Journal of African Earth Sciences 206 (2023) 105020 Contents lists available at ScienceDirect Journal of African Earth Sciences journal homepage: www.elsevier.com/locate/jafrearsci Geospatial modelling of mineral potential zones using data-driven based weighting factor and statistical index techniques Prince Ofori Amponsah a, Eric Dominic Forson b,* a Department of Earth Science, School of Physical and Mathematical Sciences, University of Ghana, P.O. Box LG 58, Legon, Accra, Ghana b Department of Physics, School of Physical and Mathematical Sciences, University of Ghana, P. O. Box LG 63, Legon, Accra, Ghana A R T I C L E I N F O A B S T R A C T Keywords: Mineral prospectivity models (MPMs) are significantly essential in delineating target zones with the optimum Mineral prospectivity modelling likelihood of containing a particular sought-after mineral deposit. This present study carried out mineral po- Weighting factor tential mapping over the Collette Prospecting Licence (PL) Area of north-western Ghana using bivariate data- Statistical index driven spatial statistical models composed of statistical index (SI) and weighting factor (WF) approaches. In Area under receiver operating characteristics Geophysical datasets the first instance, the geographic coordinates of variously known locations of artisanal mining operations as well as high Au concentration locations were mapped during a field survey. As a result, 181 known locations of Au occurrences were identified, out of which 127 (70%) were selected randomly for training and creating the mineral prospectivity models, whereas the remaining 54 (30%) were used to assess and validate the accuracy of the predictive models produced. The efficacy of mineral prospectivity models generated enormously depends on the appropriate selection of mineral-related factors. In this study, the following mineral-related condition factors (evidential layers) comprising analytic signal, lineament density, uranium-thorium ratio, uranium, potassium- thorium ratio, potassium, reduction-to-equator, and geology were used. The aforementioned evidential layers were derived and sourced from geophysical and geological datasets, which were later prepared for the generation of the models in a geographic information systems (GIS) environment. Finally, the validation of the mineral prospectivity models generated was carried out by applying the receiver operating characteristics (ROC) curve. The estimated results based on the ROC plots obtained for the predictive models showed that the area under the ROC curve (AUC) scores obtained for the SI-based and WF-based mineral prospectivity models were respectively, 0.780 and 0.733. Hence, it can be concluded that both mineral predictive models created in this study produced reasonably good accuracy (AUC score greater than 0.7) in predicting the potential zones of gold mineralisation occurrences within the Collette PL Area of north-western Ghana. These MPMs can serve as essential models for mineral exploration programmes within the study area. 1. Introduction generation of mineral prospectivity models (MPM) a multi-criteria de- cision-making (MCDM) activity whose output outlines prospective zones Mineral prospecting fundamentally concerns the exploration and of mineral occurrence (e.g., Carranza and Laborte, 2015; Yousefi and delineation of new areas characterised by ore-bearing minerals over a Carranza, 2015; Kashani et al., 2016; Forson et al., 2020; Forson and specified region of interest. The aforementioned basis for mineral pro- Menyeh, 2023). In synthesising geospatial datasets towards the gener- specting dwells on differentiating highly prospective zones from pro- ation of a predictive model in a geoscientific context, many approaches spectively low regions of a particular sought-after mineral within the have been developed and have been generally grouped into area explored. In carrying out predictive modelling over a specified area knowledge-driven and data-driven approaches (e.g., McKay and Harris, based on the objective of delineating new zones that are prospectively 2016; Sun et al., 2020; Parsa and Carranza, 2021; Zhang et al., 2021; characteristic of the target mineral, various geospatial exploratory Amponsah et al., 2022a,b; Forson et al., 2022b). Predictive modelling datasets comprising geophysical, geological, and geochemical layers that employs the knowledge-driven technique solicits the opinions of ought to be acquired, analysed, and synthesised. This makes the geoscience experts in relation to the sought-after natural resource * Corresponding author. E-mail addresses: edforson@ug.edu.gh, ericdforson@gmail.com (E.D. Forson). https://doi.org/10.1016/j.jafrearsci.2023.105020 Received 16 January 2023; Received in revised form 14 July 2023; Accepted 28 July 2023 Available online 1 August 2023 1464-343X/© 2023 Elsevier Ltd. All rights reserved. P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 1. Map of the north-western Ghana showing various administrative districts (Study area is marked in red). (mineral, groundwater, etc.) to assign weights to various Zhenjie et al., 2021; Zhang et al., 2022). geoscientific-derived evidential layers that are to be integrated to The study focuses on identifying locations for areas for exploratory generate the model. Commonly used knowledge-driven approaches in drilling within a large region in the Collette PL Area of the North- geospatial predictive modelling that have been applied in recent liter- western part of Ghana by incorporating diverse evidential layers ature comprise of the analytical hierarchy process (Barati et al., 2018; sourced from geological and geophysical datasets. Furthermore, this Akinlalu et al., 2021), the fuzzy analytical hierarchy process (Forson study compares the performance of mineral prospectivity models that et al., 2020; Khosravi et al., 2021), fuzzy logic (Sanusi and Amigun, would be generated based on the use of data-driven statistical tech- 2020; Abdelkareem and Al-Arifi, 2021), and the technique for order niques comprising weighting factor (WF) and statistical index (SI) ap- preference by similarity to the ideal solution (TOPSIS) (Mansouri et al., proaches that have been successfully employed in determining the 2017; Rahimi et al., 2020). The other category of data synthesis tech- influence of various evidential layers that are to be integrated into a niques (known as the data-driven method) employed in predictive prospectivity model in various geoscience contexts (Meinhardt et al., modelling determines the spatial correlation between each of the 2015; Khosravi et al., 2016). evidential layers used with respect to the known location of the occur- rence of the sought-after natural resource (such as gold, copper, and 2. Study area and geological setting water). Thus, in the data-driven method, known locations of occurrence of the sought-after mineral are employed as training points to determine 2.1. Study area the weight or influence of each of the evidential layers to be incorpo- rated in the predictive model to be produced (Carranza, 2008; Forson The Collette Prospecting Licence Area is located approximately 800 et al., 2022b). In view of this, a fundamental drawback in the application km NNW of Accra in the Upper West region of Ghana, specifically the of data-driven approaches is the unavailability of known locations of the Wa-East district (Fig. 1). The area lies within longitudes 2◦06′ and 2◦01′ sought-after minerals locally at the deposit scale. However, when and latitudes 10◦9′ and 10◦5’ using the WGS datum ellipsoid. The employed over regions where a considerable amount of known mineral regional capital of Wa is located 60 km southeast of the tenement. There occurrences are found, these data-driven methods are able to generate are no communities in the tenement area. Access from Wa is through the objective weights for the evidential layers to be integrated rather than Wa-Sandema road to Bulenga, and south of the prospecting licence area, relying on expert opinions as in the case of knowledge-driven ap- access is via networks of graded laterite roads that link the major towns proaches. Examples of data-driven MPM techniques include frequency and small villages. There are also several poorly graded third-class ratio (Mathew and Ariffin, 2018; Kusuma et al., 2019), weight of evi- laterite roads that become impassable during the peak periods of the dence (Zhang et al., 2016; Fu et al., 2021), evidence belief function June to October wet season. The area is relatively flat, with low east- (Carranza, 2015; Ford et al., 2016), statistical index (Ghasemzadeh west rolling hills occupying the Julie belt. The climate in the arid et al., 2022), weighting factor (Esmaeiloghli et al., 2021), Shannon en- Sahel Belt of northern Ghana is characterised by annual rainfall between tropy (Al-Abadi, 2017), and machine learning methods (Lin et al., 2021; 1000–1250 mm/year, largely falling in the rainy season from June to 2 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 2. Map of the west african craton (WAC). October (that peaks in August–September; Dickson and Benneh (1988)). According to Agyei-Duodu (2009) the metasediments were deposited around 2139 ± 2 Ma, with the late intrusive granitoids around 2104 ± 1 Ma. Juxtaposing the Wa-Lawra to the east is the Koudougou-Tumu 2.2. Regional geology of the study area domain, bounded by the Jirapa shear zone. This domain is composed of high-grade granulite facies 2187 ± 3 Ma to 2165 ± 7 Ma granitoid The Precambrian West African shield is composed of both Archean genies, biotite schist, 2134 ± 1 Ma hornblende-biotite tonalite, and and Paleoproterozoic nuclei, which form two major domains known as 2124 ± 2 Ma biotite hornblende monzodiorite. These rocks have been the Reguibet shield (nWAC) to the north and the Leo-Man shield (sWAC) intruded by late plutonic suites composed of 2112 ± 1 Ma biotite to the south. The Archean portion of the sWAC is termed the Kénéma granite, 2106 ± 1 Ma biotite granite, hornblende biotite tonalite, and Man Shield, and the Paleoproterozoic portion is termed the Birimian or quartz diorite (Agyei-Duodu, 2009; Block et al., 2016; Amponsah et al., the Baoulé-Mossi domain (Rocci, 1965; Baratoux et al., 2011; Block 2016). Lying east-west within the Koudougou-Tumu domain is the Julie et al., 2016; Masurel et al., 2022). The Birimian domain crops out in Belt. This belt is composed of basalt, andesite, and volcaniclastic sedi- countries such as Ghana, Burkina Faso, Mali, Niger, Mali, Senegal, Ivory ments. They are marked by allochthonous bodies or thrust nappes as a Coast, and Guinea (Jessell et al., 2012; Amponsah et al., 2015; Diatta result of low-angle thrust faults, with the fault planes developing shear et al., 2017; Eglinger et al., 2017; Feng et al., 2018), and it is composed zones that host the Julie gold mineralisation (Amponsah et al., 2015; of arcuate granite-greenstone belts and sedimentary basins (Feng et al., Block et al., 2016; Nunoo et al., 2022) as well as the Collette main de- 2019). In Ghana, there are six (6) of these arcuate granite-greenstone posit (Azumah Resources Limited, 2018). South of the Julie belt is the belts, namely the Kibi-Winneba belt, the Ashanti belt, the Sefwi belt, Bole-Bulenga domain. This domain is composed of 2150 ± 4 Ma biotite the Bui belt, the Bole-Navrongo belt, and the Wa-Lawra belt, and granitoid, k-feldspar-rich granitoid, and hornblende tonalite. These architecturally, all the belts have a northeast-southwest structural grain rocks are overlain by patches of Tarkwaian detrital sediments composed (Fig. 2; Agra et al., 2023), except for the Wa-Lawra Belt, which has a of sandstones and conglomerates. These detrital sediments are highly north-south orientation and is part of the larger Boromo belt, which rich in magnetite. South and southeast of the Bole-Bulenga domain and extends into Burkina Faso (Amponsah et al., 2016; Salvi et al., 2016; the Koudougou-Tumu domain (Fig. 3a), respectively, is the crustal-scale Feng et al., 2019; Sapah et al., 2021; Nunoo et al., 2022; Amponsah Bole-Nangodi greenstone belt. The main structural grain of this shear et al., 2023). The geology of northwest Ghana is defined by the zone is northeast-southwest. The belt is composed of 2179 ± 2 Ma to north-south oriented Wa-Lawra belt and the northeast-southwest ori- 2156 ± 1 Ma hornblende biotite granitoid, 2159 ± 4 Ma volcaniclastic ented Bole-Nangodi greenstone belts that meet at a tangent in north- sediments, 2139 ± 2 Ma wacke-dominated sediment, basalts, andesites, western Ghana. The Wa-Lawra greenstone belt is composed of rhyolites, and banded manganese formations. These rocks have been metasedimentary rocks (mainly shales and greywackes), metavolcanics intruded by late 2112 ± 2 Ma hornblende biotite tonalites, 2120 ± 1 Ma (basically basalts, rhyolites, and andesites), and intrusive granitoids. 3 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 3. (a) Regional geological map of the north-western Ghana showing the study area in red (modified after Amponsah et al., 2015) (b) simplified geological map of the study area (thus the Collette tenement) showing the lithologies and structural measurement. biotite muscovite granites, and 2134 ± 6 Ma tonalites (Agyei-Duodu, structural junction where the E-W oriented dextral jog shear zone of the 2009). The southern margins of this belt have been covered by the Julie belt tangentially meets the NE trending shear zone of the Neoproterozoic Volta Basin sediments (Fig. 3a). Bole-Bulenga domain. The E-W oriented shear zone, which is denoted by the N–S shortening in the Julie belt, marks the initial deformation (D1) 2.3. Structural framework of the study area in the Julie belt, which was formed as a result of the nappe stacking event that affected all the rocks in the belt (Amponsah et al., 2015). This ◦ ◦ The summary of the structural disposition of the Collette deposit is shear zone dips 35–70 to the north, and the L1 lineation trends 025 mainly from the work of Amponsah et al. (2015) and Block et al. (2016). with plunges between 35 ◦ to 80◦. Boudinage syn-deformational quartz Architecturally, the Collette deposit, just like the Julie deposit, is located veins oriented in the direction of (or parallel to) the E-W shear zone were on the E-W shear zone of the Julie belt, but the former occurs at a observed. In the volcaniclastic rocks, F1 vertical plunging folds verging 4 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 towards the east form E-W trending shear foliation, which is parallel to 3. Methodology the axial plane of the F1 folds. The NE trending shear zone Bole-Bulenga domain, according to 3.1. Geoscientific database construction using GIS Block et al. (2016), marks a N–S directed extension event and represents a second deformational event (D2) that affected all the rocks in the study The fundamental procedure of a mineral prospectivity model is the area. This shear zone denotes the boundary between the low-grade Julie collection of spatial data and the creation of a geospatial database from belt and the high-grade Bole-Bulenga domain. This shear zone has dips which the essential mineral-based geoscientific evidential layers can be ranging from 40 to 50◦ to the north and has affected the Tarkwaian derived. A geospatial database comprising associated factors of the polymitic conglomerates and the high-strained migmatitic, ortho-, and sought-after mineral was created in this study. It is noteworthy that the para-gniesses. L2, which is stretching mineral lineation found on the reliability of a mineral prospectivity model (MPM) produced is enor- shear zones or foliation, trends north with a 20◦ plunge. Both shear mously dependent on the quality and quantity of data available, the zones exhibit brittle-ductile transition characteristics with deformed chosen scale for the study area, the appropriate factors chosen as well as north-side-up sigmoidal shear zone indicators (mainly from rotated, the data integration methods selected for the analysis and modelling. In deformed mineral fabrics with pressure shadows). view of this, eight geoscientific evidential layers comprising the analytic The third deformation (D3) event observed in the study area is a signal, lineament density, lithology, potassium (K) concentration, brittle late-stage event marked by NE-SW and NW-SW shortening events potassium-thorium (K-eTh) ratio, magnetic intensity-based reduction- represented by the numerous late NE-SW and NW-SE oriented faults to-equator (RTE), uranium (eU) concentration, and uranium-thorium with dips ranging from 36◦ to 60◦ to the NW (Fig. 3b; Amponsah et al., (eU-eTh) ratio were generated from airborne magnetic, airborne 2015). radiometric, and geological datasets. The magnetic and radiometric datasets were obtained from Azumah Resources Limited (ARL; Azumah 2.4. Mineralisation style of the study area Resources Limited, 2018). The radiometric-based evidential layers, which number up to four layers (K concentration, K-eTh ratio, eU con- The mineralisation in Collette is predominantly quartz vein-hosted, centration and eU-eTh ratio), were generated by using the minimum which occurs within metavolcaniclastics rocks (vs) and sheared curvature gridding method to grid K and eU concentration channels in potassic-altered granites and gneisses (comp-gn). Primarily, the meta- Geosoft Oasis Montaj software. This aforementioned software was also volcaniclastic rocks are composed of shale and siltstone facies with its used to generate the two ratio layers (K-eTh and eU-eTh ratios). In the mineralogy composed of quartz and minor k-felspars, and that of the case of the airborne magnetic data, the magnetic intensity data acquired comp-gn is composed of plagioclase, hornblende, and biotite with over the study area was first gridded using the minimum curvature accessory minerals composed of titanite, perthitic alkali feldspars, and gridding method. Since the study area is in a low magnetic latitude zone, zircon (Amponsah et al., 2015; Azumah Resources Limited, 2018). The the observed magnetic intensity values and their corresponding comp-gn has a primary tonalitic composition overprinted by a migma- geological responses are asymmetric in nature. Thus, the titic metamorphic assemblage of epidote, calcite, fine-grained biotite, reduction-to-equator filter was applied to the magnetic intensity grid so and rutile, and that of the volcaniclastics are graphite, calcite, and as to remove the asymmetric effects associated with magnetic intensity chlorite. This mineral assemblage is characteristic of greenschist facies datasets collected at low magnetic latitudes. Subsequently, the and indicates that the mineralisation is under greenschist facies condi- reduction-to-equator evidential layer was generated, which character- tions. Meso- and micro-structural analysis during geological mapping ises a symmetric distribution of magnetic responses over their respective and drill core logging in the Collette Prospective Licence Area identified sources. The analytic signal layer was created by applying the analytic three main deformational events, which coincide with the D1, D2, and signal filter to the magnetic intensity grid. The CET (Centre for Explo- D3 phases described by Block et al. (2016) and Amponsah et al. (2015). ration Targeting) grid analysis technique was applied to the RTE layer to Gold mineralisation at Collette occurs as micron-sized inclusions and generate the lineament density layer that outlines the intensity of free gold within fractures of euhedral pyrites that have precipitated occurrence of lineaments such as dikes, faults, fractures, etc. over a re- within fracture-controlled crystalline quartz-carbonate-tourmaline gion of interest. All the evidential layers sourced from the geophysical veins associated with D1 within the sheared comp-gn as well as trans- datasets (magnetic and radiometric) were pre-processed and processed posed quartz veins that occur within the east -west shear zone that has in the Geosoft Oasis Montaj software. These aforesaid affected the volcaniclastic rocks (vs) and the Tarkwaian sediments geophysically-extracted layers were then exported to raster format for (Tarkw). Gold in the comp-gn rocks is associated with other metals and further processing in ArcGIS 10.8. The geological layer (Fig. 3b), which REE minerals within pyrite grains such as silver, bismuth, tellurium, comprises various lithological classes within the study area, was xenotime, bastnäsite, and trace amounts of base metals such as copper extracted from geological data compiled by the Ghana Geological Sur- and lead (Amponsah et al., 2015). The ore body defined in Collette is vey Authority (Agyei-Duodu, 2009) in vector shapefile format. In order made up of a series of plunging ore shots (mostly likely of cylindrical to ensure that all evidential layers are in the appropriate format for shape) which trend E-W and dip between 50◦ and 70◦ to the north. The further processing and analysis, the geology layer was also converted to bulk of the gold inventory is found at the intersection between the E-W a raster format. The lithological classes within the study area comprise shear zone defined by the Julie structural grain and the northeast bas (basalts), comp-gn (granite/gneisses), G2 (tonalite/granitoid), G4 structure defined by the Bole-Bulenga domain. The ore thickness ranges (granitoid), tarkw (Tarkwaian sediments), and vs (volcaniclastic sedi- from 10 to 20 m. The sulphide assemblage in the gold-bearing quartz ments with siltstone and shale facies). Afterwards, the evidential layers veins is dominated by pyrite in the tonalites and arsenopyrites in the were imported into a GIS environment (ArcGIS 10.4 and ArcGIS Pro) for sediments. In the comp-gn, the pyrite occurs as single euhedral crystals, further processing and rescaling towards the attainment of the same cell disseminated grains within vein alteration selvedges, and irregularly size of 2 m (yielding a total area pixel count of 10,630,746) for each of shaped “clots”. Chalcopyrite is quite rare and is interpreted to have the evidential layers used. formed at the same time as pyrite (and gold; Amponsah et al. (2015)). The bulk of the gold inventory, as well as high-grade mineralisation, is 3.2. Training and validation datasets found at the intersection between the D1 E-W shear zone and the D3 late northeast or northwest brittle faults. The ore thickness ranges from 10 to The use of bivariate statistical modelling in a data-driven based 20 m. mineral potential mapping essentially thrives on the presence of known locations of the mineral, whose spatial distribution (in terms of their respective X and Y coordinates) is to be determined over a region of 5 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 interest. Hence, 181 locations of known gold occurrences, which have the statistical information technique, n is the total number of evidential been derived from historically mapped locations of artisanal mining layers, and i represents the classes in each evidential layer j. operations as well as locations of anomalously high Au geochemical assay values in parts per billion (obtained from Azumah Resources 3.3.2. Weighting factor (WF) model Limited), were analogously converted to obtain a total Au pixels of 723 The weighting factor modelling method, which was first introduced counts. These Au occurrences were used to train and validate the min- by Cevik and Topal (2003), is a modified edition of the statistical index eral prospectivity models to be produced. A proportion of the known method, which has been employed successfully in geospatial modelling locations of gold occurrences (equal to seventy percent) were employed (Meinhardt et al., 2015; Khosravi et al., 2016). This is because, in car- in the training of the predictive models generated and are thus referred rying out the weighting factor method, weights determined are based on to as training data. The remaining thirty percent of the known gold lo- the utilisation of the WSI values obtained. In carrying out the weighting cations were used to validate the predictive models produced based on factor technique for an i-th class of a given evidential layer, the total the use of the data-driven bivariate methods discussed under Section weighting index value for that class (TSi is determined as a product of the 3.3. class WSI value and the number of pixels associated with the known Au occurrences within the said class i (as shown in equation (3)). Thus, for a 3.3. Data-driven geospatial models class i in a given evidential layer j, the TSi represents the rank of the class i among all the other classes within the evidential layer. In this study, predictive models generated for the favourable zones of ∑n { ( } mineral occurrence over the Collette PL Area were carried out using two TSIi = Npix Auij ×WSI (3) bivariate statistical models comprising the statistical index and the i=1 weighting factor. Where Npix (Auij) is the number of pixels of known Au occurrences 3.3.1. Statistical index (SI) model within class i and n represents the number of classes within the chosen In terms of geospatial modelling, the introduction of the statistical evidential layer. Afterwards, the weight factor of each of the evidential index as a bivariate statistical modelling technique was first employed in layers j to be synthesised is generated based on the min-max approach as the delineation of susceptible zones of landslide occurrences by Van shown in equation (4). Westen (1997). For its employability in geospatial modelling, this TSi − min TSi method essentially requires that a geospatial modeller selects and maps WFi = × 9 + 1 (4) max TSIi − min TSi the relevant locations of a known natural resource or hazard (in this study, gold occurrence) with respect to various classes within a chosen The expression in equation (4) stretches the values of each evidential evidential layer (Khosravi et al., 2016). In this study, this was carried out layer to a range of 1–10. min TSi and max TSi represent respectively the by overlaying the training data on each of the eight evidential layers, minimum and maximum values of the total weighting index. Thus, in the which have been discretised into their respective classes. Thus, for each generation of the mineral prospectivity model based on the weighting class of an evidential layer, SI-based weight values obtained are the factor method, each of the evidential layers used was multiplied by its natural logarithm of the intensity of occurrence of gold mineral within corresponding weighting factor value; after which all the weighted that class divided by the total mineral intensity of the study area. The SI evidential layers were integrated in a GIS environment as expressed in method and its implementation are summarised in the expression shown equations (5) and (6). in equation (1), as introduced by Van Westen (1997). ∑n [ ] [ / ] MPMWF = EVnorm,j ×WFi (5) M Au Au WSI = ln ij ij T = ln / (1) j=1 M Aij AT EVi − EVmin From equation (1), WSI denotes the SI-based weight that is computed EVnorm,j = (6) EV − EV for each class i of and evidential layer j; Mij depicts the intensity of max min mineral occurrence within a chosen class i of an evidential layer j; M where MPMWF is the mineral prospectivity model generated based on represents the total intensity of mineral occurrence within the evidential the weighting factor model. EVnorm,j denotes a normalised evidential layer. Auij captures the number of known gold (Au) occurrences layer, EVmin and EVmax represent the minimum and maximum values (training data) within class i of a given evidential layer j; AuT comprises within a particular evidential layer, EVi. the total number of known Au occurrences within the training data; Aij is the areal size of class i of a chosen evidential layer j and AT represents the total areal size of the evidential layer j. 3.4. Validation of the mineral prospectivity models For each class i of an evidential layer j, a positive WSI value attained is an indication that the known locations of Au mineral and the afore- An important task in carrying out mineral prospectivity modelling mentioned class i are properly and robustly associated. Negative values over this study area was to validate the predicted model outputs. In the of a computed WSI means there is a poor or weak relationship between geoscientific sense, the geospatial models produced are not of much locations of known Au occurrences and class i of a given evidential layer essence and are without any meaningful relevance unless they are j. No relationship or correlation is established between class i of a given validated (Bourenane et al., 2016). In this study, the validation process evidential layer and the known location of Au occurrence if no training was carried out based on the application of the receiver operating data is found in the said class i. When generating a mineral prospectivity characteristics (ROC) curve. During the use of the ROC technique, the model based on the statistical index method, each class within a given known location of Au occurrence (validation data) was matched with evidential layer is assigned their respective WSI value. Afterwards, all the the mineral prospectivity models produced. This results in the deter- evidential layers are synthesised to generate the SI-based MPM as shown mination of the distribution of Au validation data in comparison with in equation (2). the value classes of the generated mineral prospectivity models to ascertain the percentage of gold occurrences that align with the highly ∑n MPM W (2) prospective zones. Values obtained from the employability of the ROC SI = SI,ij j=1 curve technique range from 0.5 to 1. For a particular model, a computed ROC value that is closer to 1 is an indication that the model produced has Where MPMSI denote the mineral prospectivity map generated based on good and higher accuracy. 6 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 4. Discretised map of the potassium concentration layer showing the spatial distribution of potassium within the Collette PL Area. Fig. 5. Discretised map of the potassium-thorium ratio layer over the Collette PL Area. 4. Results and discussion carried out by employing eight geospatial evidential layers sourced from radiometric (layers derived comprised potassium concentration, ura- 4.1. Characteristics of geoscientific evidential layers generated nium concentration, potassium-thorium ratio, and uranium-thorium ratio layers), magnetic (layers extracted consist of the analytic signal, The creation of predictive models over the study area using the lineament density, and reduction-to-equator evidential layers), and weighting factor (WF) and the statistical index (SI) techniques to geological (from which the geology layer was derived) datasets. In delineate prospective zones viable for the mineral occurrence was mineral prospecting, the use of evidential layers sourced from 7 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 6. Discretised map of the uranium concentration layer showing the spatial distribution of uranium over the Collette PL Area. Fig. 7. Discretised map of the uranium-thorium ratio layer over the Collette PL Area. radiometric data is very useful because of their ability to delineate respect to distance from a mineralisation region (Dentith and Mudge, alteration zones with enormous relevance to mineralisation over a 2014; Forson et al., 2021). Also, potassium-thorium antagonism, which designated area. Furthermore, in hydrothermal alteration zones, an arises due to the enhancement of potassium and depletion of thorium enhancement or depletion is observed in practically every integration of over a specified area, is an indication of a possible occurrence of min- potassium, thorium, and uranium (the three main radiometric ele- eralisation. Regions with increased uranium concentrations point to the ments). It is noteworthy that potassium enrichment is observed in sit- possible occurrence of mineralisation. In terms of the uranium-thorium uations of intense hydrothermally altered zones, and thus it reduces with ratio, areas observed to be highly anomalous suggest the possible 8 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 8. Discretised map of the lineament density layer showing the intensity of occurrence of lineaments over the Collette PL Area. Fig. 9. Discretised map of the RTE-based magnetic intensity layer which depicts the distribution of magnetic intensity responses over the Collette PL Area. occurrence of mineralisation and are deemed worthy target zones density layer (shown in Fig. 8) was sourced from the use of the CET (Dentith and Mudge, 2014). Thus, for the four radiometrically-sourced (Centre for Exploration Targeting) grid analysis technique on the mag- layers, only the potassium concentration layer (shown in Fig. 4) was netic data over the study area. This layer outlines the intensity of discretised into six classes, with the rest (potassium-thorium ratio, geological structure occurrences within various regions over a specified uranium, and uranium-thorium ratio) discretised into classes of five as area and thus depicts the extent of the endowment of various fracture shown in Figs. 5–7, respectively, using the Jenks natural breaking zones within a geologic environment. Lineament density is essential in classification technique (Jenks, 1963; Chen et al., 2013). The lineament mineral prospecting because it characterises low-pressure zones, which 9 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 10. Discretised map of the analytic signal layer depict the spatial distribution of magnetic intensity gradient over the Collette Area. Fig. 11. Map of showing geology layer showing various lithological classes characterising the Collette PL Area. act as conduits or points of convergence for hydrothermal fluids (Forson study area was vital. In low magnetic latitude regions such as the et al., 2020, 2021). Gold mineralisation occurrences within the Collette north-western part of Ghana, within which this study was undertaken, PL Area and north-western Ghana are generally related to quartz veins the magnetic intensity responses observed are generally not character- (made up of various structural features such as faults, dykes, etc.) istic of subsurface geology due to asymmetric problems that are asso- (Amponsah et al., 2016). Thus, the lineament density layer (shown in ciated with magnetic data acquired over low magnetic latitudes (Forson Fig. 8) was classified into five classes. The spatial distribution of mag- et al., 2021). Hence, the magnetic intensity data was reduced-to-equator netic responses is very essential towards delineating prospective zones (RTE) to make the observed magnetic intensity responses characteristic of mineral occurrence because mineral occurrence within the study area of subsurface geology. This resulted in the generation of the is strongly related to indicator minerals with a magnetic character such reduction-to-equator evidential layer shown in Fig. 9. This task carried as arsenopyrite, pyrite, and magnetite. Hence, the inclusion of evidential out by the RTE was analogously carried out by the analytic signal (AS) layers that outline the distribution of magnetic responses within the filtering technique to generate geologically-characteristic magnetic 10 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Table 1 Statistical Index and Weighting Factor scores of the Geoscientific evidential layers used. EVs Class Total Number of Pixels Au occurrence pixels SI WF Geological Layer G4 490,039 4 − 1.7593 1 tarkw 771,629 84 0.8312 vs 2,015,277 72 − 0.2829 bas 6,447,210 304 − 0.0055 comp-gn 356,547 20 0.1682 G2 549,870 20 − 0.265 Analytic Signal Layer (nT m− 1) 0.0021–73.7297 7,927,794 428 1.1387 8.66 73.7297–242.2501 2,199,613 68 0.6521 242.2501–621.4210 421,134 4 0.2003 621.4210–2.685.7956 82,031 4 1.0285 RTE Layer (nT) 11,454–17,368 3,084,187 88 − 0.5078 2.47 17.368–20.545 1,710,402 44 − 0.6114 20.545–24.075 1,561,787 32 − 0.8389 24.075–26.988 2,605,859 296 0.8738 26,988–33,872 1,668,337 44 − 0.5865 Lineament Density Layer (km− 2) 0–17 4,109,889 192 0.9854 8.68 17–44 2,500,998 152 1.2819 44–72 2,220,323 116 1.102 72–108 1,325,318 40 0.6366 108–255 474,044 4 1.178 Potassium Concentration Layer (%) 16.15–18.89 1,730,023 40 0.4877 10 18.89–22.61 144,889 0 0 22.61–25.98 299,118 0 0 25.98–28.51 2,802,809 88 0.6622 28.51–30.41 2,635,542 140 1.1204 30.41–34.06 3,018,191 236 1.6493 Potasium-Thorium Ratio Layer (%/ppm) 0.49–0.74 1,033,728 36 0.7346 8.38 0.74–0.95 2,687,025 156 1.2246 0.95–1.10 5,533,838 264 1.0062 1.10–1.48 1,234,104 44 0.752 1.48–2.00 141,877 4 0.5947 Uranium Concentration Layer (ppm) 15.26–20.62 894,701 16 0.3772 8.82 20.62–25.89 839,547 40 1.0049 25.89–28.94 2,355,978 76 0.6804 28.94–31.00 2,572,949 132 1.0821 31.00–36.29 3,967,397 240 1.2759 Uranium–Thorium Ratio Layer 0.46–0.81 896,673 48 1.1291 8.76 0.81–1.01 4,687,297 268 1.206 1.01–1.20 3,697,917 164 0.9354 1.20–1.54 964,729 24 0.5247 1.54–2.25 383,956 0 0 intensity gradients over the study area (captured on the analytic signal experienced intense E-W shearing, mylonitization with steep dips, and layer in Fig. 10). Although the analytic signal and the isoclinal folds with their axial planes parallel to the shear foliation as a reduction-to-equator correct and mitigate the asymmetric effects in low result of thrust-related deformation (Amponsah et al., 2015). These magnetic latitude regions, the former does not take into account the shear zones have permitted the plumbing of hydrothermal fluids, direction of magnetisation. The RTE and AS evidential layers were dis- creating wall rock alteration haloes. The widespread alteration within cretised respectively, into classes of four and five for the implementation volcanosedimentary rocks consists of graphitization, silicification of the bivariate data-driven methods. Geologically, the Collette PL Area (quartz veining), sulphidation (mainly arsenopyrites and pyrites), and is generally characterised by six lithological classes consisting of bas that of the comp-gn is associated with potassic alteration that accom- (basalts), comp-gn (composite gneisses), G2 (granodiorite with diorite), panied the gold mineralisation. The intense zone of shearing is G4 (granite and granodiorite), tarkw (Tarkwaian sediments), and vs approximately 150 m in width. Also, the Tarkwaian sediments (2120 (volcanosedimentary rocks) as shown in Fig. 11. Ma) of quartzites and matured diamictite conglomerates are molassic sediments of the Comp-gn (granite and gneisses) and the volcaniclastic 4.2. Mineral prospectivity models rocks and therefore may carry free gold around the valves of the pebble clast in the conglomerates (Nunoo et al., 2022). For this reason, Tarkw 4.2.1. Statistical index-based mineral prospectivity model and comp-gn have a correlation with gold and a positive statistical By employing the statistical index model, the scores obtained for value. G2 and G4 are late intrusions which usually occur as rounded each class of various geoscientific layers are shown in Table 1. Based on moulds and intrude the early comp-gn and tarkw sediments with no the results in Table 1, four geological layer classes consisting of bas, G2, association with gold mineralisation or sulphidation. The relevance of G4, and vs were observed to have their computed SI scores being, the two lithological classes with class names tarkw and comp-gn were respectively, − 0.0055, − 0.2650, − 1.7593, and − 0.2829. The negative deemed to be strong towards gold mineralisation occurrence within the scores obtained for the aforementioned lithological classes indicate a study area due to the positive SI score obtained (which are respectively weak correlation with respect to the known gold occurrences within the 0.8312 and 0.1682). In the case of the four-classified analytic signal study area. The geology of the Collette PL Area is dominated by comp-gn layer, statistical index scores obtained for all the classes with range of − 1 − 1 − 1 (composite granite and gneissess; 2196 - 2193 Ma) with a klipper of values 0.0021 nT m - 73.7297 nT m , 242.2501 nT m - 621.4210 − 1 − 1 volcaniclastic rocks (mainly composed of siltstone and shale facies; nT m , 73.7297 nT m - 242.2501 nT m − 1 and 621.4210 nT m− 1 - − 1 2139 Ma), G2 (tonalite and granitoid), and G4 (granitoids; 2111 ± 7 Ma) 2.685.7956 nT m were respectively 1.1387, 0.6521, 0.2003 and rocks. The Comp-gn with klipper of volcanosedimentary rocks has 1.0285. The SI scores obtained for the classes within the analytic signal 11 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 12. Statistical index-based mineral prospectivity model. layer mean that each of these classes has a strong correlation with respect to the known gold occurrences within the study area. In the case Table 2 of the reduction-to-equator evidential layer, four out of the total five Area extent and percentage of mineral prospectivity classes. classes with a range of magnetic intensity values of 11,454 nT - 17,368 MPM Statistical Index Weighting Factor nT, 17.368 nT–20.545 nT, 20.545 nT–24.075 nT, and 26,988 nT - 33, Class Area of class Percentage Area of class Percentage 872 nT were observed to show very weakened relevance towards the (km2) (%) (km2) (%) known gold occurrences within the study area with statistical index Very Low 2.91 6.85 2.61 6.14 scores of − 0.5078, − 0.6114, − 0.8389, and − 0.5865 respectively. The Low 8.72 20.53 10.23 24.08 RTE layer class with a magnetic intensity value of 24.075–26.988 nT Moderate 13.72 32.3 11.2 26.37 showed a strong correlation with respect to the gold occurrences within Moderately 13.3 26.6 14.9 35.08 High the study area with a positive statistical index score of 0.8738. In the High 5.83 13.72 3.54 8.33 case of the lineament density layer, all five classes show a positive correlation to gold mineralisation occurrences within the study area because the statistical index values computed were all positive. It is, respect to the known gold occurrences. This trend of uranium concen- however, worth noting that the class with a lineament density range of tration values also corroborates the assertion that regions with high 17–44 km km− 2 was adjudged to be the class with the highest correla- uranium concentrations suggest a highly possible occurrence of miner- tion to mineral occurrence within the study area with an SI score of alisation (Dentith and Mudge, 2014). With the exception of the 1.2819. In the case of the potassium concentration layer, two classes uranium-thorium class with a range of values of 1.54–2.25 which with concentrations of 18.89–22.61% and 22.61–25.98% were statisti- showed no correlation with respect to gold occurrences within the study cally analysed to show no correlation with the known gold occurrences area, the other four classes were observed to exhibit a positive correla- within the study area, owing to the zero SI scores obtained. The other tion to gold mineralisation occurrences within the study area with potassium concentration classes were found to exhibit a positive corre- positive SI values. The SI scores obtained were assigned to their lation with respect to gold occurrences. It can further be observed that respective classes and synthesised to generate the statistical index-based potassium classes with high potassium concentration values showed a mineral prospectivity model (SI-based MPM). The aforementioned higher correlation to gold mineralisation within the study area and vice model (shown in Fig. 12) characterises five classes of mineral pro- versa. This observation corroborates the literature assertion that po- spectivity zones within the study area. It can be observed that the very tassium enrichment is evidence of intense hydrothermal alteration low, low, moderate, moderately high, and high prospective zones occurrence and an indication of a highly probable occurrence of min- covered an area of 2.91 km2, 8.72 km2, 13.72 km2, 11.30 km2, and 5.83 eralisation (Dentith and Mudge, 2014; Forson et al., 2021). For the km2 respectively, as shown in Table 2. potassium-thorium ratio layer, all five classes show a positive correla- tion to gold occurrence due to the positive SI values attained. For the 4.2.2. Weighting factor-based mineral prospectivity model uranium concentration layer, though all the five classes were observed Whereas the statistical index technique analysed and determined the to show a positive correlation to gold mineralisation occurrences within spatial correlation of each class within a given evidential layer towards the study owing to the positive SI value computed, the classes with the gold occurrences in the generation of a mineral prospectivity model, higher uranium concentration values (28.94–31.00 ppm and the weighting factor technique was implemented to determine the in- 31.00–36.29 ppm) were observed to have higher SI scores (1.0821 and fluence of each of the evidential layers employed towards the generation 1.2759 respectively); an indication of a very strong correlation with of mineral prospectivity models over the study area based on the known 12 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Fig. 13. Weighting factor-based mineral prospectivity model. mineral occurrences used. The generation of weight factor scores for summarises the performance of a predictive model based on the area each of the eight geoscientific evidential layers was carried out based on under the ROC curve (AUC), which indicates the efficacy of a con- the expression in equation (4). The weighting factor results obtained for structed model for predicting a particular natural resource potential or the evidential layers ranged from 1 to 10; an evidential layer with a WF geohazard susceptibility over an area of interest. AUC scores greater score of 1 is the layer with the least contribution towards the predictive than 0.7 are considered efficient (Forson et al., 2022a). In this study, model to be generated, whereas an evidential layer with a WF score of AUC based on ROC curves produced for the SI-based MPM (Fig. 14a) and 10 has the highest influence on the mineral prospectivity model pro- WF-based MPM (Fig. 14b) using the training datasets of gold occurrence duced. Weighting factor scores obtained for the evidential layers in this within the study area were found to be respectively 0.780 and 0.733. study indicate that the potassium concentration layer and the litholog- The AUC scores obtained indicate that the use of a statistical index for ical layer, respectively, have the highest and the least influence on the producing MPM provides a more accurate prediction in comparison with mineral prospectivity model produced based on the weighting factor the weighting factor. technique (WF-based MPM). The WF scores obtained for the other six layers comprising reduction-to-equator, potassium-thorium ratio, ana- 5. Conclusion lytic signal, lineament density, uranium-thorium ratio, and uranium were respectively 2.47, 8.38, 8.66, 8.68, 8.76, and 8.82. The WF-based In the delineation of prospective zones of natural resource occur- MPM (shown in Fig. 13) was produced by multiplying each evidential rence, various methods have been applied worldwide to synthesise layer by their respective WF scores and synthesising them. The resulting various geospatial evidential layers in a data-driven manner. Popular mineral prospectivity model in Fig. 13 has been discretised into five among these data-driven methods are the statistical index and the distinct classes of prospectivity using the Jenks natural breaking clas- weighting factor approaches. This study employed the statistical index sification technique. From the output of the WF-based MPM, an area of and weighting factor approaches to prepare mineral prospectivity 2.61 km2, 10.23 km2, 11.20 km2, 14.90 km2, and 3.54 km2,respectively, models over the Collette PL Area of the north-western part of Ghana was delineated to characterise the very low, low, moderate, moderately using eight evidential layers sourced from magnetic, radiometric, and high, and high prospective zones of gold mineralisation within the study geological datasets. Furthermore, the predictive models produced based area (shown in Table 2). on these data-driven methods were compared to each other in this study. In the case of the statistical index technique, MPM was produced by assessing the coherence of each class within each of the eight evidential 4.3. Validation of the mineral prospectivity models produced layers used with respect to the known location of gold mineralisation occurrences within the study area, based on the SI values obtained. For In mineral prospectivity modelling, an important task that ought not the weighting factor method, the MPM produced was preceded by the to be overlooked is the validation of predicted results. It is noteworthy generation of individual weights for each of the eight evidential layers that predictive models are less useful and lack any meaningful scientific used by incorporating the known location of gold mineral occurrences relevance if they are not validated (Chung and Fabbri, 2003). In within the study area. Results obtained for the SI-based MPM indicate assessing the performance of a predictive model, the use of the receiver that 6.85%, 20.53%, 32.30%, 26.60%, and 13.72% of the total area size operating characteristics (ROC) curve has proven very useful and of the study area were delineated as prospectively very low, low, mod- worthy of being applied. During the application of the ROC curve vali- erate, moderately high, and high, respectively. For the weight factor- dation technique, known locations of gold occurrence data are based MPM, the very low, low, moderate, moderately high, and high compared with the MPM produced. Thus, the ROC curve quantitatively 13 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Acknowledgements The authors would like to thank Azumah Resources Limited for making the data available for this study. The authors are grateful to the University of Ghana-Carnegie Corporation and Building a New Gener- ation Africa (BaNGA-Africa) for their immense support in making this study a success. References Abdelkareem, M., Al-Arifi, N., 2021. Synergy of remote sensing data for exploring hydrothermal mineral resources using GIS-based fuzzy logic approach. Rem. Sens. 13 (22), 4492. Agra, N.A., Elburg, M.A., Vorster, C., 2023. Constraints on paleoproterozoic crustal growth from Birimian supergroup lavas of the bui belt (Ghana) in the west african craton. Precambrian Res. 384, 106926. Agyei-Duodu, J., 2009. Geological Map of Ghana 1: 1 000 000. Geological Survey Department. Akinlalu, A., Olayanju, G., Adiat, K., Omosuyi, G., 2021. Mineralisation potential assessment using analytical hierarchy process (AHP) modeling technique: a case study of ilesha schist belt, southwestern Nigeria. Results in Geophysical Sciences 7, 100026. Al-Abadi, A.M., 2017. Modeling of groundwater productivity in northeastern Wasit governorate, Iraq using frequency ratio and shannon entropy models. Appl. Water Sci. 7 (2), 699–716. Amponsah, P.O., Forson, E.D., Sungzie, P.S., Loh, Y.S.A., 2022a. Groundwater prospectivity modeling over the akatsi districts in the volta region of ghana using the frequency ratio technique. Modeling Earth Systems and Environment 1–19. Amponsah, P.O., Salvi, S., Béziat, D., Siebenaller, L., Baratoux, L., Jessell, M.W., 2015. Geology and geochemistry of the shear-hosted julie gold deposit, nw Ghana. J. Afr. Earth Sci. 112, 505–523. Amponsah, P.O., Salvi, S., Didier, B., Baratoux, L., Siebenaller, L., Jessell, M., Nude, P.M., Gyawu, E.A., 2016. Multistage gold mineralization in the wa-lawra greenstone belt, nw Ghana: the bepkong deposit. J. Afr. Earth Sci. 120, 220–237. Amponsah, T.Y., Danuor, S.K., Wemegah, D.D., Forson, E.D., 2022b. Groundwater potential characterisation over the voltaian basin using geophysical, geological, hydrological and topographical datasets. J. Afr. Earth Sci. 192, 104558. Amponsah, P.O., Kwayisi, D., Awunyo, E.K., Sapah, M.S., Sakyi, P.A., Su, B.X., Lu, Y., Nude, P.M., 2023. New evidence for crustal reworking and juvenile arc-magmatism during the Palaeoproterozoic Eburnean events in the Suhum Basin, South-east Ghana. Geol. J. In Press. Azumah Resources Limited, 2018. The Julie Mineral Resource Estimate. Unpublished internal report. Barati, M., Ostadhosseini, A., Rasa, I., Yazdi, M., 2018. Determination of cr geochemistry anomaly zones in the orzooiyeh area, hormozgan province using analytical hierarchy Fig. 14. Receiver operating characteristics (ROC) curve for (a) SI-based MPM process (AHP). Journal of Economic Geology 10 (1), 47–59. (b) WF-based MPM. Baratoux, L., Metelka, V., Naba, S., Jessell, M.W., Grégoire, M., Ganne, J., 2011. Juvenile paleoproterozoic crust evolution during the eburnean orogeny (2.2–2.0 ga), western Burkina Faso. Precambrian Res. 191 (1–2), 18–45. prospective classes covered 2.61%, 24.08%, 11.20%, 14.90%, and Block, S., Jessell, M., Aillères, L., Baratoux, L., Bruguier, O., Zeh, A., Bosch, D., Caby, R., 3.54% of the study area, respectively. By using the area under the Mensah, E., 2016. Lower crust exhumation during Paleoproterozoic (eburnean) receiver operating characteristics curve validation technique, the per- orogeny, NW Ghana, West African craton: interplay of coeval contractional deformation and extensional gravitational collapse. Precambrian Res. 274, 82–109. formance of the SI-based MPM and WF-based MPM produced were Bourenane, H., Guettouche, M.S., Bouhadad, Y., Braham, M., 2016. Landslide hazard accordingly assessed to have AUC scores of, 0.780 and 0.733 respec- mapping in the Constantine city, northeast Algeria using frequency ratio, weighting tively. The AUC scores obtained (greater than 0.7) indicate that the SI factor, logistic regression, weights of evidence, and analytical hierarchy process methods. Arabian J. Geosci. 9 (2), 1–24. and WF models are good predictors of mineral occurrence potential over Carranza, E.J.M., 2008. Geochemical Anomaly and Mineral Prospectivity Mapping in the study area. Based on the MPM outputs generated for the SI and WF, GIS. Elsevier. 19.13 km2 and 18.44 km2 were delineated as the highly probable zones Carranza, E.J.M., 2015. Data-driven evidential belief modeling of mineral potential using (representing moderately high and high classes) of mineral occurrence few prospects and evidence with missing values. Nat. Resour. Res. 24 (3), 291–304. Carranza, E.J.M., Laborte, A.G., 2015. Random forest predictive modeling of mineral over the Collette PL Area of north-western Ghana. It is also noteworthy prospectivity with small number of prospects and data with missing values in abra to acknowledge that, these aforesaid data-driven spatial statistical (Philippines). Comput. Geosci. 74, 60–70. techniques perform poorly or are non-functional in instances where Cevik, E., Topal, T., 2003. Gis-based landslide susceptibility mapping for a problematic segment of the natural gas pipeline, Hendek (Turkey). Environ. Geol. 44 (8), known locations of mineral occurrences are rare or non-existent in the 949–962. region to be studied. Chen, J., Yang, S., Li, H., Zhang, B., Lv, J., 2013. Research on geographical environment unit division based on the method of natural breaks (jenks). Int. Arch. Photogram. Rem. Sens. Spatial Inf. Sci. 3, 47–50. Declaration of competing interest Chung, C.-J.F., Fabbri, A.G., 2003. Validation of spatial prediction models for landslide hazard mapping. Nat. Hazards 30 (3), 451–472. The authors declare that they have no known competing financial Dentith, M., Mudge, S.T., 2014. Geophysics for the Mineral Exploration Geoscientist. Cambridge University Press. interests or personal relationships that could have appeared to influence Diatta, F., Ndiaye, P.M., Diène, M., Amponsah, P.O., Ganne, J., 2017. The structural the work reported in this paper. evolution of the dialé-daléma basin, kédougou-kéniéba inlier, Eastern Senegal. J. Afr. Earth Sci. 129, 923–933. Dickson, K., Benneh, G., 1988. A New Geography of ghana. Longman group uk limited. Data availability Eglinger, A., Thébaud, N., Zeh, A., Davis, J., Miller, J., Parra-Avila, L.A., Loucks, R., McCuaig, C., Belousova, E., 2017. New insights into the crustal growth of the Data will be made available on request. paleoproterozoic margin of the Archean kéména-man domain, West African craton (Guinea): implications for gold mineral system. Precambrian Res. 292, 258–289. 14 P.O. Amponsah and E.D. Forson J o u r n a l o f A f r i c a n E a r t h S c i e n c e s 2 06 (2023) 105020 Esmaeiloghli, S., Tabatabaei, S.H., Carranza, E.J.M., Hosseini, S., Deville, Y., 2021. mapping, a case study in sarvian area, markazi province, Iran. Bulletin of Mineral Spatially-weighted factor analysis for extraction of source-oriented mineralization Research and Exploration 155 (155), 147–160. feature in 3d coordinates of surface geochemical signal. Nat. Resour. Res. 30 (6), Masurel, Q., Eglinger, A., Thébaud, N., Allibone, A., André-Mayer, A.-S., McFarlane, H., 3925–3953. Miller, J., Jessell, M., Aillères, L., Vanderhaeghe, O., et al., 2022. Paleoproterozoic Feng, X., Wang, E., Ganne, J., Amponsah, P., Martin, R., 2018. Role of volcano- gold events in the southern West African craton: review and synopsis. Miner. sedimentary basins in the formation of greenstone-granitoid belts in the West African Deposita 57 (4), 513–537. craton: a numerical model. Minerals 8 (2), 73. Mathew, T.G., Ariffin, K.S., 2018. Gold potential mapping in Kelantan (Malaysia) using Feng, X., Wang, E., Amponsah, P.O., Ganne, J., Martin, R., Jessell, M.W., 2019. Effect of ArcGIS and excel applying frequency ratio model. In: 2018 National Geoscience pre-existing faults on the distribution of lower crust exhumation under extension: Conference. numerical modelling and implications for NW Ghana. Geosci. J. 23 (6), 961–975. McKay, G., Harris, J., 2016. Comparison of the data-driven random forests model and a Ford, A., Miller, J.M., Mol, A.G., 2016. A comparative analysis of weights of evidence, knowledge-driven method for mineral prospectivity mapping: a case study for gold evidential belief functions, and fuzzy logic for mineral potential mapping using deposits around the huritz group and nueltin suite, nunavut, Canada. Nat. Resour. incomplete data at the scale of investigation. Nat. Resour. Res. 25 (1), 19–33. Res. 25 (2), 125–143. Forson, E.D., Amponsah, P.O., Hagan, G.B., Sapah, M.S., 2022a. Frequency ratio-based Meinhardt, M., Fink, M., Tünschel, H., 2015. Landslide susceptibility analysis in central flood vulnerability modeling over the greater accra region of ghana. Modeling Earth vietnam based on an incomplete landslide inventory: comparison of a new method to Systems and Environment 1–20. calculate weighting factors by means of bivariate statistics. Geomorphology 234, Forson, E.D., Menyeh, A., 2023. Best worst method-based mineral prospectivity 80–97. modeling over the central part of the southern Kibi-Winneba belt of Ghana. Earth Nunoo, S., Hofmann, A., Kramers, J., 2022. Geology, zircon u–pb dating and εhf data for Science Informatics 1–20. the julie greenstone belt and associated rocks in nw Ghana: implications for Forson, E.D., Menyeh, A., Wemegah, D.D., 2021. Mapping lithological units, structural birimian-to-tarkwaian correlation and crustal evolution. J. Afr. Earth Sci. 186, lineaments and alteration zones in the southern Kibi-winneba belt of Ghana using 104444. integrated geophysical and remote sensing datasets. Ore Geol. Rev. 137, 104271. Parsa, M., Carranza, E.J.M., 2021. Modulating the impacts of stochastic uncertainties Forson, E.D., Menyeh, A., Wemegah, D.D., Danuor, S.K., Adjovu, I., Appiah, I., 2020. linked to deposit locations in data-driven predictive mapping of mineral Mesothermal gold prospectivity mapping of the southern kibi-winneba belt of Ghana prospectivity. Nat. Resour. Res. 30 (5), 3081–3097. based on fuzzy analytical hierarchy process, concentration-area (ca) fractal model Rahimi, E., Shekarian, Y., Farahani, S., Asgari, G., Nakini, A., 2020. New approach in and prediction-area (pa) plot. J. Appl. Geophys. 174, 103971. application of the ahp–fuzzy topsis method in mineral potential mapping of the Forson, E.D., Wemegah, D.D., Hagan, G.B., Appiah, D., Addo-Wuver, F., Adjovu, I., natural bitumen (gilsonite): a case study from the gilan-e-gharb block, the Otchere, F.O., Mateso, S., Menyeh, A., Amponsah, T., 2022b. Data-driven multi- kermanshah, west of Iran. Am. J. Eng. Appl. Sci. 13 (1), 96–110. index overlay gold prospectivity mapping using geophysical and remote sensing Rocci, G., 1965. Essai d’interpretation de mesures geochronologiques. la structure de datasets. J. Afr. Earth Sci. 190, 104504. l’ouest africain. Sci. Terre 10 (3–4), 461–478. Fu, C., Chen, K., Yang, Q., Chen, J., Wang, J., Liu, J., Xiang, Y., Li, Y., Rajesh, H., 2021. Sanusi, S.O., Amigun, J.O., 2020. Logistic-based translation of orogenic gold forming Mapping gold mineral prospectivity based on weights of evidence method in processes into mappable exploration criteria for fuzzy logic mineral exploration southeast asmara, Eritrea. J. Afr. Earth Sci. 176, 104143. targeting in the kushaka schist belt, north-central Nigeria. Nat. Resour. Res. 29 (6), Ghasemzadeh, S., Maghsoudi, A., Yousefi, M., Mihalasky, M.J., 2022. Information value- 3505–3526. based geochemical anomaly modeling: a statistical index to generate enhanced Salvi, S., Amponsah, P.O., Siebenaller, L., Béziat, D., Baratoux, L., Jessell, M., 2016. geochemical signatures for mineral exploration targeting. Appl. Geochem. 136, Shear-related gold mineralization in Northwest Ghana: the Julie deposit. Ore Geol. 105177. Rev. 78, 712–717. Jenks, G.F., 1963. Generalization in statistical mapping. Ann. Assoc. Am. Geogr. 53 (1), Sapah, M.S., Agbetsoamedo, J.E., Amponsah, P.O., Dampare, S.B., Asiedu, D.K., 2021. 15–26. Neodymium isotope composition of palaeoproterozoic birimian shales from the Wa- Jessell, M.W., Amponsah, P.O., Baratoux, L., Asiedu, D.K., Loh, G.K., Ganne, J., 2012. Lawra belt, north-west Ghana: constraints on provenance. Geol. J. 56 (4), Crustal-scale transcurrent shearing in the paleoproterozoic sefwi-sunyani-comoe 2072–2081. region, west africa. Precambrian Res. 212, 155–168. Sun, T., Li, H., Wu, K., Chen, F., Zhu, Z., Hu, Z., 2020. Data-driven predictive modelling Kashani, S.B.M., Abedi, M., Norouzi, G.-H., 2016. Fuzzy logic mineral potential mapping of mineral prospectivity using machine learning and deep learning methods: a case for copper exploration using multi-disciplinary geo-datasets, a case study in seridune study from southern jiangxi province, China. Minerals 10 (2), 102. deposit, Iran. Earth Science Informatics 9 (2), 167–181. Van Westen, C., 1997. Statistical landslide hazard analysis. Ilwis 2, 73–84. Khosravi, K., Pourghasemi, H.R., Chapi, K., Bahri, M., 2016. Flash flood susceptibility Yousefi, M., Carranza, E.J.M., 2015. Prediction–area (p–a) plot and c–a fractal analysis to analysis and its mapping using different bivariate models in Iran: a comparison classify and evaluate evidential maps for mineral prospectivity modeling. Comput. between shannon’s entropy, statistical index, and weighting factor models. Environ. Geosci. 79, 69–81. Monit. Assess. 188 (12), 1–21. Zhang, P., Zhang, Z., Yang, J., Cheng, Q., 2022. Machine learning prediction of ore Khosravi, V., Shirazi, A., Shirazy, A., Hezarkhani, A., Pour, A.B., 2021. Hybrid fuzzy- deposit genetic type using magnetite geochemistry. Nat. Resour. Res. 1–18. analytic hierarchy process (ahp) model for porphyry copper prospecting in simorgh Zhang, S., Carranza, E.J.M., Wei, H., Xiao, K., Yang, F., Xiang, J., Zhang, S., Xu, Y., 2021. area, eastern lut block of Iran. Mining 2 (1), 1–12. Data-driven mineral prospectivity mapping by joint application of unsupervised Kusuma, K., Chaitanya, S., Guru, B., et al., 2019. Frequency ratio modelling using convolutional auto-encoder network and supervised convolutional neural network. geospatial data to predict kimberlite clan of rock emplacement zones in dharwar Nat. Resour. Res. 30 (2), 1011–1031. craton, India. Int. J. Appl. Earth Obs. Geoinf. 74, 191–208. Zhang, Z., Zuo, R., Xiong, Y., 2016. A comparative study of fuzzy weights of evidence and Lin, N., Chen, Y., Liu, H., Liu, H., 2021. A comparative study of machine learning models random forests for mapping mineral prospectivity for skarn-type fe deposits in the with hyperparameter optimization algorithm for mapping mineral prospectivity. southwestern fujian metallogenic belt, China. Sci. China Earth Sci. 59 (3), 556–572. Minerals 11 (2), 159. Zhenjie, Z., Qiuming, C., Jie, Y., Guopeng, W., Yunzhao, G., 2021. Machine learning for Mansouri, E., Feizi, F., Rad, A.J., Arian, M., 2017. A comparative analysis of index mineral prospectivity: a case study of iron-polymetallic mineral prospectivity in overlay and topsis (based on ahp weight) for iron skarn mineral prospectivity southwestern fujian. Earth Sci. Front. 28 (3), 221. 15