalues from 1 to 8, which represent processes contained in the data. The lesser eigenvalues likely represent under-sampled processes or random artefacts in the data\u003c/p\u003e","description":"","filename":"Slide9.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/de2852eddfbaa09d35c2379a.png"},{"id":71730193,"identity":"afda514f-27f0-45f1-b846-84e2d46ed6a0","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":230768,"visible":true,"origin":"","legend":"\u003cp\u003eBiplot of PC1 vs, PC2 for a correlation-based metric. There is a contrast between intrusive rocks along the negative portion of PC1 axis andvolcanic rocks along the positive PC1 axis. The negative PC1-PC2 quadrant reflects relative enrichment of Al, S, Si, Fe, K, Cr, Ba, reflecting a mixture of both felsic and mafic intrusive rocks. The positive PC1, negative PC2 quadrant reflects relative enrichment in Mg. The positive PC2 axis shows relative enrichment in chalcophile elements, with the exception of Cu, suggesting that the fewer principal component scores that plot above the zero value of PC2 as possibly associated with alteration systems associated with Au or massive sulphide mineralization\u003c/p\u003e","description":"","filename":"Slide10.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/bd503ee15cb443b5c18089b2.png"},{"id":71732149,"identity":"52ed651e-cd4f-4b11-979c-13a4976bdd4b","added_by":"auto","created_at":"2024-12-18 06:59:36","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":220239,"visible":true,"origin":"","legend":"\u003cp\u003eBiplot of PC2 vs. PC3. A chalcophile trend occurs along the positive PC2 axis, however, there is an inverse association between Sn-Sb with As-Zn-Bi-W-Pb along the PC3 axis, suggesting two different environments\u003c/p\u003e","description":"","filename":"Slide11.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/84553fc708b7a9a07d6cc5d9.png"},{"id":71730196,"identity":"4a69da97-6624-446c-a967-fa5ff458f707","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":241376,"visible":true,"origin":"","legend":"\u003cp\u003eBiplot of PC3 vs. PC4. PC4 shows an association of Bi-Pb-W-Na-K-As-Ti-Zn-Sr and Cr along the negative PC3 axis. This represents an association of chalcophile elements with potential association with mineralization with the Na-K association of felsic (granitoid rocks).\u003c/p\u003e","description":"","filename":"Slide12.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/f253f4cb42a87cab29f63321.png"},{"id":71730201,"identity":"e51a4646-1a5f-47e0-baf8-8d0e046ae74c","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":406051,"visible":true,"origin":"","legend":"\u003cp\u003eA map of the scores for PC1 where the colours of the symbols reflect a contrast between intrusive rocks (negative PC1) with volcanic rocks (positive PC1)\u003c/p\u003e","description":"","filename":"Slide13.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/69152a2478bad85591525ea2.png"},{"id":76591242,"identity":"0954b83c-878a-4d26-9e6d-ea87b8d0e22f","added_by":"auto","created_at":"2025-02-18 17:03:08","extension":"jpg","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":530244,"visible":true,"origin":"","legend":"\u003cp\u003eA map of PC2 scores where positive values reflect relative increases in the chalcophile elements W-Zn-Sb-Bi-Sb-As-Pb-Te. The map also shows the status of the mineral inventory for the area. Developed prospects, past and present producing mines are shown as distinct symbols. There is a clear association of positive PC2 values associated with the Ishkoday [Au] past producer, Paulpic [Au], Headway-Onaman [Zn, Pb], Lynx-Dejour-Reynolds North and South, Jacobus [Cu, Ni], Brookbank [Au], Nortoba-Tyson (No.3 vein)[Mo] prospects. Other elevated values of PC2 are also shown without any close geospatial association with known mineral occurrences\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"fig12.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/3451526cb65f57d00c97e78d.jpg"},{"id":71730202,"identity":"77f892e3-9b6b-435f-8c83-7f17cc64ab27","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":417845,"visible":true,"origin":"","legend":"\u003cp\u003eA map of PC3 scores where negative scores shown a relative increase in values of Sn-Sb and positive scores show a relative increase in values of W-Bi-Zn-As-Pb\u003c/p\u003e","description":"","filename":"Slide15.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/edfe97e14a781e32d1931750.png"},{"id":71730200,"identity":"74f87e4a-1df3-46c2-a24f-a513d7e13d1f","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":423808,"visible":true,"origin":"","legend":"\u003cp\u003eA map of PC4 where there is a contrast between relative enrichment of Sn-Sb-Zn-Cu (negative scores) with Te-Bi-Pb-W (positive scores).\u003c/p\u003e","description":"","filename":"Slide16.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/8cc02a66145c4dafcda73e1f.png"},{"id":71730923,"identity":"08f2b03a-5a8d-438a-8381-7c49b9d3a632","added_by":"auto","created_at":"2024-12-18 06:51:36","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":509770,"visible":true,"origin":"","legend":"\u003cp\u003eAnomalous lithogeochemical points derived from PCA analysis and using upper breakpoints on probability plots of the PCA data – points buffered to 1 km as a zone of influence\u003c/p\u003e","description":"","filename":"Slide17.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/d5c26ec622b3a826707a76f9.png"},{"id":71730197,"identity":"01da96b7-ae47-405e-9032-55e931b304a2","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":420871,"visible":true,"origin":"","legend":"\u003cp\u003eGeophysical data (a) density at the 750m level, (b) susceptibility at the 4250 m level, (c) resistivity at the 2250 m level\u003c/p\u003e","description":"","filename":"Slide18.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/2cdbf8c9748ef68db6c0f6db.png"},{"id":71730926,"identity":"f9c1f0dc-19ba-429c-82e2-203c019a9588","added_by":"auto","created_at":"2024-12-18 06:51:42","extension":"png","order_by":17,"title":"Figure 17","display":"","copyAsset":false,"role":"figure","size":451001,"visible":true,"origin":"","legend":"\u003cp\u003e(a) mapped faults, (b) interpreted faults from airborne magnetic data – all buffered at 1 km as a zone of influence\u003c/p\u003e","description":"","filename":"Slide19.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/b09fc90b44a92ab8c8f01c0e.png"},{"id":71730199,"identity":"79215cfe-7e33-4e62-b9f1-9417d50acc8c","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":18,"title":"Figure 18","display":"","copyAsset":false,"role":"figure","size":872486,"visible":true,"origin":"","legend":"\u003cp\u003eMineral Prospectivity Maps (MPMs), (a) produced using weighted Au mines, (b) produced using Au developed prospects, (c) produced using all Au data – units are probability (0 to 1)\u003c/p\u003e","description":"","filename":"Slide20.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/80d9e5c74a0705609c4f181d.png"},{"id":71730206,"identity":"fb3768b3-091b-4e40-afdb-dc0a3473a256","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":19,"title":"Figure 19","display":"","copyAsset":false,"role":"figure","size":120037,"visible":true,"origin":"","legend":"\u003cp\u003eEfficiency of classification curves\u003c/p\u003e","description":"","filename":"Slide21.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/cb4bb50f3f8764afe3388ea4.png"},{"id":71730922,"identity":"d054a0e8-7e8a-47c8-8cc2-67fb50f74f4d","added_by":"auto","created_at":"2024-12-18 06:51:36","extension":"png","order_by":20,"title":"Figure 20","display":"","copyAsset":false,"role":"figure","size":114535,"visible":true,"origin":"","legend":"\u003cp\u003eEfficiency of classification curves for 3 different methods of defining non-deposit points for generating MPMs – based on A MPM produced using all the Au deposits, prospects and occurrences\u003c/p\u003e","description":"","filename":"Slide22.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/ff1a6a01c2bfd63fa4aef608.png"},{"id":71730205,"identity":"078bcf2d-0504-4c24-9df8-0fa06b17243b","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":21,"title":"Figure 21","display":"","copyAsset":false,"role":"figure","size":745615,"visible":true,"origin":"","legend":"\u003cp\u003eMPMs produced using an ensemble processing of maps based on the mines, developed prospects and occurrences and associated efficiency of classification curves (a) average of all 3 MPMs, (b) based on a weighted average where the Au mines were given greater weight\u003c/p\u003e","description":"","filename":"Slide23.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/f79227bd36dc261cad47ee0c.png"},{"id":71730203,"identity":"3c352ea2-6a4d-4d9a-b0c7-ed5c5abb1857","added_by":"auto","created_at":"2024-12-18 06:43:36","extension":"png","order_by":22,"title":"Figure 22","display":"","copyAsset":false,"role":"figure","size":435745,"visible":true,"origin":"","legend":"\u003cp\u003eTernary map showing the top 5% of the most prospective areas for Au exploration based on the MPMs shown in Figure 18\u003c/p\u003e","description":"","filename":"Slide24.png","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/5b105f4f2c52c180c9a023ac.png"},{"id":80082351,"identity":"2791ce7c-7c2a-4502-a890-2c8b12102ec1","added_by":"auto","created_at":"2025-04-07 16:08:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":9758376,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5384377/v1/ce3a0711-919d-40ce-992f-ffa68240f535.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Orogenic Gold Mineral Prospectivity Mapping (MPM) of the Geraldton Area, Ontario","fulltext":[{"header":"Introduction","content":"\u003cp\u003eMachine learning, a significant aspect of the Artificial Intelligence (AI) phenomenon, has become increasingly common for creating Mineral Prospectivity Maps (MPMs). Mineral Potential Modeling (MPM) involves inputting geoscience data\u0026mdash;represented as predictor (evidence) maps that illustrate geological, geophysical, and geochemical vectors related to mineralization\u0026mdash;into a machine learning algorithm. This process also incorporates training data, which consists of the locations of known deposits, to generate a map highlighting areas with greater potential for mineral exploration for specific commodities (e.g., Au, Cu, Pb, Zn)\u003c/p\u003e \u003cp\u003eMineral Prospectivity Mapping can be approached through two primary methods: data-driven and knowledge-driven modeling. Data-driven methods require a set of known mineral deposits (occurrences) to train a machine learning algorithm (e.g., random forest, neural network) for producing prospectivity maps based on the predictor maps. In contrast, knowledge-driven approaches do not depend on known deposits; instead, each predictor map is assigned an importance weight by the geologist, and these maps are combined using either additive or multiplicative algorithms (e.g., Boolean and index overlay). Both methodologies aim to generate a map that accurately predicts known mineral occurrences while also identifying prospective areas without known mineralization. Validation techniques\u0026mdash;such as ROC and classification efficiency curves, as well as cross-validation\u0026mdash;are crucial to the MPM process and should always be conducted.\u003c/p\u003e \u003cp\u003eNumerous studies have produced MPMs utilizing data-driven methods (Bonham-Carter et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e1988\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Bonham-Carter, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Harris et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2001\u003c/span\u003e; Harris at al, (2006,2008) ; Carranza, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Harris et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2015a\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003eb\u003c/span\u003e ; Rodrigues et al. (2014, 2015), Carranza and Laborte (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2015a\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003ec\u003c/span\u003e), and Harris et al., (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) have identified random forests as a highly effective modeling technique. Consequently, this paper focuses on employing random forests as a data-driven method to generate orogenic gold MPMs. Comprehensive summaries of machine learning algorithms, including random forests, have been provided by various authors (Carranza and Laborte, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2015a\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003eb\u003c/span\u003e; Doan and Foody, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Harris et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ford, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Zuo, \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Zuo and Carranza, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), so a detailed review will not be included here.\u003c/p\u003e \u003cp\u003eThe objectives of this paper are as follows:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eDiscussion of Key Issues\u003c/b\u003e: A variety of issues relevant to the MPM modeling process are addressed throughout this paper, and solutions to these salient challenges are proposed. These include a novel application of natural language processing to generate predictor maps from geological data, methods for defining non-deposit training data for input into the RF algorithm, a weighting method for known gold mines, and a demonstration of combining MPMs in an ensemble approach.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eApplication of Random Forests\u003c/b\u003e: Random Forest (RF) (Breiman, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) is utilized as a machine learning algorithm to produce several MPMs for orogenic gold in the Geraldton area, part of the Abitibi Tectonic Subprovince of Ontario, Canada.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"Study Area","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eRegional Geology\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe granite-greenstone Wabigoon subprovince of the Archean Superior Province is bounded to the south by the metasedimentary Quetico subprovince and to the north by the metasedimentary English River subprovince and the gneissic-granitoid Winnipeg River subprovince (Beakhouse, 1991; Blackburn et al., 1991). The eastern Wabigoon, where the study area is located, extends over more than 180 km east of Lake Nipigon. It consists of two greenstone belts, the Onaman-Tashota greenstone belt to the north and the Beardmore-Geraldton Belt to the south, separated by a major structure called the Paint Lake Fault (\u003cstrong\u003eFig. 1\u003c/strong\u003e). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 1\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe Beardmore-Geraldton Belt is a transitional terrane at the boundary between the Wabigoon and Quetico subprovinces. The belt is East-trending and consists of three metasedimentary rock panels (southern, central, northern sedimentary units), imbricated and in fault contact with three metavolcanic rock panels (southern, central, and northern volcanic units). The southern and northern volcanic units are composed of\u0026nbsp;massive and pillowed basaltic and andesitic flows, whereas the central volcanic unit is dominated by andesitic and dacitic pyroclastic rocks and flows (Shanks, 1993; Tomlinson et al., 1996). They yielded identical crystallization ages of 2724.9 \u0026plusmn; 1.2 Ma obtained from samples of a massive felsic flow in the central volcanic unit and synvolcanic feldspar\u0026ndash;quartz porphyry dike in the northern volcanic unit (Hart et al., 2002). The metasedimentary rocks panels were deposited above the older metavolcanic rocks as fluvial to alluvial fan deposits (northern sedimentary unit), deltaic to subaqueous fan deposits (central sedimentary unit), and deeper water turbidite deposits (southern sedimentary unit) (Devaney and Williams, 1989). The northern sedimentary unit consists of polymictic conglomerate and minor sandstone (Mackasey 1975, 1976; Mackasey et al. 1976). The southern sedimentary unit is dominated by turbiditic sandstone with minor horizons of polymictic conglomerate and banded iron formation, and the central sedimentary unit is transitional between the northern and southern units (\u003cem\u003eibid\u003c/em\u003e). The age of the three metasedimentary units is constrained between ca. 2700 Ma, the age of the youngest detrital zircon population, and ca. 2694 Ma, the age of a cross-cutting feldspar-quartz porphyry intrusion (T\u0026oacute;th et al., 2022). \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe Onaman-Tashota greenstone belt comprises metavolcanic and metasedimentary assemblages of both Mesoarchean and Neoarchean ages. \u0026nbsp;Their ages and descriptions are from Stott et al. (2002a). \u0026nbsp;Along the north side of the greenstone belt, where East-trending metavolcanic rocks border the Ombabika batholith to the south, the Mesoarchean Toronto assemblage (ca. 2922 Ma) occurs as enclaves within the batholith and as narrow slivers of massive to pillowed basaltic flows overlain by quartz porphyritic intermediate to felsic pyroclastic rocks and flows. In the centre of the belt, where the metavolcanic rocks trend roughly North-South in-between the Ombabika batholith and Elbow Lake pluton to the West and the Onaman pluton to the East (\u003cstrong\u003eFig.\u003c/strong\u003e \u003cstrong\u003e1\u003c/strong\u003e), the Mesoarchean Tashota assemblage (ca. 2975 Ma \u0026ndash; ca. 2968 Ma) flanks the Elbow Lake Pluton as a sequence of dacitic felsic tuffs intercalated with pillowed basaltic flows and massive basaltic sills or flows. \u0026nbsp;On the East side of this North-South trending segment, the Onaman pluton is bordered by massive and pillowed basaltic flow and banded iron formation of the Neoarchean Onaman assemblage (ca. 2770 Ma \u0026ndash; ca. 2780 Ma). \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eYounger Neoarchean metavolcanic assemblages overlie those older metavolcanic assemblages. In the centre of the belt, pillowed tholeiitic basaltic flows of the Willet assemblage (ca. 2740 Ma) unconformably overlie or are in fault contact with the Tashota assemblage (Mark, 2024). Intermediate to felsic assemblages of similar age occur along the southern margin (Elmhirst-Rickaby) and northern margin (Marshall) of the belt. The Elmhirst-Rickaby assemblage consists of calc-alkaline basaltic to andesitic flows overlain by dacitic to rhyolitic flows and pyroclastic rocks, whereas the Marshall assemblage consists mainly of calc-alkalic dacitic flows and pyroclastic rocks. Volcanism in the belt continued with the deposition of dacitic to rhyolitic flows and pyroclastic rocks of the Metcalfe-Venus assemblage (ca. 2722 Ma \u0026ndash; ca. 2734 Ma) and quartz porphyritic dacite of the Humboldt assemblage (\u0026lt;2713 Ma), and was capped by the deposition of turbiditic sandstone and conglomerate similar in age (\u0026lt;2710 Ma) to those in the Beardmore-Geraldton Belt. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDeformation of the Beardmore-Geraldton Belt and Onaman-Tashota Belt began in response to south-directed D\u003csub\u003e1\u003c/sub\u003e accretion along the southern margin of the Wabigoon subprovince during the ca. 2690 Ma Shebandowanian Orogeny (Percival et al., 2012). This resulted in the interleaving of metasedimentary and metavolcanic panels of the Beardmore-Geraldton Belt (Devaney and Williams, 1989; Lafrance et al., 2004; T\u0026oacute;th et al., 2023), and in the infolding of metavolcanic rocks between the Ombabika batholith and Onaman pluton in the Onaman-Tashota Belt (Mark, 2024). The latter was coeval with the formation of a foliation (S\u003csub\u003e1\u003c/sub\u003e) and high strain zones at intrusion-metavolcanic rock contacts. Renewed North-South shortening during a second D\u003csub\u003e2\u003c/sub\u003e deformation event (\u0026lt; ca. 2690 Ma) strongly affected the Beardmore-Geraldton Belt and the southern and northern sides of the Onaman-Tashota Belt and produced regional East-West trending F\u003csub\u003e2\u003c/sub\u003e folds with a prominent axial planar S\u003csub\u003e2\u003c/sub\u003e foliation. The intensification of this foliation formed East- to Southeast-trending deformation zones with a sinistral transcurrent component, such as the Humboldt Bay deformation zone (Fig. 1) in the Onaman-Tashota Belt (Culshaw et al., 2006) and the Paint Lake Fault (Fig. 1) and the Tombill-Bankfield deformation zone in the Beardmore-Geraldton Belt (T\u0026oacute;th et al., 2023). \u0026nbsp;Later D\u003csub\u003e3\u003c/sub\u003e dextral transpression reactivated these deformation zones and East-trending lithological contacts as dextral transcurrent faults and was more pronounced in the Beardmore-Geraldton Belt where it produced multiple outcrop- and map-scale Z-shaped F\u003csub\u003e3\u003c/sub\u003e folds and a regional S\u003csub\u003e3\u003c/sub\u003e cleavage oriented anticlockwise to bedding and S\u003csub\u003e2\u003c/sub\u003e foliation (Devaney and Williams, 1989; Lafrance et al., 2004; T\u0026oacute;th et al., 2023). Regional greenschist facies metamorphism affected the Beardmore-Geraldton Belt and was higher in the Onaman-Tashota Belt, where most metasedimentary and metavolcanic rocks were metamorphosed to the greenschist-amphibolite facies transition and amphibolite facies (Haataja, in prep.). \u0026nbsp;As all metasedimentary and metavolcanic rocks are metamorphosed, the prefix \u0026ldquo;meta\u0026rdquo; is dropped for brevity. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFelsic magmatism in the Onaman-Tashota Belt range in age from ca. 2922 Ma to ca. 2680 Ma (Stott et al., 2002a). \u0026nbsp;It occurred during volcanism with the emplacement of the syn-volcanic ca. 2777 Ma Onaman tonalite in the Onaman assemblage and the ca. 2731 Ma \u0026ndash; ca. 2738 Elmhirst pluton in the Elmhirst-Rickaby assemblage. Syn to late tectonic, ca. 2700 Ma \u0026ndash; ca. 2680 Ma, calc-alkalic to sanukitoid plutons, cut across the S\u003csub\u003e1\u003c/sub\u003e foliation in the Onaman-Tashota Belt and Beardmore-Geraldton Belt and provide minimum and maximum ages for the D\u003csub\u003e1\u003c/sub\u003e and D\u003csub\u003e2\u003c/sub\u003e events, respectively. Proterozoic olivine diabase dikes and sills (ca. 1109 Ma) associated with the mid-continental rift are abundant close to Lake Nipigon. \u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGold Mineralization\u003c/em\u003e\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGold deposits in the Beardmore-Geraldton Belt are located within two main areas: the eastern Geraldton camp near the town of Geraldton and the western Beardmore camp immediately east of Lake Nipigon. \u0026nbsp;The Bankfield-Tombill deformation zone hosts most of the deposits in the Geraldton camp including, from west to east, the Tombill, Bankfield, Consolidated Magnet, Consolidated Mosher Longlac, MacLeod-Cockshutt and Hard Rock deposits. \u0026nbsp;The Hard Rock and McLeod-Cockshutt deposits were two of the three most prolific mines along the Bankfield-Tombill deformation. They produced 269,081 and 1,366,404 ounces of gold, respectively, from 1938 to 1966 (Mason and McConnell, 1983; Mason and White, 1986), and are being redeveloped as an open pit mine that will produce 5 million ounces of gold over the next 14 years (https://www.equinoxgold.com). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe Bankfield-Tombill deformation zone is within the top northern section of the southern sedimentary unit, where intercalated gabbro, mafic volcanic rocks, feldspar-quartz porphyry, turbiditic sandstone, conglomerate and banded iron formation, are folded by tight isoclinal F\u003csub\u003e2\u003c/sub\u003e folds and transposed parallel to the strong east-southeast-striking foliation (S\u003csub\u003e2\u003c/sub\u003e) that defines the Bankfield-Tombill deformation zone. \u0026nbsp;Gold mineralization is present within all rock types and is preferentially associated with quartz-carbonate veins along sheared east-southeast-striking contacts between feldspar-quartz porphyry and sandstone, with quartz-carbonate veins and sulphide (pyrite, arsenopyrite) replacement lenses in banded iron formation, and with quartz-carbonate veins in massive sandstone and feldspar-quartz porphyry (Horwood and Pye, 1955; Pye, 1952; Anglin, 1987; Macdonald, 1988; Lavigne, 2009; T\u0026oacute;th, 2019). The quartz-carbonate veins were deposited during the D\u003csub\u003e1\u003c/sub\u003e event and folded during the D\u003csub\u003e2\u003c/sub\u003e event (T\u0026oacute;th, 2019). They are surrounded by sericite-carbonate-pyrite \u0026plusmn; albite\u0026ndash;rutile alteration halos. Other quartz-carbonate veins with similar alteration haloes and tourmaline-rich veins with carbonate-tourmaline-pyrite alteration halos were emplaced during the D\u003csub\u003e2\u003c/sub\u003e event (T\u0026oacute;th, 2019). \u0026nbsp;The deposition of these veins collectively produced a large, up to 250 m wide, sericite-carbonate alteration envelope surrounding the deposit and characterized by elevated S, Te, As, W, and Bi. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLeitch, Sand River, and Northern Empire are the main deposits and past-producing mines in the Beardmore camp. \u0026nbsp;The Leitch and Sand River mines collectively produced 897,356 ounces of gold over 29 years of production from 1936 to 1965 (Ferguson, 1967). \u0026nbsp;Gold is present in quartz-carbonate veins with wall-parallel, sericitic and chloritic, internal laminae and wallrock selvedges. The veins are in sandstone and minor iron formation of the southern sedimentary unit in the hinge area of a regional Z-shaped F\u003csub\u003e3\u003c/sub\u003e fold (Lafrance et al., 2004). The latter is defined by the folded contact between the southern sedimentary unit and volcaniclastic rocks of the central volcanic unit (Lafrance et al., 2004). The veins are grouped in two sets based on their orientation (Ferguson, 1967). One set consists of straight to curviplanar, continuous veins, which occupy eastnortheast-striking D\u003csub\u003e3\u003c/sub\u003e dextral shear zones that are parallel to the axial plane of the F\u003csub\u003e3\u003c/sub\u003e fold (Lafrance et al., 2004). The second set consists of strongly folded veins oriented roughly perpendicular to the axial plane of the F\u003csub\u003e3\u003c/sub\u003e fold. 20 km to the east, the Brookbank zone, which has an indicated resource of 600,000 ounces of gold (https://www.equinoxgold.com), was emplaced along similar dextral shear zones at the contact between polymictic conglomerate of the northern sedimentary unit and basalt of the northern volcanic unit (DeWolfe et al., 2007). Little information is available on the Northern Empire mine, which is located 1 km eastnortheast of the town of Beardmore. The mine produced 149,492 ounces of gold from steeply dipping, eastnortheast-striking quartz veins in sheared mafic pillowed volcanic rocks of the southern volcanic unit (Mason and White, 1986). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMost gold occurrences in the Onaman-Tashota Belt are located in the Onaman assemblage (Tashota-Nipigon mine) and Elmhirst-Rickaby assemblage (Brenbar mine, Sturgeon River mine). The Tashota-Nipigon mine produced 12,356 ounces of Au, 14,527 ounces of Ag, and 575,000 lbs of Cu from eastnorteast-striking quartz-pyrrhotite-chalcopyrite veins and ore zones plunging 55\u0026deg;-60\u0026deg; to the northwest (Thurston, 1980). The Brenbar mine is hosted within intermediate to felsic volcaniclastic rocks of the Elmhirst-Rickaby assemblage and produced 134 ounces of Au from ENE-striking and steeply-dipping laminated quartz veins surrounded by alteration haloes of sericite, carbonate, and pyrite (Mackasey, 1975). The Sturgeon River Mine, which is located 2 km east of the Brenbar mine, produced 73,738 ounces of Au and 15,922 ounces of Ag from NNE-striking laminated quartz veins surrounded by alteration haloes of sericite and chlorite (Mackasey, 1975). Other smaller occurrences and artisanal mines from the early 1900\u0026rsquo;s are hosted by the roughly north-striking Tashota deformation zone in the centre of the Onaman-Tashota Belt. \u0026nbsp;The Tashota deformation zone straddles the contact between the Ombabika batholith and volcanic rocks of the Tashota and Willet assemblages. Two of these occurrences, the Adair and Wascanna prospects, consists of quartz-chlorite \u0026plusmn; muscovite \u0026plusmn;ankerite veins, which are surrounded by alteration halos of chlorite, sericite and ankerite and are isoclinally folded parallel to the strong S\u003csub\u003e1\u003c/sub\u003e foliation along the deformation zone (Mark, 2024).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn summary, gold mineralization is preferentially located in zones of structural complexity, such as the hinge of F\u003csub\u003e2\u003c/sub\u003e and F\u003csub\u003e3\u003c/sub\u003e folds, and lithological complexity due to shearing along lithological contacts. It is also located in east-northeast- to east-southeast shear zones and deformation zones in association with quartz-carbonate veins of similar orientation. The majority of Au mines, developed prospects and occurrences are found in the Beardmore- Geraldton greenstone belt (\u003cstrong\u003eFig. 2\u003c/strong\u003e) \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eINSERT FIG. 2\u003c/em\u003e\u003c/p\u003e"},{"header":"Methodology and Experiments (objective 1)","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 1 \u0026ndash; weighting of gold mines\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHaving a sufficient number of training points (e.g., Au mines) for input to machine learning algorithms is an important issue discussed by many different authors (Carranza and Laborte, and references therein, 2015c; Zuo et al, 2015). Following the methodology presented by Harris et al (2006, 2008) we spatially weight the three most prominent gold mines in the study area by the number of ounces of Au produced before input to the modelling exercises. By adding additional points surrounding (within 200 m) each mine. We have also increased the number of training points from 12 to 21 thus providing a more robust training data set for modeling purposes. We then compare and contrast how well these MPMs perform using efficiency of validation curves (Chung and Fabri, 2003; Harris et al, 2006, 2008). We examine how well each MPM predict weighted Au mines, developed prospects and occurrences.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 2 \u0026ndash; production of random non-deposit sites for RF modelling\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe also investigate a number of different methods for generating non-deposits required for RF modeling. The methods we propose in this paper include the following:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eSelect a random number of non-deposit points equal to the number of Au mines\\prospects\\occurrences (total of 189) without geographic restrictions.\u003c/li\u003e\n \u003cli\u003eSelect a random number of points equal to the number of Au mines\\prospects\\occurrences (total of 189) with geographic restrictions (\u0026gt; 2 km from a known Au mine).\u003c/li\u003e\n \u003cli\u003eSelect a random number of Au mines\\prospects\\occurrences (total of 189) but geographically restricted to lithologies not known to contain Au occurrences (low potential lithologies).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 3 \u0026ndash; Ensemble classification\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn keeping with the theory of ensemble classification (RF is an ensemble classifier by nature), we conduct a 5-fold repletion of our RF modeling using a separate set of randomly selected non-deposit points (total of 189 matching the total number of weighted Au mines, prospects and occurrences), greater than 2 km from a known mine for each repetition. This enables us to bracket the variability in a 5-fold repetition of the RF modelling. The variability refers to the \u003cem\u003eoob\u0026nbsp;\u003c/em\u003eand overall accuracies as well the best predictors. We then take both the average and weighted average of the 5 MPMs generated from the 5-fold repetition and compare, using overall accuracies and efficiency of classification curves, how well they predict the known Au mines, developed prospects and occurrences.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 4 \u0026ndash; Natural Language processing\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe also introduce the concept of Natural Language Processing (NLP) to extract and apply unstructured and semi-structured text data for the purposes of RF modelling. Bedrock geology maps represent an important source of geoscientific information, including rock types, geological ages, mineralogy, textures, and cross-cutting relationships that are essential for mapping the different component of gold-bearing mineral systems. These unstructured forms of geoscientific text data, which commonly occur in large volumes, are essential to the map-building process because they contain the concepts and observations underpinning the preferred map representation (Brodaric et al., 2004; Pavlis et al., 2010; Mantovani et al., 2020). For the Geraldton map area, text data is contained within attributes that are linked to geology polygons in GIS (i.e., \u0026quot;UNITNAME_P\u0026quot;, \u0026quot;ROCKTYPE_P\u0026quot;, \u0026quot;SUPEREON_P\u0026quot;, \u0026quot;EON_P\u0026quot;, \u0026quot;ERA_P\u0026quot;, \u0026quot;PERIOD_P\u0026quot;, \u0026quot;EPOCH_P\u0026quot;, \u0026quot;RegionClas\u0026quot;, \u0026quot;SubClass\u0026quot;, \u0026quot;RAge\u0026quot;). Text data from each of these attributes was concatenated prior to applying the NLP methodology described in Lawley et al. (2022): (1) text data were converted to word tokens using the \u0026ldquo;tidytext\u0026rdquo; (Silge and Robinson, 2016) and \u0026ldquo;tokenizer\u0026rdquo; (Lincoln, et al., 2018) packages in R; (2) uninformative word tokens were removed using the tidytext list of English \u0026ldquo;stop words\u0026rdquo; (n = 1149) and a list of North American place names; (3) plural terms were relaced using the Harman (1991) method to focus text analysis on root words; and (4) numbers and word tokens with two or fewer characters were removed from further analysis.\u003c/p\u003e\n\u003cp\u003eEach of the processed word tokens were then joined with a geoscience Global Vectors for Word Representation (GloVe) model (Lawley et al., 2022). This model was re-trained on public geoscientific documents from Canada and a subset of international and peer-reviewed publications, and has been shown to outperform more complex language models on geoscience analogy, clustering, relatedness, and nearest neighbour tasks (Lawley et al. 2022). Individual word vectors were then averaged to generate one word embedding for each map polygon. The cosine similarity between each map polygon embedding and the terms \u0026ldquo;igneous\u0026rdquo;, \u0026ldquo;metamorphic\u0026rdquo;, and \u0026ldquo;sedimentary\u0026rdquo; from the geoscience GloVe model were used as input for RF modelling. Unlike most previous studies that use text as a form of categorical data or apply one-hot encodings to create binary predictor maps, cosine similarities represent a continuous variable (-1 to 1) and are based on all of the available geoscientific information. Cosine similarities represent a form of semantic search and can be applied to map the most likely sources and traps of mineral systems.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData processing for producing MPMs (objective 2)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eRandom Forest Algorithm\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRandom forest (Breiman, 2001) is based on producing a forest of individual decision trees and data (predictor maps) are passed through the trees to create a final mineral prospectivity map (MPM) through majority vote for either the presence or absence of a deposit. Two aspects of randomness are introduced to the algorithm one through bootstrapping and the other through permutation of the predictor maps passing through each tree. The user defines the number of trees to build as well as the percentage of training samples to use for the construction of each tree. Typically for each tree 2/3 of the training data are selected with replacement for prediction and 1/3 are left for validation (\u003cem\u003eboot strapping\u003c/em\u003e). This process allows for an \u003cem\u003eoob\u003c/em\u003e (\u003cem\u003eout-of-bag-accuracy\u003c/em\u003e) to be calculated. Furthermore, the geologist chooses the number of predictor maps that are used to pass through each tree in the validation and information gain calculation process that is used to rank the importance of the variables. Typically, the square root of the number of predictor maps are passed down each tree. The trees are kept short although the trees are not pruned. This characteristic, as well as bootstrapping, helps prevent overfitting the model.\u003c/p\u003e\n\u003cp\u003eFurthermore, RF provides a measure of importance for each predictor map. In this study we used a technique called permutation importance whereby a baseline classification accuracy is established by passing \u003cem\u003eout-of-bag (oob)\u0026nbsp;\u003c/em\u003esamples through the random forest. The column of a single predictor map is permuted and then the samples are run through the random forest. The importance of a predictor map is determined by noting the difference between the baseline accuracy and the drop in accuracy by permuting the column. This procedure is computationally more complex but the results are mor reliable.\u003c/p\u003e\n\u003cp\u003eFinally, an MPM is produced by calculating the probability of an Au deposit at a given location (pixel) by dividing the number of votes for Au by the total number of trees. For example, if a pixel location had 1 vote for Au and 10 trees the probability would be 1/10 which equal 0.01. Conversely, if the pixel had 10 votes that probability would be 10/10 which equals 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGeneration of training data for input to RF\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGold was divided into 3 categories: occurrences (total of 166), developed prospects (11) and mines (12). However, following the methodology proposed by Harris et. (2006, 2008) we have weighted the mines that have produced more than 500,000 ounces for gold. These 3 mines are the Leitch mine (1,366.404 oz Au), the Beardmore mine (861,982 oz Au) and the Geraldton mine (605,449 oz Au). These mines developed prospects and occurrences are shown in \u003cstrong\u003eFigure 2.\u003c/strong\u003e Each mine was weighted by adding an additional point within 200m of the actual location of the mine as illustrated in \u003cstrong\u003eFigure 3\u003c/strong\u003e. Thus, Leitch received a weight of 4 additional points, Beardmore, 3 points and Geraldton, 2 points. In addition to weighting these mines by importance, addition of these extra points increased the number of points for Au mines from 12 to 21, a number felt adequate for training and RF modelling. In addition, RF requires a set of non-deposit points equal in number to the number of Au mines, prospects and occurrences. The different methods of generating these randomly selected points are discussed below.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 3\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGeneration of Predictor maps for input to RF\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirstly, predictor maps (total of 31) for predicting orogenic Au occurrences are prepared for modelling. \u003cstrong\u003eTable 1\u003c/strong\u003e provides a summary of the predictor maps used in this paper.\u003c/p\u003e\n\u003cp\u003eINSERT Table 1\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eLithology\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNatural language processing was used to extract textual information from the existing lithological map and descriptive legend, resulting in 5 predictor maps: (1) lithologic units modelled as categorical data (n = 13); (2) lithologic sub-units modeled as categorical data (n = 13); and (3) the cosine similarity between the word embeddings of each map polygon and \u0026ldquo;sedimentary\u0026rdquo;, \u0026ldquo;metamorphic\u0026rdquo; and \u0026ldquo;igneous\u0026rdquo; terms included within the geoscience GloVe model. Cosine similarities were added to the RF modelling as continuous variables. These five predictor maps are shown in \u003cstrong\u003eFigure 4a,b\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;c\u003c/strong\u003e. Overall, the NLP methodology is based on 8447 words, corresponding to approximately 23 tokens for each map polygon. Archean (n = 686), rock (n = 568), mafic (n = 447), volcanic (n = 428), and Precambrian (n = 361) represent the most frequently used words. Gold occurrences tend to be associated with mafic volcanic rocks and/or carbonaceous sedimentary rocks, which is expected given that gold is often transported as sulphide complexes that become de-stabilized during interaction with iron- or carbon-rich lithologies. Iron formation-hosted gold deposits in the southern Beardmore greenstone belt represent an important example of the relationship between gold mineralization and suitable trap rocks. Thus, iron formations were extracted from the lithology map and buffered to 1 km representing a zone of influence on alteration/mineralization determined from geological field observations (\u003cstrong\u003eFig. 5a\u003c/strong\u003e). The same is true for lithologic contacts; they were extracted from the lithologic map and buffered to 1 km (\u003cstrong\u003eFig. 5b\u003c/strong\u003e), representing possible channel ways for mineralized Au-bearing fluids.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eINSERT FIG.4 and 5.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGeochemistry\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRock-sample analysis data of 652 samples was extracted from the Metal Earth database. Major elements (Si, Al, Fe, Mn, Mg, Ca, Na, K, Ti, P, S) and trace elements (Ag, Au, Cu, Pb, Zn, Ni, Co, Cr, Hg, Sr, Ba, Sn, Sb, Mo, Te, Bi, W, and As) were selected based on their importance in gold mineralization and availability of data. \u003cstrong\u003eFigure 6\u003c/strong\u003e shows the spatial distribution of the lithogeochemical data. The data is geographically restricted to the central/east portion of the study area. This spatial distribution will have a negative effect on the RF modelling which will be discussed later. Oxides first were transformed to element content and then all data were transformed to ppm. The log-ratio Expectation-Maximization method (IrEM function, implemented in zCompositions R package) was used to impute the missing and below detection limit (BDL) values (Mart\u0026iacute;n-Fern\u0026aacute;ndez, et al., 2003; Palarea-Albaladejo et al., 2014). IrEM algorithm replaces BDL values with a value between 0 and the lower detection limit of each analysis for elements with \u0026lt;40% missing values. Applying the algorithm resulted in a complete geochemical dataset with 27 elements and 621 samples. Au and Ag were removed from the analysis because of containing \u0026gt;40% missing values. To deal with the closure problem of compositional data a centered log-ratio (clr) transformation (Aitchison, 1986) was applied on the imputed data.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 6\u003c/p\u003e\n\u003cp\u003ePrincipal component analysis (PCA) was applied to the logcentred transformed imputed data (621 analyses) using measures of correlation and covariance. The principal component screeplot of Figure 7 indicates a steep decay of values from principal component 1 through 5. This reveals that most of the significant correlations and variability of the elements occurs within the first five components. The lesser eigenvalues likely represent under-sampled processes or random artefacts in the data (Grunsky, 2010).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 7\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 8\u003c/strong\u003e shows a biplots of the PC1 vs. PC2. The principal component loadings (elements) are coloured according to a generalized Goldschmidt classification. The biplot shows a contrast between intrusive rocks along the negative portion of PC1 axis and volcanic rocks along the positive PC1 axis. The negative PC1-PC2 quadrant reflects relative enrichment of Al, S, Si, Fe, K, Cr, Ba, reflecting a mixture of both felsic and mafic intrusive rocks. The positive PC1, negative PC2 quadrant reflects relative enrichment in Mg. The positive PC2 axis shows relative enrichment in chalcophile elements, with the exception of Cu, suggesting that the fewer principal component scores that plot above the zero value of PC2 as possibly associated with alteration systems associated with Au or massive sulphide mineralization.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 8\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 9\u003c/strong\u003e shows a biplot of PC2 vs. PC3 where there is a chalcophile trend along the positive PC2 axis, however, there is an inverse association between Sn-Sb with As-Zn-Bi-W-Pb along the PC3 axis, suggesting two different environments.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 9\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 10\u003c/strong\u003e shows a biplot of PC3 vs.PC4. An association of Bi-Pb-W-Na-K-As-Ti-Zn-Sr and Cr occurs along the negative PC3 axis. This represents an association of chalcophile elements with potential association with mineralization with the Na-K association of felsic (granitoid rocks).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 10\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 11\u003c/strong\u003e shows a map of the scores for PC1, derived from the correlation-based PCA, where the colours of the symbols reflect a contrast between intrusive rocks (negative PC1) with volcanic rocks (positive PC1).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 11\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 12\u003c/strong\u003e shows a map of PC2 scores, derived from the correlation-based PCA, where positive values reflect relative increases in the chalcophile elements W-Zn-Sb-Bi-Sb-As-Pb-Te. The map also shows the status of the mineral inventory for the area. Developed prospects, past and present producing mines are shown as distinct symbols. There is a clear association of positive PC2 values associated with the Ishkoday [Au] past producer, Paulpic [Au], Headway-Onaman [Zn, Pb], Lynx-Dejour-Reynolds North and South, Jacobus [Cu, Ni], Brookbank [Au], Nortoba-Tyson (No.3 vein) [Mo] prospects. Other elevated values of PC2 are also shown without any close geospatial association with known mineral occurrences.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 12\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 13\u003c/strong\u003e shows a map of PC3 scores [correlation-based PCA] where negative scores show a relative increase in values of Sn-Sb and positive scores show a relative increase in values of W-Bi-Zn-As-Pb\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 13\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 14\u003c/strong\u003e shows a map of PC4 where there is a contrast between relative enrichment of Sn-Sb-Zn-Cu (negative scores) with Te-Bi-Pb-W (positive scores).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 14\u003c/p\u003e\n\u003cp\u003eBased on the above analysis we selected components 2,3,4 and 5 to be used as input to the RF algorithm. However, because the data was not equally spread throughout the study area, we chose not to interpolate the data but rather buffer each sample point (shown in Fig. 6) to 1 km acting as a zone of influence. This approach has been previously used by Harris et al. (2022). The anomalous samples (\u003cstrong\u003eFig.15\u003c/strong\u003e) were selected greater than 3 standard deviations above the population mean for each component. As seen in \u003cstrong\u003eFigure 15\u003c/strong\u003e, all the anomalous PCA geochemical sample points fell in the central-west portion of the study area and missing in the east portion of the study area.\u003c/p\u003e\n\u003cp\u003eBased on the above analysis we selected components 2,3,4 and 5 to be used as input to the RF algorithm. However, because the data was not equally spread throughout the study area, we chose not to interpolate the data but rather buffer each sample point (shown in Fig. 6) to 1 km acting as a zone of influence. This approach has been previously used by Harris et al. (2022). The anomalous samples (\u003cstrong\u003eFig.15\u003c/strong\u003e) were selected greater than 3 standard deviations above the population mean for each component. As seen in \u003cstrong\u003eFigure 15\u003c/strong\u003e, all the anomalous PCA geochemical sample points fell in the central-west portion of the study area and missing in the east portion of the study area. This will have negative effects on how useful the lithogeochemical data will have on the modelling procedure using the RF algorithm (incomplete coverage).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 15\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGeophysics\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe geophysical data analytic was carried out on 3D density, susceptibility, and resistivity models. The 3D density model was inverted from the gravity Bouguer anomaly (GSC, 2020) on a horizontal grid of 1x1 km. The 3D susceptibility model was generated by inverting the total magnetic anomaly (GSC, 2021) on a horizontal grid of 250x250 m. The Magneto-Telluric (MT) data acquired as part of the Metal Earth project was inverted to generate the 3D resistivity model on a horizontal grid of 1.5x1.5 km. The vertical grid of the 3D density, susceptibility, and resistivity models was exponentially increased toward the deeper parts of the model. All of the geophysical inversion tasks were carried out by imposing model smoothness constraints. Next, all three of the 3D models were re-sampled/interpolated into a regular grid of 500x500x500 m to generate co-located volumes of density, susceptibility, and resistivity suitable for analysis\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 16\u003c/strong\u003e shows example of 3 geophysical images used in the RF modelling: (a) density from the surface to 750m deep, (b) susceptibility \u0026ndash; 4250m deep and (c) resistivity \u0026ndash; 2250m deep. Not all the geophysical images (Table 1) used in RF modelling are shown to save on space.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 16\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFaults\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTwo sets of faults were used: mapped faults and interpreted faults from the airborne magnetic data. These faults were separated into cardinal strike directions and buffered using a 1 km zone of influence based on field observations with respect to the distance to which alteration effects could be mapped. \u003cstrong\u003eFigure 17\u003c/strong\u003e shows the mapped and interpreted faults buffered to a distance of 1 km. as a zone of influence based on field observations of significant alteration. Although they are much younger (Proterozoic) buffered dikes (not shown) were used as were the faults as possible channels for Au-bearing fluids. The prominent Paint Lake Fault (\u003cstrong\u003eFig 1\u003c/strong\u003e) is also labelled on \u003cstrong\u003eFigure 17\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 17\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 1 \u0026ndash; MPMs based on Au mines, developed prospects and occurrences\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 18\u003c/strong\u003e presents the 3 MPMs generated from Randon Forests; (a) based on weighted mines (21), developed prospects (11) and all Au mines, prospects and occurrences (166). The correlations between these maps are low (\u003cstrong\u003eTable 2\u003c/strong\u003e) indicting overall spatial dissimilarities. The highest correlation is between the MPM based on the occurrences vs. all occurrences (0.81). This is to be expected due to the greater number of occurrences when comparing to all the Au points.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 18\u003c/p\u003e\n\u003cp\u003eIt is apparent that the Beardmore greenstone belt is more prospective for gold than the Geraldton belt when assessing the MPM based on mines. The out-of-bag (\u003cem\u003eoob\u003c/em\u003e) error rates and accuracy of classification are listed in \u003cstrong\u003eTable 3\u003c/strong\u003e. The MPM based on developed prospects shows poor \u003cem\u003eoob\u003c/em\u003e accuracy but perfect classification accuracy. This is in part due to an insufficient number of training samples (11) and, therefore, the DVP MPM is not robust.\u003c/p\u003e\n\u003cp\u003eINSERT TABLE 2\u003c/p\u003e\n\u003cp\u003eINSERT TABLE 3\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 19\u003c/strong\u003e presents efficiency of classification curves for the various combinations of gold and derived MPMs. The best performing models in order of prediction rate, as one might expect, involves the MPM based on the Au mines predicting the mines, followed by the developed prospects (DVPs) MPM predicting the DVPs and then the MPM based on the Au occurrences predicting the occurrences. All have areas under the curves more than 0.9 which represent good prediction results. With respect to mines approximately 92% of the mines are predicted in 2% on the most prospective areas whereas the prediction results for the DVP at 92% of the deposits is 12% of the most prospective areas and the occurrences are 14% of the area.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 19\u003c/p\u003e\n\u003cp\u003eHowever, when considering whether the MPMs based on the mines DVPs and occurrence; they are not highly predictive of the Au points not involved in producing the MPM (Fig. 19) For example, the MPM based on the mines do not predict the DVPs nor occurrences very well. The same result (low prediction rate) based on the DCP MPM versus the Au occurrences, can be seen in Figure 19. These prediction rates are all characterized by less than 0.7 area under the efficiency of classification curve. The prediction rate for the above MPMs is not much better than random.\u003c/p\u003e\n\u003cp\u003eThe prediction rate for the MPM based on the occurrences versus the mines and DVPs is moderately better yielding areas under the curve of .887 and .873, respectively. The prediction rate for the DVP MPM versus mines is only moderate at best.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 2 \u0026ndash; Non-deposit point selection \u0026ndash; MPMs based on 3 different methods for calculating non-deposit Au points for RF modelling\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 20\u003c/strong\u003e shows the efficiency of classification curves for the 3 different methods of defining random non-deposit points discussed above. The curves are based on an MPM produced using all Au training data, DVPs and mines (total of 189) produced using a 5-fold repetition of the RF modelling as mentioned above. This was in part done to address how variable the results are from individual runs of the RF modelling and also to produce 5 MPMs that could be combined in an ensemble fashion (discussed in the next session).\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 20\u003c/p\u003e\n\u003cp\u003eThe curves show that the MPM (not shown) based on all the Au data when predicting the mines are very similar and quite strong from a predictive point of view with areas under the curve greater than 0.9. It is interesting to note that best MPM in terms of prediction of all the Au data was based on the selection of non-deposit points restricted so that none fell within 2 km of an Au point. This was followed by non-deposit points restricted to lithologies not expected to contain orogenic Au, followed by non-deposit points selected randomly without any spatial restrictions. In fact, the MPM based on all the Au data was a strong predictor of mines, DVPs and all the Au data.\u003c/p\u003e\n\u003cp\u003eWith respect to the variability in results from the 5-fold repetition of the RF modelling, \u003cstrong\u003eTable 4\u003c/strong\u003e shows the variability in the out-of-bag (oob) error, overall classification accuracy and strongest predictors (predictor maps) over 5 repetitions.\u003c/p\u003e\n\u003cp\u003eINSERT TABLE 4\u003c/p\u003e\n\u003cp\u003eWith respect to out-of-bag and overall classification accuracies, they are very similar and stable over the 5-fold repetition, as would be expected given the RF is based on random selection of training areas (boosting) and predictor maps (bagging) for each run of RF. However, there is some variability in the best predictors over the 5-fold repetition. The geological-sub units and density at the 750m and 7250m levels are important predictors in all runs of RF. The important predictor maps will be further discussed below.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 3 \u0026ndash; Ensemble Methods\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing the 5-fold repetition,5 MPMs were generated using all the Au training points (189). Two ensemble methods were used to create combined MPMs:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eMethod 1\u003c/em\u003e \u0026ndash; average = (mines + developed prospects + occurrences)/3\u0026hellip;\u0026hellip; \u003cem\u003eeqn1\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eMethod 2\u003c/em\u003e \u0026ndash; weighted average = ( (mines * 3) + developed prospects + occurrences)/3 \u0026hellip;.. \u003cem\u003eeqn2\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThese MPMs are shown in \u003cstrong\u003eFigure 21\u003c/strong\u003e along with their associated efficiency of classification curves. In both maps the MPMs based on the mines showed the greatest prediction rates. The MPM based on the average of the 5 MPMs predicts 84% of the mines in 2% of the most prospective areas and 100% of the mines in 25% of the area. These prediction results improve for the weighed average MPM (as would be expected) resulting in again, 84% of the mines predicted in the top 2% of the area of the MPM but 100% of the mines in only 10% of the area. The results are less strong for the Au developed prospects and Au occurrences as well as all the 189 Au training points.\u003c/p\u003e\n\u003cp\u003eINSERT FIG. 21\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExperiment 4 \u0026ndash; NLL\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe natural language processing was successful as the predictor maps resulting from this processing were among the most important predictors (geological units, geological sub-units \u0026ndash; (see \u003cstrong\u003eTable 1 and 3a\u003c/strong\u003e for listing of predictor maps). One can see from Table 3a that the NLL has produced important predictor maps of orogenic gold especially geologic sub-units.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eBest Predictors\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere is some variability in the best predictors, as would be expected due to the random nature of the RF modelling algorithm, between all experiments. This has already been seen in Table 4 discussed above\u003cem\u003e.\u0026nbsp;\u003c/em\u003eIn\u003cem\u003e\u0026nbsp;\u003c/em\u003eaddition to \u003cstrong\u003eTable 4\u003c/strong\u003e, \u003cstrong\u003eTable 5\u003c/strong\u003e presents the 6 best predictors for the MPMs generated from the mines and developed prospects and all Au training data (see \u003cstrong\u003eFig 18\u003c/strong\u003e for MPMs).\u003c/p\u003e\n\u003cp\u003eINSERT TABLE 5\u003c/p\u003e\n\u003cp\u003eFirstly, the top predictors for each MPM are quite different (suggesting a difference in deposit model \u0026ndash; iron formation vs. orogenic, structurally controlled models). The best predictors for the mines which are considered the most important of the RF models are the geophysical data (susceptibility especially) followed by iron formations and NLL produced geologic sub-units.\u003c/p\u003e\n\u003cp\u003eOverall, the best predictors, regardless of the MPM, are geological sub-units, susceptibility at the 4250m depth level and density at the depth 750m level.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eConsidering the MPMs based on the Au mines, developed prospects and occurrences (\u003cb\u003eFig.\u0026nbsp;18\u003c/b\u003e), the most predictive map, through evaluation of efficiency of classification curves (\u003cb\u003eFig.\u0026nbsp;19\u003c/b\u003e), \u003cem\u003eoob\u003c/em\u003e and classification accuracies (\u003cb\u003eTable\u0026nbsp;3\u003c/b\u003e), is the MPM generated from the Au mines followed by all the Au training data and developed prospects. However, the MPM generated from the developed Au prospects is not robust, nor stable, due to too few training points. When considering a cross-comparison between all MPMs (e.g., mines vs. developed prospects, occurrences vs. prospects etc.) the predictive results are very much weaker (Fig.\u0026nbsp;19). This is somewhat contrary to the results achieved by Harris et al. (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2006\u003c/span\u003e, 2008) in a study of the Red Lake greenstone belt in Ontario where the cross-comparisons were much more predictive. This may be due, in part, to differences in the Au deposit models.\u003c/p\u003e \u003cp\u003eThis study has shown that the selection of training points is critical as the results from the RF modelling can be significantly different depending on the points chosen (e.g., mines, developed prospects, occurrences). In this study we consider the MPM based on the Au mines as being the most robust and predictive. It can be noted that the MPMs are somewhat blocky in appearance. This is due to incomplete coverage of the entire study area by the 3D geophysical data. However, this does not affect the RF modelling other than highlighting the boundaries of the geophysical data.\u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure 22\u003c/b\u003e is a map that combines the most prospective areas derived from the MPMs shown in \u003cb\u003eFig.\u0026nbsp;1\u003c/b\u003e8. The most prospective areas were derived by taking greater and equal to 3 standard deviations above the mean probability from each MPM and creating a ternary combination whereby a green colour represents an area of 1 overlap (e.g., present on only one of the MPMs), and a purple colour represents areas where 2 MPMs overlap. There are no areas where all 3 MPMs overlap. Table\u0026nbsp;6 presents a geological summary of the each of the most prospective zones shown on \u003cb\u003eFig.\u0026nbsp;22\u003c/b\u003e. Areas C and D are of interest as they comprise iron formations. Area B is associated with the Paint Lake Fault (see \u003cb\u003eFigs.\u0026nbsp;1\u003c/b\u003e and \u003cb\u003e17\u003c/b\u003e for the location). Area A is of interest as it predicts 2 major mines; the MacLeod-Cockshutt and Geraldton mines.\u003c/p\u003e \u003cp\u003eINSERT TABLE 6\u003c/p\u003e \u003cp\u003eINSERT FIG. 22\u003c/p\u003e \u003cp\u003eThe majority of the most promising areas for Au exploration are found in the Beardmore-Geraldton greenstone belt, located in the southern part of the study area.\u003c/p\u003e \u003cp\u003eConsidering the method of generating non-deposit points our study indicates that a random selection of points equal to the number of training points but restricted to areas greater than 2 km from a training point (e.g., Au mine) produces the most predictive results (\u003cb\u003eFig.\u0026nbsp;19\u003c/b\u003e). This method is recommended over a selection of random points without spatial restrictions.\u003c/p\u003e \u003cp\u003eThe weighted average ensemble of all MPMs (mines, developed prospects, occurrences -\u003cb\u003eFig.\u0026nbsp;21\u003c/b\u003e) provides strong predictions of the Au mines which is slightly better than the MPM generated from only the mines. This is not unexpected as the mines were more heavily weighted by a factor of 3 (see \u003cem\u003eEq.\u0026nbsp;2\u003c/em\u003e). However, the average and weighted average ensemble maps (\u003cb\u003eFig.\u0026nbsp;21\u003c/b\u003e) do not predict very well the MPMs generated from the developed prospects and occurrences alone (Fig.\u0026nbsp;18).\u003c/p\u003e \u003cp\u003eThe most variable of the MPM parameters between different runs or especially when using different training areas, is the resulting most important predictor (predictor) maps. The strongest predictors for all variations of the MPM modelling process included the lithological data, obtained through NLL and the 3D geophysical data obtained at different depths. It is somewhat disappointing that the lithogeochemical data, processed using PCA, were not strong predictors due to relatively low number of sample points that were not distributed evenly across the study area.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eThis study has produced a number of Au MPMs of the Geraldton area in Ontario, Canada. Various experiments have been conducted that have relevance to the MPM RF modelling process. With respect to these experiments a number of conclusions can be summarized:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThe training data should be divided by importance (e.g., Au mines, developed prospects, occurrences). The MPMs generated by this procedure will be different as reflected by derived variations in measures of validation and visually with respect to prospective areas emphasized. These differences my shed light on slight differences in Au deposit models or tectonic environments in which the Au was deposited.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAs expected, the MPMs generated using the same training data you are trying to predict offer the best results. When cross-comparing MPMs based on different training data, the predictive results are lower as might be expected. One solution to this problem is to calculate an ensemble weighted average of the MPMs derived from different Au training datasets. In this study we recommend firstly weighting the more important Au deposits (e.g., by tonnage) and secondly providing more weight to the Au mines in generating the weighted ensemble MPM.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eNatural Language processing of textual geologic maps and legends provides valuable information for producing predictor maps for input to the MPM RF modelling process.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWe have identified a number of prospective zones in the study area for Au exploration follow-up. The majority of these areas occur in the Beardmore-Geraldton greenstone belt which appears more fertile than the Onaman-Tashota Belt.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe lithogeochemical data were not strong predictors of the Au data due to an incomplete and sporadic coverage over the study area. Although PC5 which is heavily weighted by As and Zn and PC2 heavily weighted by chalcophile elements (W-Zn-Sb-Bi-Sb-As-Pb-Te) do show a weak correlation with Au mines as 2 of the 12 (non-weighted) Au mines fall directly on the buffered PC5 and PC2 anomalies (see \u003cb\u003eFig.\u0026nbsp;15\u003c/b\u003e).\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003e\"J.H ., B.L. , C.L. , M.N., M.P., E.G., P.B., prepared the manuscript; J.H., E.G., B.P prepared the figures; J.H . , B.L., E.G., C.L., conceived the idea of the research\"\u003c/p\u003e\u003ch2\u003eACKNOWLEDGEMENTS\u003c/h2\u003e \u003cp\u003eThe authors would like to thank Metal Earth (Laurentian University, Sudbury, Ontario, Canada) for supporting this research. The Metal Earth publication number for this paper is \u003cb\u003eMERC-ME-2024-32.\u003c/b\u003e We would also like to thank the anonymous reviewers whose comments resulted in a much clearer rendition of our work.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eFunding Declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by the Metal Earth Project, Laurentian University, Sudbury, Ontario, Canada. Grant # CFREF- 2015-00005.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAitchison, J., 1986. The Statistical Analysis of Compositional Data. Chapman and Hall, London, 416 p.\u003c/li\u003e\n \u003cli\u003eAnglin, C.D., 1987. Geology, structure and geochemistry of gold mineralization in the Geraldton area, Northwestern Ontario. Unpublished Master\u0026rsquo;s Thesis. Memorial University of Newfoundland, St. John\u0026rsquo;s, Newfoundland. 283 p.\u003c/li\u003e\n \u003cli\u003eBeakhouse, G.P., 1991. Winnipeg River Subprovince, in: Thurston, P.C., Williams, H.R., Sutcliffe, R.H., Stott, G.M. (Eds.), Geology of Ontario. Ontario Geological Survey, Special Volume 4. Part 1. pp. 279\u0026ndash;301.\u003c/li\u003e\n \u003cli\u003eBlackburn, C.E., Johns, G.W., Ayer, J., Davis, D.W., 1991. Wabigoon Subprovince, in: Thurston, P.C., Williams, H.R., Sutcliffe, R.H., Stott, G.M. (Eds.), Geology of Ontario. Ontario Geological Survey, Special Volume 4. Part 1., pp. 303\u0026ndash;381.\u003c/li\u003e\n \u003cli\u003eBonham-Carter, G. F., Agterberg, F. P., Wright, D. F. 1989,. Weights of evidence modelling: a new approach to mapping mineral potential. In F. P.\u003c/li\u003e\n \u003cli\u003eBonham-Carter, G.F. (1994). \u003cem\u003eGeographic Information Systems for Geoscientists: Modeling with GIS\u003c/em\u003e. Pergamon, Elsevier Science Inc., New York.\u003c/li\u003e\n \u003cli\u003eBonham-Carter, G.F., Agterberg, F.P., Wright, D.F. (1988). Integration of geological datasets for gold exploration in Nova Scotia., Photogrammetric Engineering and Remote sensing. 54(11), pp. 1585-1592.\u003c/li\u003e\n \u003cli\u003eBreiman, L. 2001. Random forests. Machine Learning, 45(1), pp. 5-32.\u003c/li\u003e\n \u003cli\u003eBrodaric, B., Gahegan, M., \u0026amp; Harrap, R. (2004). The art and science of mapping: Computing geological categories from field data. \u003cem\u003eComputers \u0026amp; Geosciences,\u003c/em\u003e \u003cem\u003e30\u003c/em\u003e(7), 719\u0026ndash;740.\u003c/li\u003e\n \u003cli\u003eCarranza, E.J.M. (2009). Controls on mineral deposit occurrence inferred from analysis of their spatial pattern and spatial association with geological features. Ore Geology Reviews, 35, 383-400.\u003c/li\u003e\n \u003cli\u003eCarranza, E.J.M. and Laborte, A.G. 2015a. Data-driven predictive modeling of mineral prospectivity using random forests: a case study in Catanduanes Island (Philippines), Nat Resource Res,25: pp. 35-50.\u003c/li\u003e\n \u003cli\u003eCarranza, E.J.M., Laborte, A.G. (2015c). Random forest predictive modeling of mineral prospectivity with small number of prospects and data with missing values in Abra (Philippines). Computers \u0026amp; Geosciences 74, 60-70.\u003c/li\u003e\n \u003cli\u003eCarranza, E.J.M.; Laborte\u003cem\u003e, A.G., 2015b\u0026nbsp;\u003c/em\u003eData-driven predictive mapping of gold prospectivity, Baguio district, Philippines: Application of Random Forests algorithm\u003cem\u003e. Ore Geol. Rev., 71, 777\u0026ndash;787.\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003eChung, C.J., and Fabri, A., 2003.Validation of Spatial Prediction for Landslide Hazard Mapping, Natural Hazards, 30(3); pp 451-472, DOI: 10.1023/B:NHAZ.0000007172.62651.2b\u003c/li\u003e\n \u003cli\u003eCulshaw, N., Purves, M., Reynolds, P., and Stott, G. 2006. Post-collisional upper crustal faulting and deep crustal flow in the eastern Wabigoon subprovince of the Superior Province, Ontario: Evidence from structural and \u003csup\u003e40\u003c/sup\u003eAr/\u003csup\u003e39\u003c/sup\u003eAr data from the Humboldt Bay High Strain Zone; Precambrian Research, v.145, pp.272-288.\u003c/li\u003e\n \u003cli\u003eDevaney, J.R., Williams, H.R., 1989. Evolution of an Archean subprovince boundary: a sedimentological and structural study of part of the Wabigoon\u0026ndash;Quetico boundary in northern Ontario. Can. J. Earth Sci. 26, 1013\u0026ndash;1026.\u003c/li\u003e\n \u003cli\u003eDeWolfe, J.C., Lafrance, B., Stott, G.M., 2007. Geology of the shear-hosted Brookbank gold prospect in the Beardmore\u0026ndash;Geraldton belt, Wabigoon Subprovince, Ontario. Can. J. Earth Sci. 44, 925\u0026ndash;946. doi:10.1139/E06-118\u003c/li\u003e\n \u003cli\u003eDoan, H.T., Foody, G.M., 2007. Increasing soft classification accuracy through the use of an ensemble classifier. Int. J. Remote Sens. 28, pp. 4609\u0026ndash;4623.\u003c/li\u003e\n \u003cli\u003eFerguson, S.A. 1967. Leitch Gold Mines Limited, Surface plan of eastern part of property, Parts of Eva and Summers Townships, District of Thunder Bay. Ontario Department of Mines, Preliminary Geological Map P.484.\u003c/li\u003e\n \u003cli\u003eFord, A. (2020). Practical implementation of random forest-based mineral potential mapping for porphyry Cu-Au mineralization in the Eastern Lachlan Orogen, NSW, Australia. \u003cem\u003eNatural Resources Research\u003c/em\u003e, 29, 267-283. doi: 10.1007/s11053-019-09598-y\u003c/li\u003e\n \u003cli\u003eGrunsky, E.C., 2010. The interpretation of geochemical survey data; Geochemistry, Exploration, Environment Analysis 10(1), p. 27-74, doi: 10.1144/1467-7873/09-210.\u003c/li\u003e\n \u003cli\u003eGSC (Geological Survey of Canada), 2020. Geophysical Data. http://gdr.agg.nrcan.gc.ca/gdrdap/dap/search-eng.php. Accessed Feb 3, 2023.\u003c/li\u003e\n \u003cli\u003eHarman, D. (1991). How effective is suffixing? \u003cem\u003eJournal of the American Society for Information Science,\u003c/em\u003e \u003cem\u003e42\u003c/em\u003e(1), 7\u0026ndash;15.\u003c/li\u003e\n \u003cli\u003eHarris, J., Grunsky, E., Behnia, P., Corrigan, D. (2015a). Data-and knowledge-driven mineral prospectivity maps for Canada\u0026apos;s North. Ore Geology Reviews 71, 788-803.\u003c/li\u003e\n \u003cli\u003eHarris, J., and Grunsky, E.C. (2015b). Predictive lithological mapping of Canada\u0026apos;s North using Random Forest classification applied to geophysical and geochemical data. Computers \u0026amp; Geosciences 80, 9-25.\u003c/li\u003e\n \u003cli\u003eHarris, J.R., M. Naghizadeh, P. Behnia, L., Mathieu, 2022. Data-driven gold potential maps for the Chibougamau area, Abitibi greenstone belt, Canada, Ore Geology Reviews, 150, 105176.\u003c/li\u003e\n \u003cli\u003eHarris, J.R., Sanborn-Barrie, M., 2008. Mineral potential mapping: examples from the Red Lake greenstone belt, northwest Ontario. \u003cem\u003ein\u003c/em\u003e: Harris, J.R. (Ed.), GIS for the Earth Sciences. Geological Association of Canada Special Paper 44, pp. 1\u0026ndash;2.\u003c/li\u003e\n \u003cli\u003eHarris, J.R., Sanborn-Barrie, M., Panagapko, D.A., Skulski, T., Parker, J.R., 2006. Gold prospectivity maps of the Red Lake greenstone belt: application of GIS technology. Can. J. Earth Sci. 43, 865\u0026ndash;89\u003c/li\u003e\n \u003cli\u003eHarris, J.R.; Wilkinson, L.; Heather, K.; Fumerton, S.; Bernier, M.A.; Ayer, J.; Dahn, R. 2001. Application of GIS Processing Techniques for Producing Mineral Prospectivity Maps\u0026mdash;A Case Study: Mesothermal Au in the Swayze Greenstone Belt, Ontario, Canada. \u003cem\u003eNat. Resour. Res\u003c/em\u003e \u003cem\u003e10\u003c/em\u003e, 91\u0026ndash;124. https://doi.org/10.1023/a:1011548709573.\u003c/li\u003e\n \u003cli\u003eHart, T.R., terMeer, M. and Jolette, C. 2002. Precambrian geology of Kitto, Eva, Summers, Dorothea andSandra townships, northwestern Ontario: PhoenixBedrock Mapping Project; Ontario Geological Survey, Open File Report 6095, 206 p.\u003c/li\u003e\n \u003cli\u003eHorwood, H.C., Pye, E.G., 1955. Geology of Ashmore Township. In: 60\u003csup\u003eth\u003c/sup\u003e Annual Report of the Ontario Department of Mines, Part 5. 139 p.\u003c/li\u003e\n \u003cli\u003eLafrance, B., DeWolfe, J.C., Stott, G.M., 2004. A structural reappraisal of the Beardmore-Geraldton Belt at the southern boundary of the Wabigoon subprovince, Ontario, and implications for gold mineralization. Can. J. Earth Sci. 41, 217\u0026ndash;235. doi:10.1139/E03-090\u003c/li\u003e\n \u003cli\u003eLavigne, M.J. 2009. Distribution of gold with respect to lithologies, metamorphic facies and strain state in the Beardmore\u0026ndash;Geraldton greenstone belt; Ontario Geological Survey, Open File Report 6241, 88p.\u003c/li\u003e\n \u003cli\u003eLawley et. al., 2022, https://www.sciencedirect.com/science/article/pii/S2590197422000064\u003c/li\u003e\n \u003cli\u003eLemkow, D.R., Sanborn-Barrie, M., Stott, G.M., Percival, J.A, Stone, D., Tomlinson, K.Y., Skulksi, T., McNicoll, V., Davis, D.W., Whalen, J.B. and Hollings, P. 2005. GIS compilation of the geology and tectonostratigraphic assemblages, Wabigoon\u0026ndash;Winnipeg River\u0026ndash;Marmion transect, western Superior Province, Ontario; Geological Survey of Canada, Open File 4971, Ontario Geological Survey, Miscellaneous Release\u0026mdash;Data 187, Scale 1:250 000\u003c/li\u003e\n \u003cli\u003eLincoln, L. A., Benoit, K., Keyes, O., Selivanov, D., \u0026amp; Arnold, J. (2018). Fast, consistent tokenization of natural language text. \u003cem\u003eJournal of Open Source Software,\u003c/em\u003e \u003cem\u003e3\u003c/em\u003e(23), 655.\u003c/li\u003e\n \u003cli\u003eMacdonald, A.J., 1988. The Geraldton Gold Camp: The Role of Banded Iron Formation; Ontario Geological Survey, Open File Report 5694, 173 p.\u003c/li\u003e\n \u003cli\u003eMackasey, W.O., 1975. Geology of Dorothea, Sandra, and Irwin Townships, District of Thunder Bay. Ontario Division of Mines, Geological Report 122, 83 p.\u003c/li\u003e\n \u003cli\u003eMackasey, W.O., 1976. Geology of Walters and Leduc Townships, District of Thunder Bay. Ontario Division of Mines, Geoscience Report 149. 58 p.\u003c/li\u003e\n \u003cli\u003eMackasey, W.O., Edwards, G.R., and Cape, D.F. 1976. Legault Township, District of Thunder Bay. Ontario Division of Mines, Preliminary Map P-1191.\u003c/li\u003e\n \u003cli\u003eMantovani, A., Piana, F., \u0026amp; Lombardo, V. (2020). Ontology-driven representation of knowledge for geological maps. \u003cem\u003eComputers \u0026amp; Geosciences,\u003c/em\u003e \u003cem\u003e139\u003c/em\u003e, 104446.\u003c/li\u003e\n \u003cli\u003eMark, B., 2024. Geology and Structural Evolution of the Tashota Deformation Zone and Implications for Orogenic Gold Mineralization. Unpublished MSc thesis, Laurentian University, Sudbury, Canada, 191 pp.\u003c/li\u003e\n \u003cli\u003eMartin-Fernandez, J.A., Barcel\u0026oacute;-Vida l, C. \u0026amp; Pawlowsky-Glahn, V. 2003. Dealing with zeros and missing values in compositional data sets using non-parametric imputation. Mathematical Geology, 35, 253\u0026ndash;278.tern Fujian Province, China. \u003cem\u003eOre Geology Reviews\u003c/em\u003e, 71, pp. 502-515\u003c/li\u003e\n \u003cli\u003eMason, J.K., McConnell, C.D., 1983. Gold mineralization in the Beardmore-Geraldton area, In: Colvine, A.C. (Ed.), The Geology of Gold in Ontario, Ontario Geological Survey, Misc. Paper 110. pp. 84\u0026ndash;97.\u003c/li\u003e\n \u003cli\u003eMason, J.K., White, G., 1986. Gold occurrences, prospects and deposits of the Beardmore-Geraldton area, District of Thunder Bay and Cochrane. Ontario Geological Survey, Open File Report 5630. 680 p.\u003c/li\u003e\n \u003cli\u003ePalarea-Albaladejo J, Mart\u0026iacute;n-Fern\u0026aacute;ndez JA, Buccianti A (2014) Compositional methods for estimating elemental concentrations below the limit of detection in practice using R. J Geochem Explor. ISSN 0375\u0026ndash;6742. https://doi.org/10.1016/j.gexplo.2013.09.003.\u003c/li\u003e\n \u003cli\u003ePavlis, T. L., Langford, R., Hurtado, J., \u0026amp; Serpa, L. (2010). Computer-based data acquisition and visualization systems in field geology: Results from 12 years of experimentation and future potential. \u003cem\u003eGeosphere,\u003c/em\u003e \u003cem\u003e6\u003c/em\u003e(3), 275\u0026ndash;294.\u003c/li\u003e\n \u003cli\u003ePercival, J.A., Skulski, T., Sanborn-Barrie, M., Stott, G.M., Leclair, A.D., Corkery, M.T., Boily, M., 2012. Geology and tectonic evolution of the Superior Province, Canada, in: Percival, A.J., Cook, F.A., Clowes, R.M. (Eds.), Tectonic Styles in Canada. Geological Association of Canada, pp. 321\u0026ndash;378.\u003c/li\u003e\n \u003cli\u003ePye, E.G., 1952. Geology of Errington Township, Little Long Lac Area, in: 60th Annual Report of the Ontario Department of Mines. Ontario Department of Mines, 140 p.\u003c/li\u003e\n \u003cli\u003eRodriguez-Galiano, V., Sanchez-Castillo, M., Chica-Olmo, M., Chica-Rivas, M. (2015). Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines. Ore Geology Reviews 71, 804-818.\u003c/li\u003e\n \u003cli\u003eRodriguez-Galiano, V.F., Chica-Olmo, M., Chica-Rivas, M. (2014). Predictive modelling of gold potential with the integration of multisource information based on random forest: a case study on the Rodalquilar area, Southern Spain. International Journal of Geographical Information Science 28, 1336-1354.\u003c/li\u003e\n \u003cli\u003eShanks, W.S. 1993. Geology of Eva and Summers townships, District of Thunder Bay; Ontario Geological Survey, Open File Report 5821, 93p.\u003c/li\u003e\n \u003cli\u003eSilge, J., \u0026amp; Robinson, D. (2016). tidytext: Text mining and analysis using tidy data principles in R. \u003cem\u003eJournal of Open Source Software,\u003c/em\u003e \u003cem\u003e1\u003c/em\u003e(3), 37.\u003c/li\u003e\n \u003cli\u003eStott, G.M., Davis, D. W., Parker, J.R., Straub, K.J., and Tomlinson K. Y., 2002a. Geology and tectonostratigraphic assemblages,eastern Wabigoon Subprovince, Ontario, Ontario Geological Survey Map P3449, scale 1:250,000, Geological Survey of Canada, Open file 4285\u003c/li\u003e\n \u003cli\u003eStott, G.M., Davis, D.W., Parker, J.R., Straub, K.H., Tomlinson, K.Y., 2002b. Geology and Tectonostratigraphic Assemblages, eastern Wabigoon Subprovince, Ontario. Scale 1:250 000. Ontario Geological Survey, Preliminary Map P. 3449. 1 sheet\u003c/li\u003e\n \u003cli\u003eThurston, P.C. 1980. Geology of the Northern Onaman Lake Area, District of Thunder Bay; Ontario Geological Survey, Report 208, 81p. Accompanied by Map 2411, scale 1:31 680, and Chart A.\u003c/li\u003e\n \u003cli\u003eTomlinson, K.Y., Hall, R.P., Hughes, D.J., Thurston, P.C., 1996. Geochemistry and assemblage accretion of metavolcanic rocks in the Beardmore\u0026ndash;Geraldton greenstone belt, Superior Province. Can. J. Earth Sci. 33, 1520\u0026ndash;1533.\u003c/li\u003e\n \u003cli\u003eT\u0026oacute;th, Z., 2019. The geology of the Beardmore-Geraldton belt, Ontario, Canada: geochronology, tectonic evolution and gold mineralization. Unpublished Ph.D. thesis, Laurentian University, Sudbury, Ontario.\u003c/li\u003e\n \u003cli\u003eT\u0026oacute;th, Z., Lafrance, B., Dub\u0026eacute;, B., 2023. Oblique lateral extrusion during dextral transpression along the Beardmore-Geraldton belt, Canada. Journal of Structural Geology, 169,104834.\u003c/li\u003e\n \u003cli\u003eT\u0026oacute;th, Z., McNicoll, v., Lafrance, B., Dub\u0026eacute;, B., 2022. Early depositional and magmatic history of the Beardmore-Geraldton belt: Formation of a transitional accretionary belt along the Wabigoon-Quetico subprovince boundary in the Archean Superior craton, Canada. Precambrian Res. 371. https://doi.org/10.1016/j.precamres.2022.106579\u003c/li\u003e\n \u003cli\u003eWang, J., Zuo, R. and Xiong, Y. (2020). Mapping Mineral Prospectivity via Semi-supervised Random Forest. Natural Resources Research, 29, 189\u0026ndash;202. https://doi.org/10.1007/s11053-019-09510-8\u003c/li\u003e\n \u003cli\u003eZhang, S., Carranza, E.J.M., Xiao, K. et al. (2021) Mineral prospectivity mapping based on isolation forest and random forest: Implication for the existence of spatial signature of mineralization in outliers. \u003cem\u003eNatural Resources Research\u003c/em\u003e, 31, 1981\u0026ndash;1999. https://doi.org/10.1007/s11053-021-09872-y\u003c/li\u003e\n \u003cli\u003eZhang, S., Xiao, K., Carranza, E. J. M., \u0026amp; Yang, F. (2018). Maximum entropy and random forest modeling of mineral potential: analysis of gold prospectivity in the Hezuo-MeiwuDistrict, West Qinling Orogen China. \u003cem\u003eNatural Resources Research\u003c/em\u003e, 28(3), 645\u0026ndash;664\u003c/li\u003e\n \u003cli\u003eZuo, R. (2020). Geodata Science-Based Mineral Prospectivity Mapping: A Review. Natural Resources Research 29, 3415-3424.\u003c/li\u003e\n \u003cli\u003eZuo, R., Carranza, E.J.M., 2011. Support vector machine: a tool for mapping mineral prospectivity. Comput. Geosci. 37, 1967\u0026ndash;1975\u003c/li\u003e\n \u003cli\u003eZuo, R., Zhang, Z., Zhang, D., Carranza, E.J.M. and Wang, H. (2015). Evaluation of uncertainty in mineral prospectivity mapping due to missing evidence: a case study with skarn-type Fe deposits in Southwestern Fujian Province, China, Ore Geology Reviews, 71, pp. 502-515\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 1\u003c/strong\u003e \u0026ndash; Summary of predictor maps (vectors to mineralization) used for modelling\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePredictor map\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eComments\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eThreshold\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 593px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eLithogeochemistry\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003ePCA2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eBuffered to 1 km\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e\u0026gt;3 stdev\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003ePCA3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eBuffered to 1 km\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e\u0026gt;3 stdev\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003ePCA4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eBuffered to 1 km\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e\u0026gt;3 stdev\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003ePCA5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eBuffered to 1 km\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003e\u0026gt;3 stdev\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 593px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e\u003cem\u003eLithology- possible sources of orogenic Au\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eIron Formations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu known to be associated with Iron formations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eBuffered to 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eLithologic units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu source?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003emulti class\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eLithological sub-units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu source?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eSedimentary units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu source?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eMetamorphic units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu source?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eIgneous units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAu source?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 593px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;\u003cem\u003eStructure \u0026ndash; possible zones of weakness for the migration of Au bearing fluids\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eInterpreted faults (mag)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e60-100\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eInterpreted faults (mag)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e40 \u0026ndash; 60\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eInterpreted faults (mag)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e100 \u0026ndash; 260\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eInterpreted faults (mag)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e0 \u0026ndash; 40\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eMapped faults\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e\u0026nbsp;60 - 100\u0026deg;strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eMapped faults\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e40 \u0026ndash; 60\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eMapped faults\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e260 \u0026ndash; 360\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eMapped faults\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e100 \u0026ndash; 260\u0026deg; strike direction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eDykes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAll directions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eLithological contacts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAll directions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003ebuffered at 1 km - binary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 593px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u003cstrong\u003eGeophysics \u0026ndash; to define lithology at different depths\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eDensity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e750m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eDensity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e4250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eDensity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e7250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eSusceptibility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e750m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eSusceptibility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e2250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eSusceptibility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e4250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eSusceptibility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e7250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eResistivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e750m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eResistivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e2250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eResistivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e4250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 268px;\"\u003e\n \u003cp\u003eResistivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e7250m depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 157px;\"\u003e\n \u003cp\u003eContinuous surface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e \u0026ndash; Correlation between MPMs generated from Au mines, developed prospects, occurrences\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 120px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMines\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 130px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDeveloped prospects\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOccurrences\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAll\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMines\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 120px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 130px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDeveloped prospects\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 120px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 130px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOccurrences\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 120px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 130px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAll\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 120px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 130px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 134px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 3\u003c/strong\u003e \u0026ndash; Out-of-bag and overall classification accuracies from MPMs generated from mines, developed prospects, occurrences\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMines\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDeveloped prospects (DVP)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOccurrences\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOut-of-Bag (oob) accuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e80%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e59%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e80%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall classification accuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e97.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e90.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 4\u003c/strong\u003e \u0026ndash; Results from 5-fold repetition of RF modelling (a) variability in best predictors, (b) variability in out-of-bag error, (c) variability in overall classification accuracy\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRank\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOriginal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRepeat 2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRepeat 3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRepeat 4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;Repeat 5\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e5\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;a) \u0026nbsp; 3 = geol_sub_unit; 25 = den_7250; 2 = geol_unit; 26 = den_4250; 13 = res_4250; 24 = \u0026nbsp; den_750; 8 = sus_7250; 5 = new_met (see Table 1 for evidence maps)\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e82.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e81.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e81.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e82.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e82.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e82.1 ave\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eb) \u0026nbsp; \u0026nbsp; \u0026nbsp;Out-of-bag error for 5-fold repetition of RF modelling\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e94.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e95.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e94.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e94.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e94.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e94.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 82px;\"\u003e\n \u003cp\u003e94.6 ave\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003ec) \u0026nbsp; \u0026nbsp;Overall classification accuracies for the 5-fold repetition of RF modelling\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 5\u003c/strong\u003e \u0026ndash; top 6 predictors (evidence maps) for MPMs generated from Au mines, developed prospects, and all Au training data\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRank\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAu Mines\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAu Developed Prospects\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAu Occurrences\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e5\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e1 = iron formations; 2 = Geological units; 3 = Geological sub-units, 4 = geological sedimentary units; 6 = Geological igneous units; 7 = Susceptibility 750m depth; 8 = Susceptibility 7250m depth; 9 = Susceptibility 4250m depth; 10= Susceptibility 2250m depth; 11= Resistivity 7250m depth; 13 = Resistivity 4250m depth; 24 = dikes; 25 = density 750m depth; 26 = density 7250m depth\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 6\u003c/strong\u003e \u0026ndash; Summary of Au high prospectivity areas shown on Fig. 22\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eLocation\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eLithological / Structural Comments\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003ePredictions\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMostly metasediments and smaller E-W trending linear volcanic belt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts many 7 producing mines (including MacLeod-Cockshutt and Geraldton mines), 2 developed prospects and 14 Au occurrences\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eB\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMostly felsic and intermediate volcanic rocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts many Au occurrences along the Paint Lake Fault\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eE-W striking metasedimentary rocks, strong association with banded iron formations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts many Au occurrences and 3 mines (including Leitch mine)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMainly E-W striking mafic and intermediate volcanic rocks, strong association with banded iron formations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredict many Au occurrences and 1 mine (Northern Empire mine)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eArea cored by folded diorite-monzodiorite-granodiorite suite Surrounded by mafic\\intermediate volcanic and metasedimentary rocks, associated with shear zones\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredict 6 Au occurrences\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eSame geology as E\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003eNo known Au\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMafic/intermediate volcanics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts 3 Au occurrences\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eG\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eE-W trending mafic/intermediate volcanic rocks with small belts of felsic volcanics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts 3 Au occurrences and 1 developed prospect\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eH\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMostly mafic\\intermediate volcanic rocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003eNo known Au\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMostly mafic\\intermediate volcanic rocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredicts 3 developed prospects and 1 Au occurrence\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eJ\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 345px;\"\u003e\n \u003cp\u003eMostly mafic\\intermediate volcanic rocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 205px;\"\u003e\n \u003cp\u003ePredict 7 Au occurrences and 1 mine (Consolidated Louanna mine)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"earth-science-informatics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"esin","sideBox":"Learn more about [Earth Science Informatics](http://link.springer.com/journal/12145)","snPcode":"12145","submissionUrl":"https://submission.nature.com/new-submission/12145/3","title":"Earth Science Informatics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Mineral Prospectivity mapping, gold exploration, greenstone belts, tectonics, machine learning, random forests","lastPublishedDoi":"10.21203/rs.3.rs-5384377/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5384377/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn this paper, we employ Random Forests (RF) (Breiman, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) to generate several Mineral Prospectivity Maps (MPMs) for orogenic gold in the Geraldton area, located within the Abitibi Tectonic Subprovince of Ontario, Canada. We address various issues pertinent to the Mineral Prospectivity Mapping (MPM) modeling process and propose solutions to these key challenges. Additionally, we utilize multiple methods to analyze text-based geoscientific information derived from geological maps, including a novel application of natural language processing to delineate the sources and traps of gold mineral systems.\u003c/p\u003e","manuscriptTitle":"Orogenic Gold Mineral Prospectivity Mapping (MPM) of the Geraldton Area, Ontario","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-18 06:43:31","doi":"10.21203/rs.3.rs-5384377/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-12-02T14:52:35+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-12-01T15:25:27+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-28T09:34:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"128339002502957879111082257224242411177","date":"2024-11-28T00:08:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-27T03:05:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"31369865012197201171071226602474948307","date":"2024-11-26T02:10:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"101464585821140299392724765754298232554","date":"2024-11-26T01:01:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"333590312829828861071189698675625826700","date":"2024-11-26T00:43:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"257982114353209943018921987862790672144","date":"2024-11-26T00:16:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"228313436528108523052849012425026665397","date":"2024-11-25T18:22:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"213855249573881654947790357578068478591","date":"2024-11-25T16:43:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"301595110184928217041136408877741886261","date":"2024-11-25T16:41:28+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-25T16:37:31+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-25T16:35:58+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-18T06:46:43+00:00","index":"","fulltext":""},{"type":"submitted","content":"Earth Science Informatics","date":"2024-11-04T03:02:21+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"earth-science-informatics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"esin","sideBox":"Learn more about [Earth Science Informatics](http://link.springer.com/journal/12145)","snPcode":"12145","submissionUrl":"https://submission.nature.com/new-submission/12145/3","title":"Earth Science Informatics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"76e092dc-48c5-475b-8bf8-f52508dd02c1","owner":[],"postedDate":"December 18th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-04-07T16:06:17+00:00","versionOfRecord":{"articleIdentity":"rs-5384377","link":"https://doi.org/10.1007/s12145-025-01831-y","journal":{"identity":"earth-science-informatics","isVorOnly":false,"title":"Earth Science Informatics"},"publishedOn":"2025-04-04 15:57:39","publishedOnDateReadable":"April 4th, 2025"},"versionCreatedAt":"2024-12-18 06:43:31","video":"","vorDoi":"10.1007/s12145-025-01831-y","vorDoiUrl":"https://doi.org/10.1007/s12145-025-01831-y","workflowStages":[]},"version":"v1","identity":"rs-5384377","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5384377","identity":"rs-5384377","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}