A Geo-Data Science Method for Assessing Unconventional Rare-Earth Element Resources in Sedimentary Systems

Abstract

Rare-earth elements (REEs) supply raw materials that constitute many of our modern critical infrastructure, defense, technology, and electrification needs. Despite REE accumulations occurring in conventional bedrock and ion-adsorption deposits sourced from weathering of igneous rocks, unconventional host materials such as coal and related sedimentary strata have been identified as promising sources of REEs to meet growing demand. To maximize the potential of unconventional resources such as REE-coal systems, new approaches are needed overcome challenges from mineral systems with no known deposits and areas with sparse geochemical data. This article presents a systematic knowledge-data resource assessment method for predicting and identifying REE resource potential and occurrence in these unconventional systems. The method utilizes a geologic and geospatial knowledge-data approach informed and guided by REE accumulation mechanisms to systematically assess and identify areas of higher enrichment. An assessment of the Powder River Basin is presented as a test case to demonstrate the method workflow and results. The key output is a potential enrichment score map reported with varying confidence levels based on the amount of supporting evidence. Results from the test case indicate several locations with promising potential for different types of coal-REE deposits, demonstrating the viability of the method for exploration and assessment of unconventional REE resources. The method is flexible by design and, with sufficient applicable knowledge and data, can be adapted for assessing critical mineral systems in other sedimentary systems as well.

Introduction

Rare-earth elements (REEs) comprise the 17 lanthanide series elements, Scandium (Sc), and Yttrium (Y). These elements serve as critical raw materials necessary for growing demand for modern technologies. The supply of REEs has historically been sourced from bedrock and regolith-hosted ion-adsorption deposits (IADs) originating from the weathering of igneous rocks (Borst et al., 2020). Recent studies have indicated that coaliferous sediments provide promising sources for potential domestic REE accumulations (Seredin & Dai, 2012; Hower et al., 2016a) and other critical minerals (DOE, 2017). However, uncertainty exists in understanding the volumetric estimates and spatial extent of these REE deposits and resources. In particular, little is known about concentrations, mineral phase, and occurrence of REEs by type—e.g., light (La, Ce, Pr, Nd, Sm; LREEs) and heavy (Ho, Er, Tm, Yb, Lu; HREEs), as defined by Seredin and Dai (2012). These factors are important as they ultimately determine the economic viability and salability of a deposit and its ability to provide a reliable source.

Chemical analysis of samples from coaliferous sediments at various US sites indicates promising REE concentrations in certain areas (Ekmann, 2012), yet these deposit types have not been considered conventional sources of REEs (Van Gosen et al., 2017). This may in part be attributed to the quality and sparseness of publicly available REE measurements in coaliferous sediments for much of the USA—and world—which hinders rigorous evaluation of the resource potential of these deposits. Given the systematic nature of geologic conditions and processes conducive to REE enrichment in sediments, as with other geologic resource systems (Magoon & Dow, 1994; Wyborn et al., 1994), deposit locations and character may be predictable (Creason et al., 2021). However, because of poorly quantified spatial resource trends, no known economically viable deposits, and the lack of an exploration method tuned to REEs in coaliferous sediments, no resource assessment method exists at present to systematically identify potential occurrences. Therefore, there exists a need for an assessment method accounting for the variation in geological mechanisms of REE enrichment within sedimentary systems that also addresses geologic uncertainty and identifies data gaps to better inform exploration efforts. Such an approach would aid the reduction of commercial exploration risk and help quantify a domestic source of economic and sustainable REE development.

Here, we present a new approach for predicting, identifying, and, with sufficient data, quantifying potential Unconventional Rare-earth element and Critical mineral (URC) occurrences in sediments using a systematic knowledge-data resource assessment method. The approach leverages lessons learned from existing resource assessment methods and complementary analytical approaches, such as geologic resource systems concepts (e.g., Magoon & Dow, 1994; Wyborn et al., 1994) and mineral prospectivity mapping (e.g., Carranza, 2008, 2017) as well as the current state of knowledge of REE enrichment in coal and related strata (Creason et al., 2021). These preceding, foundational methods and techniques are reviewed and summarized to provide the basis formulation of the URC assessment method; however, the new approach differs from other data-driven and knowledge-driven assessment methods (cf. Carranza, 2017; and references therein) in that there are no known significant deposits in US sedimentary basins. Thus, the results from the URC assessment method, presented as a Potential Enrichment (PE) score and Confidence maps, are intended to prioritize and highlight the most prospective sites for future investigation and additional data collection for these unconventional resources. We demonstrate the method using a test case in the Powder River Basin (Wyoming and Montana, USA) and discuss how the results can inform future, targeted data collection efforts which can, in turn, yield more robust assessments.

Background

Types of REE Deposits

Despite their name, REEs occur in significant concentrations in a variety of geologic materials. At present, the main geologic sources of REE ore are from carbonatite and peralkaline intrusive rock complexes and their weathered by-products—termed weathered crust elution deposits (WCEDs) (Balaram, 2019). These types of deposits are found throughout the world; however, mining of economically viable ores is rare. Carbonatites contain the mineral phases bastnaesite and monazite, both generally having high REE + Y concentrations of 1–10% in ore. The Mountain Pass mine located in San Bernardino County, CA, was historically a world-leading producer of LREEs from carbonatite deposits (Olson et al., 1954). Mining activities ceased at Mountain Pass in 1998 due to competition from international sources, primarily China. Recently, carbonatite deposits have received renewed attention with the development of more cost-effective separations (enrichment and recovery) technologies (Weng et al., 2015; Anenburg et al., 2020) and management of waste. WCEDs with ion exchangeable REEs are sources for all REEs (LREEs, HREEs). Mining of WCEDs has been economically feasible in China due to the large-scale of the deposits, simplicity of REE extraction, and the economic benefit of recovering critical HREEs. However, environmental concerns continue to hinder carbonatite and WCED production as many deposits contain significant concentrations of uranium and thorium, further complicating waste stream management. In response to these challenges, unconventional sources of REEs, such as those in sedimentary settings, have garnered increased interest from the global community.

REEs in Sedimentary Systems

Given the number of sedimentary basins in the USA, REE deposits found within sedimentary strata could provide a secure supply (Coleman Jr & Cahan, 2012). REE occurrences in sedimentary systems are generally classified into two groups: primary- or secondary-type emplacement. Primary-type sedimentary REE emplacements involve concurrent, or syn-depositional, emplacement of an REE-rich source material and a host sediment, such as an ash fall within a peat bog or mineral detritus within a fluvial sand. Post-deposition, or secondary-type emplacement, depends on a series of events in succession. For example, REEs might be mobilized from a source, transported, and experience ideal geochemical conditions for enrichment of a certain horizon or zone. Heavy-mineral sands (HMS) occur in the USA and serve as the primary source of titanium (ilmenite), zircon, and monazite. HMS accumulate in two depositional environments: continental shelf or coastal plain and alluvial deposits. Coastal deposits constitute the vast majority of HMS globally (Van Gosen et al., 2014). In the USA, monazite HMS deposits are found in large concentrations in alluvial channels of North and South Carolina, beach deposits of Florida and Georgia, and stream deposits of southeast Idaho (Long et al., 2010; and references therein).

REEs in Coal and Coal-Bearing Strata

Coal and coal by-products have been stated to contain notable concentrations of rare earths and critical minerals (Seredin & Dai, 2012; Hower et al., 2016a; Lin et al., 2017; Montross et al., 2018). Hard coal (e.g., bituminous and anthracite) tends to have the highest REE content of all geologic material (Rudnick & Gao, 2003; Ketris & Yudovich, 2009). For example, global averages for cerium (Ce, used as a metric for LREE concentrations in global reserves) for soft coal (brown lignite) and hard (including bituminous to anthracite) are 22 and 23 ppm, respectively. Typical continental crust values for Ce are 64 ppm (Rudnick & Gao, 2003). Brown and hard coal fly ash, produced from the combustion of coal for power generation, contain nearly 6 × this amount of Ce (Ketris & Yudovich, 2009). Late Permian Heshan coals in China are found in carbonate successions and represent another lithofacies in which REE contents are enriched. The coals are low volatile bituminous coals containing 5–12% sulfur, of which 90% is organic sulfur (Shao et al., 2003; Dai et al., 2013). The coal strata are intercalated with marine carbonates and detrital terrigenous minerals such as quartz and albite. Work by Shao et al. (2003) determined that a variety of factors control the geochemical and mineralogical composition of coal preserved within marine carbonate sequences. The most prevalent factor in this case is multi-stage hydrothermal activity coupled with inputs of organic sulfur and elevated B, Mg, K, Sr, and Rb from seawater.

The relative distribution of REEs associated with organic and inorganic material in coal is dependent on the depositional environment and presence of molecules or mineral phases that can capture REEs (Given & Miller, 1987). Lin et al. (2017) showed that the REEs in coal are not only associated with inorganic components (e.g., detrital minerals), but also the organic matter, and the organic matter is relatively enriched in HREEs compared with the inorganic matter. Many coals are slightly enriched in HREEs relative to LREEs, compared to chondrite and shale (Eskenazy, 1999; Lin et al., 2017). Eskenazy (1999) hypothesized that fractionation of HREEs/LREEs may be linked to the complexation of HREEs with organic compounds that are more stable than those formed by LREEs. The authors demonstrated the ability for organic matter to adsorb up to 0.2 mEq/0.5 g of xylain, using organic molecules xylain and humic acids, and demonstrated the cation exchange mechanism in which Na, K, Ca, and Mg bound to –COOH and –OH groups were replaced by REE cations. However, there was no difference in HREEs/LREEs in ion exchange and therefore no evidence that fractionation of HREEs and LREEs are controlled by organic material in the sample. Experimental work by Eskenazy (1999) and Sonke & Salters (2006) confirmed that strong complexation of REEs occurred when humic substances were present. The coupled results indicate that humic substances play an important role in the fractionation of REEs in geologic systems.

Seredin and Dai (2012) proposed four distinct mechanisms for the accumulation of REEs in coal: (1) terrigenous: input by surface water; (2) tuffaceous: connected with falling and leaching of acid and alkaline volcanic ash; (3) infiltrated: meteoric (ground) water driven; and (4) hydrothermal: connected with ascending flows of thermal mineral water and deep reduced fluids.

Inter-seam volcanic ash, or tonstein, layers have been found in some coal seams around the world (Zhou et al., 2000; Spears, 2012; Dai et al., 2014; Rozelle et al., 2016). Dai et al. 2014 investigated the mineralogy and geochemistry of dacite magma derived tonstein layers in Chinese coal. The tonstein layers were dominated by kaolinite, mixed layer illite/smectite, and minor quartz. Hydrothermal leaching of the buried volcanic ash layer led to the removal of certain ions resulting in Nb/Ta, Zr/Hf, and U/Th ratios much lower in the tonsteins than in the adjacent coal beds or floor/roof rock. The Middle Pennsylvanian-Fire Clay coal is a source of REEs (Zhang et al., 2018; Hower et al., 2020) and is distributed across the Central and Southern Appalachian basin, predominately concentrated in the central and southern coal regions of WV, KY, and AL. Coals within Fire Clay seams in Kentucky have distinct REE and Sc concentrations that are related to the position of the lithotype within the coal seam (Hower et al., 2020). Geochemical relationships such as the ratio of light to heavy REEs are also related to the lithotype position within the coal seam. Hydrothermal alteration may also play a role in the enrichment of REEs in certain layers that are in proximity to the tonstein (Hower et al., 2016b; Hower et al., 2020). Fire Clay coal is enriched in REEs and other critical minerals via emplacement by volcanic ash fall into the coal mire. Studies have demonstrated these tonstein layers can be horizontally pervasive throughout coal seams and partings (Bohor & Triplehorn, 1993) and are the source of REEs in the rock partings and within the coal itself (Seredin and Dai, 2012; Hower et al., 2018). The predominant mechanism of REE enrichment in the Fire Clay is from tuffaceous/volcanic, infiltrational/leaching, and hydrothermal processes that began in the coal depositional environment during coal formation and continued through subsequent diagenetic (e.g., dissolution and weathering in the subsurface) events (Hower et al., 2016a).

Geologic Resource Assessment Methodologies

Geologic resource assessment methodologies are used for exploration and quantification of critical resources, such as precious metals or fossil hydrocarbons. Over the past several decades, systematic geologic resource assessment methodologies have proven crucial for predicting subsurface properties and evaluating in situ resources. For example, prior to 1970, the chance of drilling a successful hydrocarbon well was comparable to a coin flip (Cochener, 2010). Following the development of the petroleum system model in the 1980s and other technological breakthroughs, exploration became more predictable and experienced increasing success rates throughout the years to above 90% today (Cochener, 2010). Thus, an effective assessment methodology for resource exploration begins with an understanding of the geologic processes and systems that produce and preserve the target resource (e.g., Singer & Berger, 2000; Meneley et al., 2003; McCuaig et al., 2010; McCuaig & Hronsky, 2014b; Hagemann et al., 2016). This knowledge base provides a conceptual foundation or framework to identify the appropriate datasets and analytics to develop the methodology (McCuaig et al., 2010; Hagemann et al., 2016). Data and knowledge frameworks on these systems are then applied using analytical techniques to either determine the prospectivity of the commodity within an area and/or obtain volumetric ore estimates (e.g., Ford et al., 2019).

Building a knowledge framework for a given mineral or hydrocarbon resource involves a spatial and temporal understanding of the geologic processes or components that contribute to or are required for accumulation and preservation. This typically requires an understanding of the source, transport mechanism, and containment for a given resource, as well as the larger-scale geologic setting. For example, the petroleum system is a conceptual process model that considers the origin and evolution of the geologic system that produced—or at least could produce—an accumulation of hydrocarbons (Magoon, 1988). Another type of knowledge framework relies less on the interdependencies between these processes or components, but instead focuses on the spatial and temporal coincidence of the critical independent contextual factors for ore genesis (Hronsky and Kreuzer, 2019). Factors for this approach include favorable lithospheric architecture, fertility (favorable geochemical conditions; transient favorable geodynamics, i.e., tectonic/stress conditions over time), and ore preservation (of primary depositional zone) (McCuaig & Hronsky, 2014a). Development of a knowledge framework often depends on the availability of knowledge or data for the specific mineral resource.

Information beneficial for a regional evaluation of mineral resources includes: (1) well-defined key descriptive terms; (2) geologic map data, to identify genetically related geologic map units and determine the geologic history; (3) mineral occurrence data, to establish attributes and locations of known mineral deposits, and relate deposits to host rocks and tectonic origins; (4) geochemical and geophysical data, to locate chemical anomalies of pathfinders on the surface and interpret the subsurface structure, respectively; and (5) exploration history and mineral deposit models of known deposits in the region, to understand outcome of previous assessments and infer the potential of nearby undiscovered deposits (Lipin & Bawiec, 2000; Nokleberg et al., 2007). Geological estimates of the deposit, such as grade, tonnage, and spatial extent, can (and often do) introduce significant amounts of error and uncertainty in assessing mineral resources and ore reserves (Dominy, 2002). Yet, these types and quantities of data can inform analytical techniques to map and quantify ore estimates.

Starting in the mid-1970s, mineral resource assessments from the US Geological Survey (USGS) adhered to a three-part quantitative assessment methodology (Singer, 1993). This approach, now seen as the “traditional approach” for quantitative assessments of nonfuel mineral deposits, uses known mineral site locations and deposit characteristics in combination with associated geologic features to estimate known resource quantities using grade and tonnage models and probabilistic calculations to estimate the number of undiscovered mineral deposits (Singer, 1993). Since the development and acceptance of the three-part assessment method, however, the advent of Geographic Information Systems (GIS) (Bonham-Carter, 1994) engendered additional methods and analytical techniques for predicting unknown mineral deposits, referred to as mineral prospectivity mapping or mineral potential modeling (MPM) (e.g., Bonham-Carter, 1994; Porwal et al., 2003; Carranza, 2008; Porwal & Kreuzer, 2010; Yousefi & Nykänen, 2017; Hronsky & Kreuzer, 2019).

Application of GIS-based MPM involves the collection, analysis, and combination of multi-source and multi-scale geologic data to integrate known mineral occurrences with spatial evidence layers to map mineral potential or prospectivity (Carranza, 2017). The MPM approach relies on a conceptual model which describes theoretical relationships between geologic controls of mineralization to inform which datasets are analyzed to create predictor maps (Carranza, 2017). Integration of the predictor maps thus yields a single map of mineral potential that can be compared with and validated against known mineral deposits.

There are two main types of MPM: data-driven models and knowledge-driven models. If a relatively large number of known mineral occurrences and their relationships to explanatory datasets are well known within a study area, data-driven modeling techniques such as weights of evidence (Nykänen et al., 2008), Monte Carlo simulations (Lysytsyn, 2015), Naïve Bayes (Zaremotlagh & Hezarkhani, 2016), logistic regression (Xiong & Zuo, 2018), or neural networks (Wang et al., 2011; Zaremotlagh & Hezarkhani, 2017) may be appropriate. Data-driven techniques allow for the quantification of uncertainty due to statistical randomness. In contrast, knowledge-driven techniques rely on expert knowledge due to few or no known mineral occurrences within a study area (Porwal et al., 2003; Carranza, 2008; Abedi et al., 2013). Knowledge-driven methods successfully applied to delineate potential mineral targets are variations of fuzzy logic (Dag & Mert, 2008; Nykänen et al., 2008; Lysytsyn, 2015; Yousefi and Carranza, 2015; Abedi et al., 2017; Leväniemi et al., 2017) and various ranking and weighting techniques (Karl et al., 2016).

Despite significant progress in the development of data-driven and knowledge-driven MPM approaches, significant challenges remain, primarily with regard to input data, which have impeded more widespread adoption of MPM by industry (Hronsky & Kreuzer, 2019). Yousefi et al. (2019) and (2021) outline a pyramid model concept of an Exploration Information System procedure to overcome some of these challenges by reevaluating the approach to the data-knowledge-information-insight progression. The model consists of defining knowledge relationships for particular mineral deposit types in a mineral systems model, leveraging the relationships to extract key data features representing proxies of the targeted mineral system and integrating these spatial data proxies in a prospectivity analysis for exploration target modeling (Yousefi et al., 2019, 2021).

As data science capabilities in processing and analyzing big data continue to evolve, so do methodologies that can further improve predictions for poorly constrained resources through the integration with data science (Gibert et al., 2018a; Singleton & Arribas-Bel, 2021). Building upon the progress and successes in data science, the nascent field of geo-data science combines the geosciences and data science to overcome challenges of integrating and analyzing large, complex, multi-source, geospatial datasets (Gibert et al., 2018b; Zuo & Xiong, 2020). Methodologies used in geo-data science combine data analysis and processing with domain expertise, translating data into knowledge that can then inform decision-making or drive further analyses (Gibert et al., 2018a). Geo-data science leverages tools and techniques such as data statistics, machine learning (e.g., clustering analysis), and GIS to efficiently collect, integrate, analyze, and mine geoscience datasets, which in turn leads to knowledge discovery and insights about spatial decisions (Gibert et al., 2018b; Zuo & Xiong, 2020). One such approach, the Subsurface Trend Analysis (STA) method developed by Rose et al. (2020), is a systematic, science- and data-driven method that utilizes geologic systems knowledge to inform statistical analyses of subsurface properties and resources. The STA workflow integrates geologic data used in traditional assessment approaches (e.g., geologic maps, in situ measurements, core) with appropriate statistical methods (e.g., geostatistics, dimensional analysis) to characterize the subsurface property of interest. Of particular relevance for mineral exploration, geo-data science methods have also led to promising developments in geochemical mapping (Zuo et al., 2019; Zuo and Xiong, 2020) and mineral prospectivity mapping (Zuo, 2020). Collectively, geo-data science capabilities enable integration of various knowledge sources and expertise, along with data handling, visualization, analysis, and knowledge production (Gibert et al., 2018b), all of which are of value for geologic modeling and resource assessments.

Thus, considering the success of existing approaches for assessing geologic properties and resources, development of a systematic approach for predicting and quantifying the occurrence of REEs in coal and related strata requires, at a minimum: 1) knowledge and data associated with coal-REE enrichment mechanisms, and 2) synthesis and integration of analytical approaches from established resource assessment methods.

Methods: Development of the URC Knowledge-Data Framework

Conceptual Model

The URC assessment method knowledge-data framework includes an understanding of the systematic nature of mineral accumulation in sediments and leverages data related to the processes of REE accumulation and in situ rock properties. In this section, a knowledge-data framework is presented for applications of the URC assessment method in coal sedimentary systems. Although the model and supporting database presented herein align to coal-REE assessments, the method is highly adaptable and applicable for other sedimentary systems as well.

For purposes of convenience, the URC assessment method knowledge-data framework reclassifies the genetic types of coal-REE accumulation defined by Seredin & Dai (2012) into six distinct occurrence types on the basis of timing of REE enrichment relative to sediment deposition, REE transport mode, and resulting REE phase (Table 1). Conceptual deposit models developed for the various coal-REE occurrence types represent the knowledge of how and why REEs accumulate in sedimentary systems. Like the petroleum system (Magoon & Dow, 1994) or other mineral deposit conceptualizations (Cox & Singer, 1986), a simplified conceptual REE-coal deposit model was developed that incorporates unique combinations of REE sources and accumulation processes (Fig. 1). Additional geological factors such as depositional history, tectonic and diagenetic influences, and geochemical alteration are also represented in the models as individual geologic components. Since the coal-REE deposit models comprise the core of the knowledge-data framework, it is critical that the models accurately depict the various accumulation mechanisms. Accordingly, the components and relationships defined in the deposit models are the culmination of extensive literature review on the current state of knowledge of the coal-REE system (Creason et al., 2021). Included in the Supplementary Information are detailed knowledge-data frameworks for each deposit type in Table 1, illustrated as unique REE enrichment process flowcharts (Online Resource 1 in Supplementary Information). As the scientific understanding of REE occurrences in coal sedimentary systems continues to evolve and improve, these deposit models will be refined, and additional deposit models may be developed.

Table 1 Classification of various REE-coal deposit types as considered by the URC assessment method. Adopted and modified from Seredin and Dai (2012)

Figure 1

Simplified flowchart of pathways for primary- and secondary-type occurrences of REEs in coal-bearing sedimentary systems. Each unique pathway (arrow path) involves an REE source and an accumulation process that can result in a coal-REE deposit. Boxes are a simplification of the geologic components evaluated by the URC assessment method. The chronological, comprehensive framework for each mechanism is presented in the Supplementary Information (Online Resource 1)

Having developed coal-REE deposit models, the key geologic components needed for REE accumulation can then be assessed per the model using a data science approach, such that associated data can be queried to test for evidence that a process or component occurred in the geologic past. Formally defining the relationships between the datasets and process of sedimentary REE accumulation is vital to a successful assessment at the basin, mine, or smaller scales.

Data Collection and Acquisition

During formulation of the URC assessment method, a collection of publicly available resources was compiled to support testing and development. This data collection, named the REE Coal Open Geo-database (Justman et al., 2018), comprises both spatial and non-spatial geological, geochemical, and geospatial datasets spanning across the US Key data include surface lithology, mineralogy, structural geology, geochemical analytical samples, and sediment provenance, from sources such as the USGS, US DOE, state geologic surveys, and other government agencies. The collection is available for download via US DOE NETL’s Energy Data eXchange (Justman et al., 2018).

The abundance of available data varies significantly across the USA (Fig. 2). Visualizing where data are available prior to implementing an assessment can help to identify suitable locations to apply the URC assessment method. However, given the proprietary nature of most local or mine scale coal data, the REE Coal Open Geo-database is mostly applicable for basin scale analyses. Note the relative abundance of data in the Powder River Basin (PRB) is largely due to the latest USGS coal resource assessment (Kinney et al., 2015; Luppens et al., 2015).

Figure 2

Count of overlapping datasets within the REE Coal Open Geo-database (Justman et al., 2018), collected for development and testing of the URC assessment method. Grid cells are 25 × 25 km. Warmer colors (shades of red) indicate a greater number of available datasets. Analysis performed using NETL’s CSIL tool (Romeo et al., 2019). The notable hotspot (red) highlights data density in the Powder River Basin of Wyoming and Montana

Implementation Process

The URC assessment method entails two primary aspects: geological characterization and spatial assessment. Geological characterization involves reconstructing depositional, structural, and post-depositional history of the region of interest using geological, geochemical, and geophysical data. The characterization aspect of the assessment informs of locations where past conditions were favorable for REE accumulation, presented as a “potential enrichment” (PE) score. In this context, the term enrichment is used inclusively to represent both pre- and post-depositional accumulation of REEs in sediment. The spatial assessment aspect integrates results from the geologic characterization with REE concentration data to estimate REE resource prospectivity. In areas with no REE measurements available, the PE score alone is reported to indicate regions with potential for increased REE enrichment solely on the basis of depositional, structural, and post-depositional history.

It is important to note that this approach differs from other geo-data science resource assessment methods due to inherent limitations of the unconventional resources being targeted. For example, geo-data science-based mineral prospectivity mapping involves testing, training, and validating predictions using known deposit locations (Zuo, 2020), which is not practicable for coal-REE deposits as there are no definitively known deposits in the USA. Other methods, such as geochemical mapping methods, involve exploring spatial geochemical patterns and inter-relations and identifying anomalies associated with mineralization (Zuo & Xiong, 2020). Given the sparseness of publicly available geochemical measurements relating to the coal-REE system, the URC assessment method focuses primarily on evaluating geologic data evidencing the processes of enrichment, with supplemental geochemical data integrated only during the final analytical step to provide confidence levels where available.

In practice, implementation of the assessment method occurs in four stages, with the first three addressing geological characterization of the assessment area and the last stage integrating spatially the predicted enrichment potential with the available REE concentration measurements (Fig. 3). Each of the REE-coal deposit types presented in Table 1 is assessed independently, as each deposit type results from unique circumstances. The URC assessment method presented herein is applied on the Powder River Basin (PRB) in Wyoming and Montana for demonstration purposes.

Figure 3

Generalized workflow for implementing the URC assessment method

Step 1: Inventory Knowledge-Data and Assess Limitations of Data Availability

In this step, individual components in the URC assessment method are related to datasets to establish the occurrence of required geologic components, as part of the geologic characterization. Datasets used in the assessment are assigned to at least one geologic component. For example, observation of a volcanic ash layer in coal core represents a potential REE source; thus, the spatial dataset can be assigned to the corresponding component (“Alkaline volcanic ash”; Fig. 4). Some components are not directly testable and thus have no related data. For example, it is challenging, if not impractical or impossible to acquire data that evidences a mire was downwind of contemporaneous volcanic activity. In select cases such as this, evidence for non-testable components may be inferred. In the ash-bearing coal example above, presence of the volcanic ash layer implies the (then) mire was downwind of contemporaneous volcanism. This also allows multiple geologic components to be represented by a single observation or dataset. Implicit relationships are also utilized to assign an occurrence of geologic components for an area. Considering again the ash-bearing coal example above, given a sufficiently thick ash and proximity to volcanism and no tectonic influences, one can assume the ash layer extends laterally beyond the immediate vicinity of the core. Thus, in the absence of data suggesting otherwise, occurrence of the ash can be inferred for a local area around the cored site. Implicit assignment of a geologic component based on spatially explicit (i.e., coal core) data permits an assessment of an area or volume using discrete measurements. Uncertainty associated with these assumptions is captured in step two of the method. See Supplementary Information for the complete table of geologic components and assigned data and the evaluation criteria used in the URC assessment of the PRB (Online Resource 2).

Figure 4

REE enrichment process flowchart illustrating the geologic components that constitute primary and secondary enrichment deposit types. Arrows denote conditional relationships between components with respect to the requirements for enrichment. Arrow lines indicate whether dependencies are related through the logical operators “AND” (solid line) or “OR” (dashed line). Colors indicate related components; lighter shaded components comprise those darker with bold text. *Labels indicate components required for primary enrichment deposit types as described in Table 1: Eo, Eolian; Fl, Fluvial. Enrichment process flowcharts for the other deposit types are included in the Supplementary Information (Online Resource 1)

Leveraging the REE enrichment knowledge base and REE enrichment process diagrams (Fig. 4; Online Resource 1 in Supplementary Information), relevant data can be collected, integrated into a database (e.g., Justman et al., 2018), and assigned to individual components for each enrichment mechanism. To understand limitations of data availability for a particular enrichment mechanism, a “data availability” (Da) metric is calculated to represent the amount of uncertainty (i.e., confidence level) in the PE score. The metric is a ratio of the number of geologic components with related and unqueried data (not filtered to represent support for a given mechanism) versus the total number of components required for a specific occurrence type (Fig. 4), expressed as:

$$Da_{m} = \frac{{\sum Da_{m,c} }}{{Dr_{m} }}$$

(1)

where Da_m is the proportion of geologic components with data available to test for occurrence of mechanism m, Da_m,c is a binary representation of whether a geologic component c has associated data available unique to mechanism m, and Dr_m is the number of geologic components required for mechanism m.

Step 2: Postulate Spatiotemporal Domains based on Shared Geologic Attributes

This involves characterizing the geologic history of the region of interest and identifying areas with common geologic attributes, such as basin sediments with shared provenance or regions with similar structural character, following the Subsurface Trend Analysis (STA) method as described in Rose et al. (2020). The STA method involves using geologic knowledge and data relating to lithology, structure, and secondary alteration—three primary factors affecting subsurface properties—to partition an area into genetically related spatial domains (Fig. 5a). The discrete thematic domains are then overlain to delineate areas with unique geologic histories.

Figure 5

Application of the Subsurface Trend Analysis (STA) method for postulating genetically related spatial domains in Step 2 of an assessment (see detailed discussion in Rose et al. (2020)). (a) STA ternary diagram with the primary factors affecting subsurface properties as end-members (colored text). Spatial domains are postulated for each geologic end-member. Black text lists example data and information utilized in an assessment, plotted according to association with geologic end-members (e.g., “fault zones” information provides constraints for structure domains). Powder River Basin lithologic (b), structural (c), and unique (d) domains are displayed in light red, green, and orange, respectively

In the Powder River Basin test case, thematic domains were configured on the basis of lithology and structure (Figs. 5b, c). Lithologic domains were differentiated based on sediment provenance using paleocurrent measurements from outcropping Late Paleocene sandstone beds (Seeland, 1993). The criteria for delineating structural domains involved structural style (i.e., fault trends, displacement sense) (Flores et al., 2010; Horton, 2017), occurrences of large-scale structural lineaments (Slack, 1981; Robbins, 1994; Sims et al., 2001; Flores et al., 2010) and geophysical gravity and magnetic anomalies (Robbins, 1994; Sims et al., 2001). Secondary alteration domains were not postulated as the available data and information were insufficient to confidently differentiate areas at the basin scale.

Step 3: Apply Fuzzy Logic to Calculate Potential Enrichment (PE) Score

Assessment datasets are queried for evidence of support for an enrichment component based on geological interpretation of the data (Fig. 4). For example, core descriptions could be queried for observations of tonstein horizons, evidencing deposition of volcanic ash and thus a potential REE source.

For each enrichment mechanism m, a metric of “data supporting” (Ds_m) is calculated using fuzzy logic, a decision-making process that can account for uncertainties and ambiguities in data and processes in geologic and mineral systems. The process is derived from the fuzzy set theory (Zadeh, 1965), based on the idea that real world phenomena do not have precisely defined criteria of membership to any one category or class (either 0 or 1), rather a degree of membership (values between 0 and 1).

The Mamdani fuzzy inference system is used due to its ability to incorporate expert knowledge into a set of easily interpretable linguistic rules (Mathworks, 2021) that can align to geological data (An et al., 1991; Carranza & Hale, 2001; Porwal et al., 2003) relating to enrichment processes (e.g., Mutele et al., 2017). The approach involves first defining a membership function for each input variable and output category (i.e. Ds is high or low). Membership functions are applied to transform and weight the range of values within a variable from zero to one, to reflect the relationship between the variable and the output category being modeled. For example, a sigmoid function can be applied to values of distance from an REE source (e.g., low distances have a higher weight than larger distances) to represent a geologic component for an enrichment mechanism (Fig. 6). Linguistic rules take the form of if–then rule statements; these are constructed to combine membership functions (components) using appropriate fuzzy operators for each variable. Fuzzy operators may take the form of “AND”, “OR”, and “PRODUCT” within the if–then rule statements to combine input membership functions to pass onto the membership functions that define the output categories (Bonham-Carter, 1994). Figure 6 illustrates an example if–then rule statement, evaluated for a given discrete location. In this example, the “AND” operator is the logical intersection (or minimum value) between two or more fuzzy membership values and is appropriate where evidence must be present together for a hypothesis to be true. Alternatively, an “OR” operator would yield the logical union (or maximum value) between two or more fuzzy membership values. The “OR” operator is appropriate where any evidence must be present for a hypothesis to be satisfied. For example, if Figure 6 instead applied an “OR” operator in the rule statement, the higher fuzzy inference arrow would determine the output value. Another potential operator is fuzzy “PRODUCT”, which yields the algebraic product between two or more fuzzy membership values. Additional operators used in fuzzy logic, are “SUM” and “GAMMA.” The “SUM” operator is complimentary to the fuzzy product, expressed as (1—(fuzzy “PRODUCT”)), and is appropriate where all evidence combined supports the hypothesis more than each piece of evidence individually. The “GAMMA” operator is the algebraic product of the fuzzy “SUM” and “PRODUCT”, both individually raised to the power of GAMMA (user-defined value of 0–1). Fuzzy “GAMMA” allows a continuous output across all other operators and useful in cases where relationships are uncertain.

Figure 6

Example of fuzzy logic rule statement application in the URC assessment method. Linguistic if–then rule statement is in bottom box (dashed boundary), red curves are membership functions and blue arrows represent the fuzzy logic inference process from input values (distances) to an output (PE score). See text for discussion

The URC assessment method takes advantage of fuzzy logic in two steps: (1) establishing component representations (spatial weighting) for explicit and implicit data, and (2) aligning components to the enrichment process flowcharts and combing them to obtain mapped Ds_m outputs.

For a given component, assigned datasets are first split into explicit and implicit data (Fig. 7). Explicit data are the raw data with exact locations and measurements (e.g., drill core). If the explicit data are associated with a spatial domain, the data are also represented by domain areas where the raw data spatially overlap. Both the explicit and implicit data are converted to a raster format representing a normalized (between 0 and 1) distance from their respective locations for the entire study area, then multiplied together to represent the weighted values of the associated component (Fig. 7). This is done to ensure the importance of the explicit data increasingly diminishes outside of the spatially associated domain boundaries.

Figure 7

Powder River Basin example displaying the process to build the fuzzy representation of a single component within a given enrichment process. Component data are split into explicit and implicit (domain-based) normalized distance-based rasters and multiplied together to represent the component weight (low to high; purple to yellow)

Once the component data are translated to component weights, these fuzzy representations are aligned to their respective enrichment processes and combined to obtain PE score maps. Combination occurs in two stages: first, the representations are integrated into two groups using unstructured and structured methods; second, the two outputs are then combined using a fuzzy sum operator.

The unstructured combination method arithmetically sums the individual component representation maps, then normalizes to total number of required components (Dr_m) to account for varying complexity between enrichment mechanisms, expressed as:

$$Ds_{m} = \frac{{\sum Ds_{m,c} }}{{Dr_{m} }}$$

(2)

where Ds_m is the unstructured cumulative map of supporting component data for mechanism m, Ds_m,c is a map representation identifying where a geologic component c in mechanism m has associated data evidence, and Dr_m is the number of geologic components required for mechanism m.

The structured combination method integrates the component representation maps using fuzzy operators based on the relationships defined in the enrichment process flowcharts (Fig. 4; Online Resource 1 in Supplementary Information). Components designated as “required” (solid black lines) are combined using the “AND” fuzzy operator to ensure the components must be present together for a higher PE score (maximum potential), while “possible” components (dashed black lines) are combined through the “OR” fuzzy operator to ensure the presence of any one of these components adds to the PE score. The structured approach accounts for the hierarchical relationships between component maps based on the type and degree of interconnectedness to other components in the enrichment process (Fig. 4; Online Resource 1 in Supplementary Information). Notably, the structured combination method excludes components deemed “not testable”, as inclusion would nullify results from the testable components. Thus, the structured method normalizes to the proportion of testable components, expressed as:

$$\widetilde{Ds}_{m} = \widetilde{Ds}_{m,c} \cdot Dr_{m}^{*}$$

(3)

where ${\widetilde{Ds}}_{m}$ is the structured cumulative map of supporting component data for mechanism m and ${\widetilde{Ds}}_{m,c}$ is a map representation identifying where a geologic component c in mechanism m has associated data evidence, and ${Dr}_{m}^{*}$ is the proportion of testable geologic components required for mechanism m.

Once cumulative output maps are obtained, the unstructured and structured map pairs are integrated with the fuzzy sum, an increasing linear combination of values, to determine a PE score for each enrichment mechanism, expressed as:

$$PE_{m} = 1 - [(1 - Ds_{m} ) \cdot (1 - \widetilde{Ds}_{m} )]$$

(4)

where Ds_m is unstructured cumulative map of supporting component data for mechanism m, and ${\widetilde{Ds}}_{m}$ is the structured cumulative map of supporting component data for mechanism m. The PE score map output represents spatially the URC resource prospectivity that can be compared across enrichment mechanisms. The fuzzy sum is applied here to increase the PE score in areas with combined evidence, on the basis that these areas have higher potential than those with only single evidence, without having a decreasing effect for areas with lesser evidence.

For a given location, each REE-coal deposit type receives a PE score value ranging from zero to one. Depending upon the available component data (i.e., Da_m), a PE score of zero indicates minimal or no potential for REE enrichment, and a PE score of one indicates high potential for REE enrichment. Results from this step indicate locations with potentially undiscovered REE occurrences and supplement the spatial assessment (Step 4).

Step 4: Spatial Integration of Predictions and Observations

Evaluation methods such success-rate and receiver operating characteristics curves and improved prediction-area plot are useful for validating predictions in areas with known deposit locations (Yousefi et al., 2019). However, since there are no known economically viable or significant deposits of REEs in coal within the USA, validation of PE score predictions is not feasible at the regional scale. Thus, each PE score is qualified with an additional confidence metric (C) using measured REE concentration data in coal samples, where available. The confidence metric is the product of the PE score and the normalized log of summed REE concentrations for a given study area, expressed as:

$$C_{m} = PE_{m} \cdot \left( {\frac{{{\text{log}}({\Sigma }LaCeY_{xi} ) - {\text{min}}\left( {{\text{log}}\left( {{\Sigma }LaCeY_{x} } \right)} \right)}}{{\max \left( {\log ({\Sigma }LaCeY_{x} )} \right) - {\text{min}}\left( {{\text{log}}\left( {{\Sigma }LaCeY_{x} } \right)} \right)}}} \right)$$

(5)

where C_m is the confidence for mechanism m, PE_m is the PE score for mechanism m, LaCeY_xi are the geochemical concentrations for La, Ce, and Y for a single coal sample, LaCeY_x are the geochemical concentrations for La, Ce, and Y for all coal samples within the study area. The sum of La, Ce, and Y concentrations is used as a proxy for total REEs given the analogous geochemical behavior (Montross et al., 2022) and the fact that many samples lack measurements for total REEs.

The confidence metric outputs range from zero to one, which translates as low to high confidence in PE scores, respectively. The intent in presenting via the confidence metric is to inform decision-making associated with further data collection and investigation (i.e., targeted sample or core collection). To spatially visualize and quantify the confidence metric, the URC assessment method applies the Variable Grid Method (VGM; Bauer & Rose, 2015). The VGM enables visualization of PE score confidence in a user-defined number of varying grid sizes that best suits the data uncertainty. Here, smaller grid cells sizes represent higher confidence values, thereby highlighting locations where evidence suggests greater enrichment potential.

Results

As a proof of concept, the URC assessment method was applied to the Powder River Basin (PRB) in Wyoming and Montana to predict areas of potential enrichment of REEs in coal and related strata. The PRB was selected due to the sizeable amount of relevant, publicly available data (Fig. 2), as well as it’s relatively simple geologic history. The assessment was implemented using open-source python libraries, including NumPy (Harris et al., 2020), pandas (Reback et al., 2020), and GDAL (GDAL/OGR contributors, 2021). NETL’s Spatially Integrated Multivariate Probabilistic Assessment tool (Wingo et al., 2019) was used to perform the fuzzy logic modeling for PE score calculation. Data from the REE Open Coal and REE Database (Justman et al., 2018) were identified, assigned, and queried to geologic components (Online Resource 2 in Supplementary Information). All raster datasets were aligned to an aerial grid cell (pixel) resolution of 1 km². All input data and spatial domains as well as output PRB results are provided in the supplementary information (Online Resources 3 and 4, respectively, in Supplementary Information).

Coal-REE concentration data for the confidence metric were obtained from the USGS CoalQual sample database (Palmer et al., 2015) and unpublished core samples from the University of Wyoming, USGS, and industry partners. Due to limited availability of PRB samples with total REE measurements, concentrations of La, Ce, and Y were used as proxies for total REEs, owing to their similar geochemical characteristics.

Figure 8 displays Da_m for the six REE-coal deposit type. Data availability ranges from moderately low (green) to moderately high (yellow-orange) for the PRB area. Fluvial enrichment yielded the lowest Da_m values (Fig. 8a) due to only two geologic components with data available out of the six required components. Fluvial enrichment also yielded no variability across the basin, owing to component 10 paired with a contiguous dataset (i.e., polygons) that extends across the basin. Although multiple datasets are aligned to this component, Da_m only tracks the presence/absence of any data, not the sum of all available datasets. The five other deposit types display bimodal data availability across the basin, owing to a combination of spatial data types for multiple components. Meteoric adsorption deposit type (Fig. 8c) yielded both the highest absolute value and the highest range in values across the basin, demonstrating the combination of local and areal data types. Hydrothermal adsorption deposit type yielded the highest cumulative Da_m across the study area.

Figure 8

Data availability (Da_m) metric for six coal-REE deposit types, indicating the proportion of data available with respect to the amount necessary to characterize an enrichment process. A value of 100% indicates availability of data for all the geologic components involved in enrichment

The PRB PE scores show distinct patterns for each of the six deposit types (Fig. 9). Primary-eolian and primary-fluvial deposit types yielded generally uncorrelated spatial distributions in the highest PE scores. Both secondary-meteoric deposit types show an east–west trend in elevated PE scores toward the north central PRB, a north–south trend of elevated PE scores along the eastern basin margin, and a cluster of highest PE scores in the southeast. The two secondary-hydrothermal deposit types show a similar but more aerially extensive east–west trend in higher PE scores in the north central basin and a north–south trend of higher PE score and clustered highs in the southeast basin. Note that the PE score confidence grid locations (Fig. 10) do not correlate with data availability for the different deposit types, but rather the spatial distribution of CoalQual samples (Fig. 11) and supplemental core samples.

Figure 9

Potential Enrichment (PE) score for six coal-REE deposit types. (a) Maximum PE score obtained among all deposit types. (b–g) PE score results for individual deposit types. See text for discussion

Figure 10

Confidence variable grid overlaying PE scores, indicating areas of lower and higher confidence with larger and smaller cells, respectively. Areas outside cells are those with PE score predictions only; REE concentration measurements are unavailable or not assessed. (a) Maximum PE score and corresponding confidence among all deposit types. (b–g) PE score and confidence for individual deposit types. See text for discussion

Figure 11

Heat map displaying the density of USGS CoalQual samples (Palmer et al., 2015) in US coal basins (dark gray). Warmer colors indicate a higher relative density of coal samples

Discussion

Like any geologic modeling effort, data availability and quality are key for the success of an URC assessment. In order to successfully predict and assess potential URC deposits at mine, seam, or basin scales, the URC assessment method relies upon fine-scale data and studies in addition to larger-scale (“regional”) data and information. Core, outcrop, and mine scale data and information are key to constraining and understanding how and why REEs accumulate in unconventional geologic systems, and the processes and conditions that result in the vertical and lateral heterogeneity of these deposits. In addition, contextual geologic data and information at these finer scales is important for constraining the lithologic, structural, and secondary alteration history of the systems to improve predictions of the occurrence and concentrations of URC deposits at various scales. At present, much of these data and knowledge are either incomplete or do not exist in forms appropriate for supporting quantitative, in situ URC resource estimates.

Since there are no known economically viable or significant deposits in the target area (sedimentary basins in the USA), established evaluation methods such as success-rate curves, ROC curves, and improved P-A plot (Yousefi et al., 2019) are not suitable to validate results from the case study. In line with these constraints, the method and case study results presented here are intended to demonstrate the ability to systematically prioritize and highlight the most prospective sites for future investigation and additional data collection for areas with no known deposits, and sparse geochemical data.

An alternative, less data-intensive approach to assessing in situ REE resources would be applying a geostatistical algorithm such as inverse distance weighting or kriging to interpolate between measured REE concentrations in coal core samples over an area or zone of interest. However, there are major drawbacks to this approach. First, few publicly available samples have the information needed and those samples are not distributed evenly across US coal basins. For example, the USGS CoalQual database (Palmer et al., 2015) contains valuable information and is considered a preeminent source for REE concentrations in US coal basins, yet these data were collected for a different purpose. The homogenized nature of the analyses does not capture intra-seam and horizontal heterogeneity and significantly limits the utility of that data for seam-scale volumetric resource assessment. Moreover, the samples are not distributed evenly across US coal basins or beds, with most of the sampling clustered in select regions (Fig. 11). Second, and perhaps more importantly, a simple interpolation approach neglects valuable information that can be gleaned from important contextual data. Interpolating across geologic domain boundaries (e.g., across a fault zone or across areas with vastly different lithologies) ignores changes in rock properties that may affect REE enrichment conditions. Conversely, as demonstrated by the success of the petroleum systems concept in petroleum exploration, incorporating geologic contextual data into an assessment can significantly improve results. With knowledge of the geologic mechanisms of REE accumulation in coal, one could predict where REEs likely occur in significant concentrations for a given area that experienced favorable conditions for accumulation, without having directly measured the REE concentrations. This consideration enables an assessment of potential enrichment for areas with sparse or even no REE concentration measurements. Hence, the URC assessment method involves analysis of both measured geochemical concentrations and contextual data and information.

Future scenarios in which sufficient data and information are available could support quantitative estimates of resources using the URC assessment method, primarily at the mine or seam scale, depending upon the data qualities. Key to this effort would be core data characterizing not only the vertical and lateral distribution of REEs, but also host rock mineral phases, geochemical composition, and lithostratigraphy to develop paragenetic relationships and identify potential enrichment mechanisms (Montross et al., 2022). Adequate and consistent spatial coverage and sampling resolution of cores would also be required to minimize the influence of sampling bias and ensure recording of small-scale variations. As an added benefit, insights from such a systematic, geologically informed sampling and characterization strategy could help refine the conceptual models for coal-REE deposits used in the URC assessment method.

PE scores from the PRB use case may also inform future data collections efforts by highlighting priority types of data. Comparing relationships between evidence for geologic components and the PE score for a given location, it is possible to identify corresponding component datasets with greater relative coincidence with high PE scores across the study area (Fig. 12). For example, for secondary-meteoric adsorption-type deposits, cells with data evidencing component 54 (“neutralizing pH”) yield nearly 2 × the PE score than mean PE score for all cells in the study area. Based on high incidence of component evidence per PE score, future smaller scale assessment efforts within the PRB could prioritize data collection for “neutralizing pH” evidence and other components with component/PE ratios greater than 1, with the understanding that evidence captured in these particular data may serve as proxies for high PE score localities. However, the component/PE ratio criteria applies only to components with data located inside the PRB boundary. Several lithologic domain components utilize data located proximal yet adjacent to the boundary to inform provenance for REE source-related components. Thus, select component-data cannot be evaluated for component influence on the PE score.

Figure 12

Impact of individual component datasets on the PE score. X-axis is individual component datasets from the REE enrichment process flowcharts, identified by component numbers and grouped by coal-REE deposit type. Components numbers are provided in the URC query table in Supplementary Information (Online Resource 2). Blue bars indicate the ratio of the mean PE score in cells with full evidence for a component (Ds_m,c = 1) versus the mean PE score for all cells in the study area. See text for discussion

Interpretation of the PE score benefits from the confidence metric in several ways. By integrating predictions of REE enrichment with measured concentrations, the PE score confidence maps present the assessment results in terms of a hypothetical prediction and geochemical observations (Figs. 10, 13). Presenting the results in this manner conveys the type of information underpinning the assessed area, such that the smallest grid cells (highest columns) indicate areas with measurements in accordance with predictions. At the same time, cells with a high PE score yet located within larger grids highlight areas with higher potential but are unconfirmed. Such locations should thus be priority targets for future sampling campaigns. Moreover, comparison of the individual PE score confidence maps offers insights into the predominant REE phase for a given location. For example, for secondary-meteoric adsorption deposit type (Fig. 10e), the southeastern PRB yields low to intermediate PE scores and large grid cells, while the same area for secondary-hydrothermal adsorption-type (Fig. 10f) yields a north–south trending PE score high coincident with the smallest grid cells. Although the measured REE concentrations alone from this region would indicate a relatively high REE occurrence, there would be no knowledge of the mode of REE occurrence for the potential deposit. Determining that the coal-REE deposits are likely hydrothermal adsorption-type informs not only predictions of LREEs/HREEs, but also the technology needed to recover and beneficiate these resources, and thus is key when assessing technical and economic recoverability.

Figure 13

Three-dimensional representation of URC assessment results for the Powder River Basin test case, based on Fig. 10a. Colors indicate maximum value PE score among the different deposit types for a given grid cell (cooler colors are lower scores, warmer colors are higher scores). Grid cells are 1 × 1 km in X,Y dimensions; Z dimension scales with PE score confidence (higher confidence = higher extrusion). See Supplementary Information for an interactive three-dimensional figure (Online Resource 5)

It is important to note that the URC assessment method presented here evaluates components based on only the presence or absence of data assigned to the components. It is recognized that this approach neglects the sequential nature of enrichment processes, such that the order of events (i.e., timing of geologic components) is not random. Temporal relationships of components for the various deposit types have been resolved (Online Resource 1 in Supplementary Information); however, the primary limitation in applying this understanding is obtaining data that both represents a component and provides temporal constraints, especially at larger scales. There are also select instances in which geologic components are not evaluated for the PE score. In these cases, components are either not testable or no data is available. If a component is not testable, the maximum PE score for that deposit type is thus limited. Applications of machine-learning algorithms such as natural language processing could soon mollify these limitations by unlocking new knowledge and data from scientific literature (e.g., Morkner et al., 2022), thereby offering new resources for evaluation.

As new studies and results emerge on REEs in sedimentary systems and the understanding of enrichment processes and deposit characteristics matures, the URC assessment method can evolve to incorporate those findings and in turn provide improved predictions (Kreuzer et al., 2020). The URC knowledge-data framework utilized in this method was designed for adaptability by accepting and fully leveraging the current state of geologic knowledge of REE occurrence and controls on distributions in sedimentary systems. Future implementations of the URC assessment method will involve testing in different geologic settings and exploring the applicability of other critical mineral deposit types.

Conclusions

The Powder River Basin case study demonstrates the utility of the URC assessment method for predicting and assessing potential unconventional resources in sedimentary lithofacies for areas with no known deposits, in this case REE-enriched coal and related strata. The output PE score and Confidence maps validate how the knowledge-data framework approach can significantly reduce search areas for coal-REE deposits by highlighting priority exploration targets for future investigation. Even in areas with no geochemical concentration data, the PE score map outputs identify target areas where data evidence support the possibility of enrichment. Additional data collected from these target areas can, in turn, be integrated in the future assessment to yield more robust results. Moreover, the URC assessment method provides a means to contextualize existing and future REE measurements in terms of potential REE phase, as evidenced by the PE score and Confidence maps. This information is key for determining the extraction and beneficiation potential of the mineral resource.

The URC assessment method is flexible by design, allowing modifications to the underlying knowledge-data framework as new insights are discovered. As new studies and results emerge on REEs and other critical minerals in sedimentary systems and the understanding of enrichment processes and deposit characteristics matures, the URC assessment method can evolve to incorporate those findings.