Applying Mineral System Criteria to Develop a Predictive Modelling for Epithermal Gold Mineralization in Northern New Brunswick: Using Knowledge-Driven and Data-Driven Methods

Abstract

Using mineral prospectivity mapping (MPM), the mineral systems approach enables the identification of geological indicators linked to ore formation. This approach streamlines exploration by minimizing the time and cost required to identify areas with the highest mineral potential. With its extensive till cover and dense forests limiting bedrock exposure, New Brunswick provides an ideal environment to test this approach. The New Brunswick portion of the Canadian Appalachians hosts a diverse range of gold deposits and occurrences that formed during various stages of the Appalachian orogeny. In northern New Brunswick and the adjacent Gaspé Peninsula, the Tobique–Chaleur Zone contains several orogenic and epithermal gold systems that are closely associated with a large-scale crustal fault and its offshoots, i.e., the long-lived trans-crustal Rocky Brook–Millstream Fault system. To identify favorable zones for epithermal gold mineralization in northwestern New Brunswick, this study employed MPM by translating key mineral system components—such as ore metal sources, fluid pathways, traps, and geological controls—into mappable criteria for regional-scale analysis. The data were modeled through the integration of knowledge-based and data-driven methods, including fuzzy logic, geometric average, and logistic regression approaches. The concentration–area (C–A) fractal model was applied to reclassify the final maps based on prospectivity values obtained from these three approaches, dividing the mineral prospectivity maps into six classes, with threshold values emphasizing high-favorability zones. The fuzzy overlay model had the highest predictive accuracy (AUC 0.97), followed by the geometric average model (AUC 0.93), whereas the logistic regression identified more tightly constrained high-potential zones. In the prospectivity models, known epithermal gold mineralization consistently overlaps with regions of high favorability. This suggests a positive result from the use of MPM, indicating that this approach could be applicable to other regions and types of ore deposits.

1. Introduction

Mineral prospectivity mapping (MPM) is a multi-criteria decision-making process that aims to assist mineral explorationists in identifying mineral targets using existing geological datasets by integrating multiple geospatial variables [1,2,3,4]. It achieves this by recognizing areas with geological characteristics similar to known mineral deposits or occurrences. MPM relies on diverse geospatial data types, including geochemical, geophysical, and geological datasets [5]. However, ensuring the validity and interpretability of predictive models remains a challenge, particularly due to uncertainty in the correlation between prospectivity maps and deposit models, which can reduce prediction reliability [6]. To address this, researchers have translated mineral system models for specific mineral deposit types to link geological features to prospectivity models [3].

The mineral systems approach helps geologists identify critical components influencing ore formation, such as metal sources, fluid pathways, and depositional traps [3,7]. These geological factors can be quantified and integrated into mineral prospectivity models. MPM follows one of two methodologies. Data-driven techniques use training points to identify patterns directly from geospatial data, making them suitable when there is a large dataset of known mineral occurrences (brownfield exploration). In contrast, knowledge-driven approaches depend on geological expertise, which is particularly valuable in areas with limited data (greenfield exploration). Knowledge-driven methods are efficient in translating mineral system components to mappable criteria and the production of potential maps [8]. These techniques have demonstrated their effectiveness in leveraging expert knowledge to test mineral system models and support MPM [3,9]. Recently, in the Kolari region, Finland, Khammar et al. [3] modified the integration of proxies using the mineral system approach and integration of evidence layers to generate four components: source, energy, pathway, and trap. These components were analyzed using fuzzy logic overlay, geometric average, fuzzy inference system, and logistic regression to create prospectivity maps, which successfully delineated areas of high potential for iron oxide–copper–gold deposits. Likewise, Skirrow et al. [10] described a method of MPM based on the mineral systems concept and mapped iron oxide–copper–gold mineral potential in Australia using the knowledge-driven methodology [10]. They concluded that the approach is most appropriately applied in cartoon- to regional-scale analysis. Nykänen et al. [9] effectively applied a knowledge-based fuzzy logic approach to delineate potential areas for Co across Finland. This method integrated parameters derived from mineral systems models to identify Co as either a primary commodity or a by-product. Despite these advancements, limited research has explored the integration of these techniques for epithermal gold mineralization, particularly in the Canadian Appalachians.

In Canada, MPM studies have addressed various mineral systems, utilizing advanced mapping techniques to uncover mineralization patterns specific to structural geology. There has also been extensive MPM research focusing on mineral-rich areas and leveraging geostatistical and machine-learning models to identify economically viable deposits [11,12,13,14,15]. However, the translation of mineral system components into a knowledge- and data-driven MPM framework for epithermal gold deposits remains underdeveloped. The aim of this study was to investigate the geological criteria controlling the formation of gold mineralization and translate them to mappable criteria and evidence layers. This was accomplished by employing various geological, geochemical, and geophysical datasets. Following this, three integration techniques were employed to generate prospectivity maps. This study applies fuzzy logic overlay and geometric average, in the light of recent advancements in knowledge-driven approaches [3,9]. Moreover, the study incorporates an empirical approach using logistic regression (LR), enabling a direct comparison between knowledge-driven and data-driven methods trained on known occurrences, thereby improving the accuracy in the identification of prospective areas. It should be mentioned that each mineral prospectivity mapping (MPM) approach—fuzzy logic, geometric average, and logistic regression—offers distinct advantages and limitations in mineral exploration. Fuzzy logic is highly effective in handling uncertainty and integrating expert geological knowledge, making it ideal for regions with sparse data. In this approach, evidential layers were assigned weights and confidence factors based on expert assessments of their relevance to epithermal gold mineralization processes. However, its reliance on subjective input can introduce bias, and it lacks a statistical framework for quantifying the significance of each predictor variable [16]. Geometric average provides a balanced integration of multiple geological layers of gold mineralization without overly emphasizing extreme values, reducing the impact of outliers. Yet, its main limitation is that it assumes equal weighting for all input layers, which may not accurately reflect geological importance [3]. Logistic regression offers a statistically robust framework for prospectivity modeling, identifying significant predictors and providing probabilistic outputs. This method excels in regions with abundant known mineral occurrences and enhances model interpretability. However, it requires a sufficiently large training dataset and assumes a linear relationship between predictor variables and mineralization, which may not be reliable in the complex geological setting of this study area [17]. By integrating these methods, this study leverages the strengths of knowledge-driven and data-driven approaches to improve prediction accuracy and reduce uncertainty in epithermal gold prospectivity mapping.

A case study was conducted in New Brunswick, a region recognized for its gold potential. Twenty-one evidence layers were created based on the conceptual model of the targeted deposit type, which were then utilized to produce prospectivity maps. This study applied the concentration-area fractal method to define threshold limits for various types of mineral prospectivity maps [18,19]. Finally, the effectiveness of the maps was assessed using the ROC evaluation method, along with mineralogical and petrological examinations [3].

2. Geology of the Study Area

The northern Appalachian belt is a complex collage of suprasubduction zone terranes and microcontinents formed during the closure of the Iapetus and Rheic oceans between the late Cambrian and the Permian [20,21]. The suprasubduction zone terranes contain various types of arc rocks that formed during the subduction of oceanic crust, whereas the microcontinents show crustal blocks of peri-Laurentian (Dashwoods) and peri-Gondwanan (Ganderia, Avalonia, and Meguma) provenance [20,21]. The major tectonic zones of the Canadian Appalachians, from northwest to southeast are Humber, Dunnage, Gander, Avalon, and Meguma. The sequence of oceanic and continental accretion and closure of the Iapetus and Rheic oceans led to several collisional events containing Taconic, Penobscot, Salinic, Acadian, and NeoAcadian orogenies, between the late Cambrian and Permian [21]. Each of these orogenies was accompanied by a wide variety of significant types of mineralization. Northern Appalachians and Maritime Canada display major tectonic units and structures related to the closure of the Iapetus and Rheic oceans, such as the Acadian, NeoAcadian, Variscan, Hercynian, and Alleghanian orogenies [20]. The northern part of the Appalachian belt displays systematically younger ages of deformation from the west to the east, as the remnants of various arcs and the microcontinents were accreted [22,23]. The Appalachians include rocks affected by various orogenies, representing the combination of folding, faulting, metamorphism, and plutonism [24] (Figure 1a). Siluro-Devonian bimodal volcanic and interbedded sedimentary rocks of the Northern Appalachians in eastern Canada and Maine (USA) are exposed in three major belts: the Piscataquis, Tobique–Chaleur zone, and Coastal volcanic belts. The study area is located within the Tobique–Chaleur zone, which stretches from southwestern New Brunswick to the Gaspé Peninsula of Québec [24,25] (Figure 1b,c).

The Canadian Appalachians host gold mineralization, which formed during various stages of the Appalachian orogeny and subsequent episodes of exhumation and erosion [21,26]. These include a wide range of gold-bearing mineral deposit types, including orogenic [27], volcanogenic massive sulfide [28,29], skarn [30,31], and epithermal systems [32].

According to Fyffe and Fricker [33], the Middle Paleozoic Matapédia Cover Sequence (MCS) is a Late Ordovician to Middle Devonian basin fill sequence. The Matapedia Cover Sequence is divided into three structures that from northwest to southeast, are (1) the Connecticut Valley–Gaspé Synclinorium, (2) the Aroostook–Percé Anticlinorium, and (3) the Chaleur Bay Synclinorium [34,35]. The southern limit of the Fortin Formation, the Matapedia fault, marks the boundary between the Connecticut Valley–Gaspé Synclinorium and the Aroostook–Percé Anticlinorium to the south [34].

In Northern New Brunswick (Figure 1), Late Silurian volcanic rocks are divided into the Petit-Rocher, Dickie Cove, and Chaleurs groups, whereas Early Devonian volcanic rocks are included in the Tobique and Dalhousie groups [36]. In the study area (Figure 1b,c) three Early Paleozoic inliers—Popelogan, Miramichi, and Elmtree—are overlain by Middle Paleozoic rocks of the Tobique Chaleurs zone and Carboniferous rocks. According to [20,37], the Acadian deformation in northern New Brunswick took place during the Early Devonian, between 415 and 395 Ma. In its later stages, this deformation was characterized by dextral, primarily strike-slip faulting, along the Rocky Brook–Millstream, Restigouche–Grand Pabos, and Catamaran Brook fault systems, as well as several associated satellite faults [22,26]. The gold occurrences investigated in this study are located in the Chaleur Bay Synclinorium and are hosted primarily by Upper Silurian to Early Devonian strata and subordinate lower Silurian strata (Figure 1b,c). The area investigated encompasses two 1:50,000 scale map sheets, covering approximately 1560 km² and containing 20 gold occurrences (Figure 1c). The Chaleur Bay Synclinorium exhibits more varied geology and a more intricate tectonic history than the adjacent Aroostook–Percé Anticlinorium. Its uppermost Silurian to Early Devonian sedimentary and volcanic rocks are divided into the Chaleurs, Petit-Rocher, Quinn Point, Dickie Cove, and Dalhousie groups north of the Rocky Brook–Millstream Fault, whereas south of the fault, early Devonian rocks are included in the Tobique Group [26].

Figure 1. Location and geology of the study area. (a) Map of the northern Appalachians and Atlantic Canada representing major tectonic zones, and distribution of Au mineralization [21]. (b) A regional geological map of western and northern New Brunswick displaying the gold deposits and occurrences under investigation in this study (modified from Wilson [37]. (c) Detailed geological map of the area investigated showing locations of gold mineralization (adapted from Wilson [37]).

Figure 1. Location and geology of the study area. (a) Map of the northern Appalachians and Atlantic Canada representing major tectonic zones, and distribution of Au mineralization [21]. (b) A regional geological map of western and northern New Brunswick displaying the gold deposits and occurrences under investigation in this study (modified from Wilson [37]. (c) Detailed geological map of the area investigated showing locations of gold mineralization (adapted from Wilson [37]).

3. Epithermal Gold System

Epithermal gold systems are shallow (1–2 km below the surface), hydrothermally formed deposits that commonly form in volcanic and geothermally active settings [38]. These systems result from the interaction of hot fluids, either magmatic or meteoric, with surrounding host rocks. The deposits are characterized by formation in a temperature range of approximately 150 °C to 300 °C [32,38]. Depending on fluid chemistry and alteration patterns, epithermal systems are divided into two types: high-sulfidation and low-sulfidation systems. High-sulfidation systems are linked to acidic, oxidized fluids derived from magmatic processes, whereas low-sulfidation systems are associated with near-neutral pH and reduced fluids, often equilibrated with host rocks [38,39,40]. The alteration patterns and mineralization styles differ significantly between the two deposit types. In high-sulfidation systems, acidic fluids cause extensive rock leaching, creating advanced argillic alteration zones characterized by minerals like quartz, alunite, and kaolinite. The leached rocks may later be mineralized with gold and other metals through magmatic fluid activity. During the early stages of the mineralizing event, high-temperature, metal-rich magmatic fluids dominate and mobilize ore metals. As the event progresses, these fluids gradually mix with cooler meteoric water, causing a decrease in temperature and shifts in pH and redox conditions. This fluid evolution leads to the formation of distinct alteration assemblages, which in turn facilitate the sequential precipitation of ore minerals, thereby contributing to the protracted nature of the mineralizing event. Conversely, low-sulfidation systems exhibit zonal alteration from quartz-adularia veins to argillic and propylitic assemblages [38,39]. These systems commonly form in structurally controlled vein networks where gold and silver precipitate due to fluid boiling and mixing. Epithermal gold deposits are highly variable in terms of grade and tonnage and are significant exploration targets globally. Epithermal systems occur in a range of tectonic settings characterized by high geothermal gradients, e.g., subduction zones, rift zones, and continental hot spots. Their mineral assemblages commonly include electrum, silver sulfides, and selenides in low-sulfidation systems, whereas high-sulfidation systems host minerals like enargite and luzonite. The understanding of alteration zoning, structural controls, and fluid dynamics within these systems is essential for effective exploration and resource assessment [38,39].

Epithermal Gold Mineralization in New Brunswick

Epithermal gold mineralization in New Brunswick is limited but geologically significant. The only confirmed high-sulfidation epithermal occurrence is at Chambers Settlement, in the Avalon Zone of southern NB, and is hosted by intensely altered Neoproterozoic felsic volcanic rocks. The alteration assemblage includes topaz, pyrophyllite, kaolinite, and quartz, with gold and silver as the primary metals. Yousefi et al. [41] investigated the potential for distal, low-sulfidation epithermal gold mineralization within the Evandale porphyry Cu–Mo–(Au) system in southwestern New Brunswick [41]. The research highlighted the significance of the Middle Devonian Evandale Granodiorite, an oxidized I-type intrusion with an adakitic characteristic, as a fertile magmatic source for such epithermal systems. Here, mineralization occurs in veins containing pyrite, chalcopyrite, molybdenite, and arsenopyrite, with anomalous gold, and is structurally controlled by regional fractures and faults. Additionally, the proximity of gold occurrences, such as those in the Clarence Stream area, supported the genetic link between the Evandale system and epithermal gold systems, establishing the Evandale Granodiorite as a significant exploration target for epithermal gold mineralization [41].

In northern New Brunswick, the northeastern part of the Chaleur Bay Synclinorium hosts several occurrences of epithermal breccias and veins containing base metals, along with gold and silver. These are commonly associated with felsic hypabyssal rocks as well as fine- to medium-grained mafic intrusions that are spatially associated with major splays of the Rocky Brook–Millstream Fault, e.g., the McCormack Brook and Ramsay Brook faults [26,42]. Sanchez Mora et al. [25,42] studied the Williams Brook area in northern New Brunswick, which hosts low-sulfidation epithermal gold mineralization hosted by Siluro-Devonian bimodal volcanic and sedimentary rocks, deposited in a transpressive tectonic setting during the convergence of Gondwana and Laurentia [25,42]. Gold mineralization is associated with felsic units and is associated with early-stage gold and galena in muscovite/illite veinlets and later-stage quartz veins containing native gold, sulfides, and oxyhydroxides. Geochemical and stable isotope analyses indicate a magmatic origin for the mineralizing fluids, supporting their classification as a low-sulfidation epithermal system. Khalifani et al. [43] investigated the Williams Brook epithermal gold mineralization, which is hosted in Siluro-Devonian volcanic–sedimentary rocks of the Wapske Formation and controlled by the Rocky Brook–Millstream Fault. This low-sulfidation system is characterized by gold hosted in quartz veins and stockworks within rhyolite [44]. Additionally, they investigated epithermal gold mineralization at the McIntyre Brook (MB), Mulligan Gulch (MG), and Simpson Field occurrences in northern New Brunswick and conducted pXRF and Micro-XRF analyses on drill cores from these occurrences [44]. Epithermal gold mineralization in the Mulligan Gulch area is associated with an intermediate sulfidation system and occurs primarily within quartz veins containing arsenopyrite, pyrite, and other sulfide minerals. Multiple generations of quartz veining, ranging from early deformation-related phases to later fault-controlled emplacement, indicate a protracted mineralizing event. Alteration phases such as sericite and smectite reflect varying thermal and chemical conditions within the hydrothermal system, with high-temperature alteration overprinted by lower-temperature assemblages. Gold precipitation is interpreted to result from fluid mixing, as evidenced by sulfide textures and the absence of boiling features and the presence of Ag, Pb, and Zn suggests a complex mineralizing fluid. Comparisons with nearby occurrences, such as Williams Brook, reveal similar structural controls and hydrothermal processes, though variations in temperature and mineral assemblages highlight local geological differences [45]. The McIntyre Brook epithermal gold mineralization is structurally controlled, occurring along the McIntyre Brook Fault and at lithological boundaries between felsic and sedimentary units. Gold is found in brecciated felsic rocks and quartz-carbonate veins and is associated with hematite, potassic feldspar, and albite alteration and is commonly associated with base metals such as copper, zinc, and cobalt. The Simpson Field occurrence is recognized as an epithermal gold system, distinguished by its development within low-grade metamorphic rocks. In this area, gold mineralization predominantly occurs in quartz–carbonate veins and is associated with wall rocks altered to biotite or actinolite. The Simpson Field epithermal gold mineralization is associated with pyritic quartz–carbonate alteration zones, which are locally silicified and carbonatized. Gold is primarily linked with pyrite and minor amounts of chalcopyrite and arsenopyrite in sedimentary rocks and mafic intrusions. In addition, some gold mineralization occurs along fault zones and contacts with gabbroic intrusions. Figure 2 illustrates drill core samples from the northern New Brunswick occurrence showing examples of gold mineralization in quartz veins, felsic porphyritic, and mafic intrusions with sulfide mineralization and alteration.

4. Materials and Methods

4.1. Framework of Model

To model epithermal gold mineralization, theoretical criteria for ore formation and associated geological processes must be established. Then, a database is created and structured to capture key parameters of the mineral system. Figure 3 presents the methodological flowchart employed in this study to process geospatial data and develop a mineral model for epithermal gold mineralization in Northern New Brunswick. Geological, geochemical, and geophysical layers were initially generated using suitable analytical methods. Logistic transformation functions were applied to normalize and assign continuous weights to spatial evidence layers. Subsequently, knowledge-driven and data-driven approaches were applied, namely, fuzzy logic, geometric average, and logistic regression.

We employed the concentration–area fractal method, originally proposed by Cheng et al. [46], to establish thresholds for classifying values in mineral prospectivity mapping. The C–A fractal model enables the identification of distinct mineralization patterns and spatial distributions by delineating thresholds based on the relationship between concentration and area [47]. These thresholds were then applied to enhance the classification of values derived from fuzzy logic, geometric average, and logistic regression methods. Finally, a receiver operating characteristics (ROC) curve was used to evaluate the effectiveness of the prospectivity models developed using fuzzy logic and the geometric average. Furthermore, the findings of the mineralogy and petrology studies corroborate the results of the mineral prospectivity analysis.

4.2. Dataset

4.2.1. Geological Data

Geological data used as evidence layers in this study include lithological map units (polygons) and structural features such as faults. These features provide critical insights into the geological processes controlling mineralization and are used to define mappable criteria for mineral prospectivity analysis. The geological data for this study were obtained from the 1:50,000 bedrock geology map published by the New Brunswick Department of Natural Resources [48]. Specifically, a digitized compilation of lithological rock unit maps for National Topographic System map sheets (21O/10 and 21O/07). Examining geological units and faults in these maps enables exploration geologists to better understand and interpret geophysical and geochemical data in terms of epithermal gold mineralization.

4.2.2. Surficial Geochemical till Data

The till geochemical data provided by the New Brunswick Department of Natural Resources [49] encompasses samples collected at 2000 m intervals, offering a comprehensive geochemical base that aids in understanding the distribution of geologic units, mineralization, and alteration in the subsurface. A dataset of 460 samples analyzed using inductively coupled plasma mass spectrometry (ICP-MS) and instrumental neutron activation analysis (INAA) was extracted from the provincial dataset to generate geochemical evidence layers, following standard preparation and Coda-based procedures.

4.2.3. Geophysical Data

Geophysical data, specifically airborne magnetometry, provides crucial information on both surface and subsurface geological units and lineaments [50,51]. In this study, the aeromagnetic geophysical datasets from Campbelltown (1986) [52] and Bathurst (1987) [53] surveys were utilized to detect faults, contacts, and intrusive bodies associated with mineralization. The Campbelltown dataset features a line spacing of 1000 m and a cell size of 200 m, whereas the Bathurst dataset was collected using a 300 m line spacing and a cell size of 75 m, making the two surveys complementary. Merging these datasets ensured full coverage of the study area, creating a unified geophysical framework to highlight key geological and structural features, critical for a well-founded MPM study. Various image-processing techniques have been crafted to delineate the boundaries of magnetic anomalies that are commonly associated with mineralization-related intrusions [51,54]. In this research, the total magnetic intensity data were reduced to the north magnetic pole (RTP) with an inclination of 72.96° and a declination of −20.65 2° [55]. To delineate lineaments from aeromagnetic data, several filters are used, including the RTP [50,56], the analytic signal (AS), the tilt derivative (TDR) [50,57], and the first vertical derivative (FVD) [50,58]. These tools are instrumental in enhancing the definition of geological structures within magnetic data, thereby improving the identification of faults, contacts, and other boundaries [50]. Airborne radiometric surveys can be especially useful for detecting hydrothermally altered areas close to gold deposits. Additionally, they can help to identify and pinpoint felsic intrusions and volcanic rocks, which commonly contain high concentrations of potassium (K), equivalent uranium (eU), and equivalent thorium (eTh) [51,58]. In this research, radiometric data from Campbelltown (1976) [59] and Campbelltown (1986) [52] surveys were used to help identify lithological boundaries and alteration zones associated with mineralization.

4.3. Methods

4.3.1. Log-Ratio Approach and Principal Component Analysis

Geochemical data is inherently compositional, meaning that its chemical components always sum to a fixed total, thereby forming a closed system [60,61]. Geochemical data requires preprocessing and normalizing prior to use in any MPM application. The closed system features and related challenges and the necessity and importance of converting closed systems to open data have been discussed in many studies [44,60,62,63]. Here, we employ a central log-ratio transformation (CLR), a form of log-ratio family, to properly handle the closed and compositional nature of the geochemical dataset [61,64,65]. This transformation converts compositional data from classical to Aitchison geometry, facilitating the identification of underlying patterns. Among several multivariate analysis techniques, principal component analysis (PCA) was applied to the geochemical data, enabling a more precise interpretation of the interrelationships among the dataset’s variables [51,60,61].

4.3.2. Logistic Transformation

Logistic transformation is a mathematical technique used to normalize continuous spatial data into a range of [0, 1], making it useful for integrating diverse datasets in MPM [19,66]. The transformation employs a logistic function, a sigmoid-shaped curve, to assign fuzzy weights to evidence values. These weights represent the relative importance of each data point in identifying mineralization potential. This approach avoids the discretization of data into arbitrary classes, a common practice in traditional methods, reducing exploration bias and allowing for a more accurate representation of the underlying evidence [19,66].

This study utilizes the logistic function previously applied by Yousefi et al. [67] and Yousefi and Carranza [68] to standardize the values of various evidential datasets into a unified space [66,67,68]. The function is expressed as:

$$F_{Ev}={\displaystyle \frac{1}{1+e^{-s}\left(Ev-i\right)}}$$

Here, $F_{Ev}$ represents the fuzzy weight in logistic space, ranging between 0 and 1. The parameters i and s correspond to the inflection point and slope of the logistic function, respectively, while Ev is the evidential value at each pixel in the input map (e.g., fault density or proximity to specific features).

4.3.3. Fuzzy Logic

Fuzzy logic is a mathematical framework built on the principles of fuzzy set theory, first introduced by Zadeh [69]. Fuzzy logic overlay entails using fuzzy logic principles within a geographic information system (GIS) environment to effectively address uncertainty in spatial data [3,9,69].

The method transforms evidential data using functions such as the logistic sigmoid or Gaussian functions, enabling normalization and comparability of diverse input layers. The logistic sigmoid function is expressed as:

$$F_x={\displaystyle \frac{1}{1+e^{-k}\left(x-c\right)}}$$

where x is the input variable (e.g., a geochemical value or a geophysical reading), c is the inflection point, i.e., the value of x at which the function’s output is 0.5, and k is the steepness (or slope) parameter, which controls how quickly the function transitions from values near 0 to values near 1 [67,70,71].

Evidential layers are combined using fuzzy operators like fuzzy AND, fuzzy OR, or gamma operators to produce a unified prospectivity map [9]. In this study, the final fuzzy overlay prospectivity map was created using a fuzzy gamma of 0.90.

4.3.4. Geometric Average

The geometric average method is a knowledge-driven approach used for mineral prospectivity mapping to integrate multiple evidential layers effectively. The geometric average is calculated by taking the nth root of the product of the values, where n is the total number of values. For dataset {v₁, v₂, …, v_n}, the geometric average (G_A), is computed accordingly [3,68,72].

$${\mathrm{G}}_{\mathrm{A}} ({\mathrm{v}}_1, {\mathrm{v}}_2, \dots , {\mathrm{v}}_{\mathrm{n}})={\left({\prod }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{v}}_{\mathrm{i}}\right)}^{{\displaystyle \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$n$}\right.}}=\sqrt[n]{v_1{v}_2\dots v_n}$$

G_A is particularly useful for handling uncertainties and combining diverse geological, geochemical, and geophysical datasets into a unified prospectivity model [68,72].

In this study, 21 evidential layers were multiplied together, and the 21st root of the resulting product was calculated [68].

4.3.5. Logistic Regression

Logistic regression (LR) is a versatile multivariate statistical technique especially suited to binary classification tasks, utilizing a logistic function to model the probability of a categorical outcome [3,73]. This method is particularly advantageous for mineral prospectivity mapping (MPM), where the presence of minerals, categorized distinctly as either present or absent, serves as the dependent variable. In this study, LR assesses and quantifies the relationships between multiple independent variables (e.g., source, pathways, traps), which function as evidential layers [74].

These variables are integral in predicting favorable mineralized zones, making LR a widely applied and powerful data-driven tool in MPM. The model effectively describes how these independent variables relate to the dependent variable—mineral deposits or occurrences, thereby providing statistically interpretable results that aid in the identification of mineral prospectivity areas. The probability of mineral occurrence is determined using the following Equation (1) [3]:

Z = b₀ + b_{1 × 1} + b_{2 × 2} + … + b_{n × n}

(1)

where:

Z = logit (linear transformation of the dependent variable);

b₀ = intercept;

b₁, b₂…, b_n = regression coefficients;

x₁, x₂…, x_n = independent predictor variables.

Probability Calculation: the probability (P) of mineral occurrence is derived from Equation (2) [3,75].

$$P={\prod }_{k=1}^n{A}_k$$

(2)

This equation maps the linear combination Z into a range between 0 and 1 (Nykänen and Salmirinne [76]; Porwal et al. [74]; Khammar et al. [3]).

4.3.6. Fractal-Based Discretization of MPM Maps

In geosciences, fractal models have been employed to describe the spatial distributions of geological features [77,78]. Numerous fractal models have been created and successfully implemented to analyze spatial features associated with mineralization, such as geochemical anomalies [18], structural elements [79], and geological units [80]. In this study, we applied the concentration–area (C–A) fractal model [46] to establish thresholds on MPM maps through both knowledge-driven and data-driven approaches. Specifically, we derive threshold values by identifying breakpoints or inflection points on the resulting multifractal curve; these thresholds effectively partition the MPM maps into discrete classes that reflect varying levels of mineral prospectivity. This classification not only enhances our ability to differentiate high-favorability zones from background noise but also improves the interpretation of spatial mineralization patterns, as demonstrated in previous studies [18,19].

4.3.7. Receiver Operating Characteristic

The Receiver operating characteristic (ROC) curve is a widely used statistical method for evaluating the predictive accuracy of MPM. The ROC curve uses the true positive rate (sensitivity) versus the false positive rate (1-specificity). Sensitivity measures the proportion of actual positive instances correctly identified by the model. It is calculated as TP/(TP + FN), where TP is true positive, and FN is false negative. The 1-specificity measures the proportion of actual negative instances incorrectly classified as positive by the model. It is calculated as FP/(FP + TN), where FP is false positive, and TN is true negative.

The ROC curve is generated by varying the decision threshold of the classifier. As the threshold changes, the true and false positive rates also change, resulting in various points along the ROC curve. AUC-ROC quantifies the overall performance of the classifier. A model with perfect discrimination has an AUC-ROC of 1, whereas a random classifier has an AUC-ROC of 0.5. Higher AUC-ROC values indicate better discriminative ability. The closer the ROC curve is to the upper-left corner, the better the model’s performance [3,81].

A diagonal line represents random guessing (no discrimination), whereas points below this line indicate a less reliable model performance. The method supports selecting an appropriate threshold for a given application based on the desired balance between sensitivity and specificity [9,82,83]. The ROC analysis not only validates the precision of the models but also enables comparisons among different approaches, ensuring more reliable delineation of exploration targets. This study utilizes 20 mineral occurrences, classified as gold epithermal mineralization, to develop fuzzy overlay and geometric average models [3].

5. Mineral System Approach

The concept of a mineral system approach was introduced by Wyborn et al. [84], McCuaig et al. [85], and Hagemann et al. [86]. It includes all geological factors that influence the information and deposition of mineral deposits, such as the processes responsible for transferring and concentrating mineral constituents from their source via the migration pathway to the depositional site.

The mineral systems approach involves the following concepts: (1) trigger: tectonic events that initiate a mineralization event e.g., subduction of an oceanic ridge; (2) source regions: any geochemical and tectonic processes that increases concentrations of metals and fluids; (3) pathways: any structural elements or basinal architecture that leads to the formation of conduits that facilitate the flow of fluids or magma, and operating at the lithospheric, crustal, province, and district scales; (4) driver: processes that trigger the movement of fluids within these conduits; (5) throttle: any mechanical and structural event that directs and guides the flow of fluids or magma into specific locations where concentrated mineral deposition can occur, commonly known as “trap” sites; (6) trap: any chemical and physical processes such as fractional crystallization, fluid–rock interaction, density separation, etc., responsible for the deposition of metals; (7) dispersion: geochemical and geophysical processes that can alter mineralized material and leave detectable features like geochemical and geophysical anomalies; and (8) preservation: the mechanisms that enhance mineralized material [84]. It also is a conceptual framework that helps map the geological processes [84,87,88,89]. In practice, there is an updated conceptual framework that simplifies the mineral system components from eight to three crucial components: source, pathway, and trap [83,84].

5.1. Mapping Components to Targeting Criteria

5.1.1. Source

Epithermal deposits are often associated with magmatic systems. Magmatic fluids originate from volcanic arcs or other tectonic settings that generate significant heat and magma [42,90,91]. Fluids and metals, including gold, are derived from a magmatic source [91]. The bimodal volcanic rocks and their hypabyssal equivalents of the Tobique and Chaleurs groups are assumed to be the magmatic source for gold and associated metals in the study area. Likewise, the magmas responsible for the formation of these rocks likely supplied the necessary heat and fluids crucial for the formation of epithermal gold systems [42]. The close spatial association of intrusions with mineralized zones implies their importance as a heat (and metal) source to drive these gold mineralizing systems [92,93].

5.1.2. Pathway

The pathway component of mineral systems for gold mineralization describes the critical role of transporting ore-forming fluids from their source to the site of deposition [32]. According to Gabo-Ratio et al. [91], structural pathways are critical for facilitating the migration of ore-forming fluids in epithermal gold systems. Major fault systems, such as the Rocky Brook–Millstream Fault in the northern New Brunswick area, act as primary conduits for hydrothermal fluids in the study area [26,94].

These faults, once formed, may be episodically reactivated during episodic tectonic activity, repeatedly creating zones of high permeability that facilitate the ascent of mineralizing fluids. In epithermal gold systems, fracture networks associated with extensional tectonic regimes further enhance fluid flow. These networks provide interconnected pathways for fluid migration, often linked to localized stress regimes and dilation zones, which are crucial for hosting gold mineralization. The structural evolution of the study area, combined with evidence of vein formation and reactivation, underscores the importance of fault and fracture systems as dynamic pathways for hydrothermal processes [32,38,42].

5.1.3. Trap

Gold and associated metals are commonly deposited in veins and breccias in a number of favorable settings including faults and fault intersections, volcanic domes, and contacts between lithologic units of contrasting rheology. In many cases gold-bearing fluids mix with meteoric waters, reducing temperature and triggering precipitation of gold and associated minerals. Physico-chemical processes, such as boiling, cooling, and fluid mixing, further enhance the trapping potential by changing fluid chemistry (e.g., pH, redox state), leading to further gold precipitation. Stable isotopic analyses (O, H, S) and fluid inclusion studies in the Williams Brook area indicate gold deposition occurred at ~200 °C and was likely triggered by fluid mixing [42]. The trap component in the mineral system approach for epithermal gold mineralization integrates the spatial distribution (enrichment and depletion) patterns of key elements (e.g., Au, As, Cu, Mo, Pb, Sb, and Zn) as critical evidence layers. These elements act as proxies for hydrothermal processes and can delineate zones where fluids were focused and gold was deposited. Multi-element anomalies of As, Sb, Pb, and Zn can outline hydrothermal alteration zones that function as chemical traps [95]. In this study, the relationship between the first two principal components (PC1 and PC2) highlights the important relationship among several elements (Figure 4).

6. Results and Discussion

MPM applied in this study effectively identified high-favorability zones for epithermal gold mineralization in northern New Brunswick. In the study area, there are 20 occurrences of epithermal gold mineralization.

Table 1. Evidence layers employed in this study.

Fundamental Processes of the Gold Mineralization	Outline of Targeting Criteria	Predictor Maps	Source of Data	Practices Employed	Figure
Source	Tobique Group and Chaleurs Group mafic and felsic intrusive rocks with Silurian-Devonian aged	Ramsay Brook Gabbro, Upsalquitch Lake Gabbro, Silurian Felsic Intrusive, Portage Brook Troctolite mafic intrusive, Mulligan Gulch Porphyry felsic intrusive, Mount Bailey Granite, Mount LaTour Gabbro, Mount Elizabeth Gabbro, Mount Elizabeth Granite	1:50,000 bedrock geology maps (21O/10 & 21O/07) [48]	The Euclidean distance was calculated using intrusive rocks, including both Granite and Gabbro units.	Figure 5a
		High values of RTP and Analytic Signal	Campbelltown aeromagnetic data with 200 m resolution [52] Bathurst aeromagnetic data with 75 m resolution [53]	After applying the reduction to the pole (RTP) and analytic signal (AS) filters, the Euclidean distance was determined from the RTP anomalous regions [19].	Figure 5b,c
		High values of K, eU, and eTh,	Campbelltown radiometric data with 100 m resolution [52] Campbelltown radiometric data with 1000 m resolution [96]	The Euclidean distance was computed using the radiometric K, eU, and eTh elements.	Figure 5d–f
Pathway of hydrothermal fluid	Faults, lineament, and lithological contacts	Proximity to mapped major, minor, and thrust faults	1:500,000 bedrock geology map [48]	Major, minor, and thrust faults were identified through geological map analysis, and the Euclidean distance from these fault structures was subsequently calculated.	Figure 5g–i
		Proximity to extracted Magnetic lineament features using the value of tilt derivative (TDR) and first vertical derivative (FVD)	Campbelltown aeromagnetic data with 200 m resolution [52] Bathurst aeromagnetic data with 75 m resolution [53]	Shallow and deep magnetic structures were highlighted using various derivative-based techniques, including the tilt derivative and first vertical derivative. Following this enhancement, the Euclidean distance to the identified structures was calculated [19].	Figure 5j,k
		Proximity to lithological contacts	1:500,000 bedrock geology map [48]	Faults were identified through the analysis of geological maps, and the Euclidean distance to these fault structures was then calculated.	Figure 5l
Trap and metal deposition and emplacement	Geochemical indicator elements	Anomalous signature of Au, As, Cu, Mo, Pb, Sb, Zn, and W elements	Geochemical till data collected at 2000 m sampling intervals [49]	Gold pathfinder geochemical signatures were analyzed after applying a central log-ratio transformation, and Euclidean distances to anomalous basins were calculated.	Figure 5m–t
	Host rocks	Proximity to Greys Gulch Formation felsic, mafic volcanic and sedimentary rocks, Wapske Formation felsic and mafic volcanic, Wapske Formation mixed volcanic and sedimentary, Free Grant Formation sedimentary rocks	1:50,000 bedrock geology maps (21O/10 & 21O/07) [48]	Euclidean distances to felsic and mafic volcanic rocks were calculated.	Figure 5u

provides a summary of the evidence layers utilized in this study, whereas each layer is shown individually in Figure 5a–u. The evidence layers indicating the likely source of mineralization in the study area encompass the Late Devonian Ramsay Brook Gabbro, Mount LaTour Granite and related gabbro, Mount Elizabeth Granite and related gabbro, Mount Bailey Granite, Portage Brook Troctolite, various unnamed Silurian felsic intrusions, and the Silurian-Devonian Mulligan Gulch Porphyry, any of which may have provided a magmatic contribution to these systems (Figure 5a). Based on RTP and AS maps (Figure 5b,c), rock units exhibiting high aeromagnetic anomalies are identified as felsic and mafic intrusions, both of which contain magnetite minerals [50].

In the study region, felsic volcanic rocks of the Benjamin and Wapske formations are associated with several Au occurrences in the study area and exhibit high K, eU, and eTh (Figure 5d–f). These combined geological and geophysical layers demonstrate the critical role of magmatic processes in generating mineralizing fluids and metals [50].

Pathways for epithermal gold mineralization in the study area are often evidenced by geological structures and geophysical indicators that facilitate the movement of hydrothermal fluids. Proximity to major and minor, faults and magnetic lineaments serve as critical evidence layers for identifying fluid conduits (Figure 5g–i). Faults create zones of high permeability that channel fluids through the crust. Likewise, proximity to lithological contacts (Figure 5l), where rocks of contrasting rheology provide fluid pathways by enhancing structural complexity and fluid flow. Geophysical evidence such as high values of the first vertical derivative (FVD) and total tilt derivative (TDR) can further delineate areas of increased areas of structural favorability, highlighting potential pathways (Figure 5j,k). These combined structural and geophysical evidence layers are essential for mapping and understanding the features formed by hydrothermal fluid migration and gold deposition.

Geochemical zonation, exemplified by the proximal distribution of elements such as Au, As, Sb, Zn, Pb, Cu, Mo, and W, serves as an indicator for mineralized traps and reflects the dynamic evolution of fluid conditions and depositional environments (Figure 5m–t).

Pathfinder elements identified through till data analyses strongly correlated with high-potential zones. These anomalies align with zones of hydrothermal alteration, particularly sericitization, and silicification, indicating their reliability as proxies for epithermal gold mineralization. For example, the Williams Brook area’s geochemical signatures exhibit elevated As and Pb, which correlated with high-potential zones delineated by the models. Alteration patterns, including sericitization, silicification, and chloritization, provided further insights into mineralizing processes [25,42] These features, often concentrated near major fault zones, are critical indicators of hydrothermal activity. Drill core samples from Williams Brook and Mulligan Gulch highlighted multiple phases of quartz veining, pyrite mineralization, and alteration halos, reinforcing the presence of active hydrothermal systems. The models’ ability to predict areas with such alteration patterns validated their effectiveness in delineating epithermal gold systems. Furthermore, the Devonian mafic and felsic volcanic host rocks of the Wapske Formation, along with the mafic volcanic rocks of the Greys Gulch Formation, play a crucial role as vectors to mineralize traps and the depositional environment (Figure 5u).

This study applied the concentration–area fractal method to classify and evaluate various MPM maps using fuzzy logic, geometric average, and logistic regression, as shown in Figure 6a,c,e [18]. Based on the C–A fractal method, we discretized the mineral prospectivity maps for fuzzy logic, geometric average, and logistic regression models (Figure 6b,d,f). Here, the MPM result was categorized into six classes according to the relevant C–A multifractal curve (Figure 6). The first and second thresholds for fuzzy logic (0.90 and 0.94), geometric average (0.92 and 0.95), and logistic regression (0.86 and 0.94), depicted in deep red and orange, signify high favorability on three MPM maps derived from fuzzy logic, geometric average, and logistic regression (Figure 6b,d,f). Yellow stars indicate the locations of epithermal gold mineralization in the study area and closely align with the highly favorable areas identified. This suggests that the prospectivity map effectively identifies high favorability values at known gold occurrences while highlighting areas of potential. The identification of favorable areas near known gold mineralization suggests accurate identification of zones exhibiting similar characteristics to the known occurrences. The fuzzy gamma map identifies broader, more continuous zones of high potential, whereas the geometric average method produced somewhat more localized but still relatively diffuse areas of potential. In contrast, the logistic regression map identifies more sharply defined high-potential zones closely aligned with known gold occurrences. Exploration efforts should focus on areas with the highest prospectivity, irrespective of the modeling method employed [9].

In this study, we utilized the ROC curve, which compares the true positive rate with the false positive rate across various classification thresholds. With an area under the curve (AUC) exceeding 0.9, the model demonstrates strong discriminative capability, effectively distinguishing between classes. Furthermore, its ROC curve lies predominantly in the top-left portion of the plot, indicative of both high sensitivity and specificity—reinforcing the model’s overall robustness. The ROC curves and corresponding mean AUC values for the two knowledge-driven models are presented, with results indicating that the fuzzy overlay model achieved the highest AUC (0.97), followed by the geometric average model (0.93). These findings demonstrate exceptionally favorable performance (Figure 7a,b).

To validate the mineral prospectivity map derived from the logistic regression model, the mineralogical and petrological characteristics of the identified promising zones were examined. Figure 8 illustrates the presence of gold and common pathfinder minerals—including pyrite, galena, and pyrrhotite—in gold occurrences such as Williams Brook, McIntyre Brook, Mulligan Gulch, and the Simpson field. The mineral indicators, observed within the high-potential areas identified by the model, provide confidence in the accuracy of the prospectivity predictions. The models demonstrated a strong alignment with known gold occurrences within the Tobique–Chaleur Zone, which is characterized by its complex geology and significant structural controls. A major focus of this study was the development of meaningful geological, geochemical, and geophysical evidential layers to model using a mineral system approach [3,87]. These layers highlight the critical role of faults as fluid pathways, the significance of felsic and mafic intrusive rocks as source components, and geochemical signatures, as well as the importance of felsic and mafic volcanic trap rocks. The MPM maps (Figure 6), further illustrate the effectiveness of comparing knowledge-driven methods (fuzzy logic and geometric average) with a data-driven method (logistic regression). This two-pronged approach ensures that both expert insights and statistical analyses contribute to the delineation of prospective areas.

The data-driven modeling approach demonstrated significant advantages in terms of reducing the exploration area while maintaining high predictive accuracy. Logistic regression is used to make robust predictions by efficiently processing large datasets and identifying meaningful patterns that traditional methods might miss, thereby improving predictive accuracy. This method provides a systematic and quantitative framework for decision-making. For example, Figure 6 reveals that known occurrences consistently overlap regions of high predicted favourability, with new target areas identified in underexplored areas. This capability not only enhances exploration efficiency but helps to mitigate risks associated with greenfield exploration.

Fuzzy logic and geometric average both draw on expert geological insights to assign weights to evidence layers, making them particularly suitable for areas with limited data. In this study, they identified extensive zones of high mineral potential, with fuzzy logic demonstrating higher predictive accuracy (AUC = 0.97) than the geometric average method (AUC = 0.93). By contrast, data-driven logistic regression relies on training data from known occurrences, capturing complex relationships among variables and generating more narrowly defined high-potential areas. However, it demands a sufficiently large dataset and risks overfitting when data are scarce. Integrating these approaches combines the breadth of expert-based knowledge with rigorous statistical mapping, improving the efficiency and reliability of exploration targeting. In addition, all three methods—fuzzy logic, geometric average, and logistic regression—prove adaptable in addressing uncertainties from sparse or variable data, underscoring their value in complex geological terrains.

These methods significantly reduced the search area, thereby streamlining exploration efforts and reducing costs. Figure 8 presents a 3D view of the study area, showing the distribution of gold occurrences and the high-potential areas identified in this study. Overall, this study underscores the importance of integrating geological knowledge with advanced computational methods to address the challenges of mineral prospectivity mapping.

7. Conclusions and Recommendations

The major accomplishments of this study can be summarized as follows:

• This study applied an MPM technique to identify high-favorability zones for epithermal gold mineralization in Northern New Brunswick. By integrating geological expertise with advanced computational methods, namely, fuzzy logic, geometric average, and logistic regression, it achieved a high level of predictive accuracy. The identified zones strongly correlated with known gold occurrences, especially in the structurally controlled Tobique–Chaleur Zone.
• Geological, geochemical, and geophysical datasets were crucial in defining evidence layers that represented the mineral system components. The alignment of high-favorability zones with key structural features, such as the Rocky Brook–Millstream Fault system, demonstrated the importance of faults as fluid pathways. The study also underscored the role of felsic and mafic intrusive rocks as essential magmatic sources, contributing to gold mineralization.
• The fuzzy gamma model delineated extensive, continuous zones with high potential. The geometric average method identified more localized but somewhat dispersed areas of interest. The logistic regression model generated sharply defined high-potential zones that closely corresponded with known occurrences. Integrating knowledge-driven approaches (fuzzy logic and geometric average) with a data-driven method (logistic regression) enhanced the accuracy and effectiveness of the prospectivity analysis.
• The high area under the curve (AUC) values obtained from the receiver operating characteristic (ROC) curves validate the robustness and reliability of the MPM methods used in this study. Specifically, the fuzzy overlay model achieved the highest AUC (0.97), followed by the geometric average model (0.93). These values indicate strong predictive accuracy, confirming that the models effectively delineate high-favorability zones for epithermal gold mineralization in Northern New Brunswick.
• The study identified indicator minerals and pathfinder elements that strongly correlate with high-potential zones. Drill core analysis confirmed the presence of gold-bearing quartz veins and associated mineralization, further reinforcing the accuracy of the MPM models.

In addition, future work might include the following:

6.

Incorporate field-based validation (e.g., additional drilling or detailed mapping) in the newly identified high-favorability zones to confirm the presence of epithermal gold mineralization.

7.

Explore machine-learning techniques beyond logistic regression (e.g., random forests or support vector machine) to capture nonlinear relationships and optimize predictive accuracy.

8.

Acquire or integrate higher-resolution geophysical and geochemical datasets—where it is feasible to further reduce uncertainty and improve the spatial resolution of prospectivity maps.

9.

Apply the presented methodology to related mineral deposit styles (e.g., orogenic gold, base-metal skarns) within the Appalachian or other structurally complex terrains, assessing model adaptability and limitations.