Rare-earth elements (REEs) are comprised of the 15 elements that make up the lanthanide group of the periodic table and also include yttrium and scandium because of their similar physical and chemical properties (Castor and Hedrick, 2006; Gambogi and Cordier, 2010; Walters et al., 2010; Liao et al., 2013; Golev et al., 2014). The elements can be grouped into two different categories based on their atomic numbers. REEs with atomic number 57–63 are classified as light-rare earths (LREEs), and REEs with atomic number 64–71 are classified as heavy-rare earths (HREEs). Scandium and yttrium are usually classified as HREEs (Humphries, 2011; Yang et al., 2013; Pothen, 2014). From other work, a third category, i.e., medium-rare earths (MREEs) is used to describe samarium, europium, and gadolinium (Chen, 2011; Hatch, 2011; Tharumarajah and Koltun, 2011). Table 1 lists the 17 REEs with their main applications, with applications related to energy in bold.

Rare-earth elements have traditionally been applied in a variety of domains and products, including catalysts, ceramics, communications, electronics, lighting, phosphors, and polishing powders (Eggert et al., 2008; Hatch, 2012; White et al., 2013). What distinguishes REEs from other elements is their particular electronic structures (i.e., 4f orbitals), which offer unique chemical, structural, and physical properties (Korinek and Kim, 2010; Graf et al., 2013; Fouquet and Martel-Jantin, 2014). The REE industry’s importance has ballooned in the last few years, emerging from an industry that was widely ignored to one that is essential to an environmentally efficient economy (Bourzac, 2010; Kara et al., 2010; Binnemans et al., 2013; Clapper, 2013; Adibi et al., 2014). Furthermore, REEs have gained worldwide attention, given their application in hybrid vehicles, wind turbines, and military systems (Hagelüken and Meskers, 2009; Grasso, 2013; Kifle et al., 2013; Lewis and Jiménez-Villacorta, 2013; Bogue, 2014). For example, it is estimated that wind turbines, incorporating NdFeB magnets, contain 171 kg of REEs per MW of capacity (Wallington et al., 2013). Also, the properties of REEs allow them be used to miniaturize components and appliances (Brown et al., 2002; Chakhmouradian and Wall, 2012). Given that REEs provide critical functionality in many applications, especially sustainable mobility and energy supply (Gibson and Parkinson, 2011; Alonso et al., 2012; Hatch, 2012), the interest and need for them is growing and will continue to grow (Bauer et al., 2010; Paul and Campbell, 2011; Moss et al., 2013; Wallington et al., 2013).

Accounting for more than 90% of the total world supply, China is currently the largest producer of REEs (Chu and Majumdar, 2012). However, in 2008, China began the process of implementing a REE production quota as well as limiting the number of permitted exporters, which has made the global REE market extremely volatile (Walters et al., 2010). This has forced other countries to examine their supply chains for potential risks (Gambogi and Cordier, 2010; Hurst, 2010; Tse, 2011; McLellan et al., 2013; He, 2014). In 2008, the European Union (EU), through the European Commission’s Raw Material Supply Group, prepared a national plan recommending direct actions be taken to increase market accessibility, increase investments in R&D, and support recycling programs (Blakely et al., 2012). The EU has also established the Strategic Energy Technology Plan, which aims to promote renewable resources and energy efficiency in order to tackle climate change (Moss et al., 2011). Japanese firms and the Japanese government have searched for mines throughout Asia, Africa, and the Americas to secure supplies of REEs (Humphries, 2011; Blakely et al., 2012). In 2011 and 2012, the U.S. Department of Energy (DOE) released the Critical Materials Strategies, calling for a reduced dependence on critical materials, as well as ensuring that energy technologies are not obstructed by future supply shortages (Grasso, 2013). The most critical materials identified by the DOE are all REEs: Dy, Nd, Tb, Eu, and Y (Bauer et al., 2010). The Critical Materials Energy Innovation hub was established recently to address challenges in mineral processing, material substitution, and reuse and recycling (EERE Network News, 2012; Goldman, 2014).

Although many REEs are critically important to sustainable mobility and energy supply, production of REEs itself incurs significant environmental damages (Schüler et al., 2011; Ali, 2014). The Mountain Pass mine in California, which operated by MolyCorp and dominated the world REE production for decades (accounting for 70% of the world’s supply in the 1970s and 1980s), closed in 2002 (Bourzac, 2010; Chakhmouradian and Wall, 2012; Shujing, 2013). One of the two major causes of this closure is environmental issues with the other being intensive price competition from Chinese producers (Meyer and Bras, 2011; Schüler et al., 2011; Fuerstenau, 2013).

Similarly, large scale REE production in China has raised serious environmental concerns, particularly with regards to heavy metal and radioactive emissions in groundwater, river ways, soil, plants, and the atmosphere around mine sites (He et al., 2004; Liang et al., 2014). The Chinese Society of Rare Earths, estimate that a ton of REE produces 8.5 kg of fluorine and 13 kg of flue dust, while refining with H₂SO₄ generates 9,600–12,000 m³ of gas containing flue dust concentrate, HF, SO₂, and H₂SO₄ (Paul and Campbell, 2011; Sonich-Mullin, 2012). Other estimates indicate that producing one ton of REEs creates 60,000 m³ of gases mixed with H₂SO₄ and HF, 200 m³ of acid water, and 1.4 tons of radioactive waste (China Daily, 2009; Blakely et al., 2012). In addition, long-term mining dust inhalation can cause pneumoconiosis (black lung) (Hirano and Suzuki, 1996; China Daily, 2009). For example, the city of Baotou, located 120 km from the Bayan Obo mine, has had considerable pollution issues due to the nearby REE processing facilities (The Guardian, 2012). The Chinese government has cited environmental damage as one of the major drivers to close heavy pollution facilities and limit production (Blakely et al., 2012; Ting and Seaman, 2013). Overall, production of REEs involves a large number of process steps, out of which many incur significant material/energy consumption and environmental release (Golev et al., 2014). A comprehensive assessment of the environmental impacts represents a key first step toward reducing the environmental damage of REE production.

Currently, the most widely accepted (and arguably the most objective) methodology for environmental sustainability assessment is life-cycle assessment (LCA) (Ness et al., 2007; Evans et al., 2009). In recent years, LCA has become a core tool in sustainable product development and environmental policy making for companies and governments all over the world (Guinee et al., 2010). This paper starts by providing a brief overview of the LCA methodology, followed by a concise introduction of major REE production routes with a focus on material/energy flows and environmental emissions. Published LCA studies related to REEs are then reviewed to identify knowledge gaps and future efforts needed.

In short, LCA takes a holistic approach and provides a complete view of the environmental impacts over the entire life cycle of a process or product, from raw material extraction and acquisition, manufacturing, transportation and distribution, use and maintenance, reuse and recycle, and all the way to disposal and waste management (Curran, 2006; Lehtinen et al., 2011; Lizin et al., 2013). LCA avoids shifting environmental consequences from one life-cycle stage to another, from one geographic area to another, and from one environmental medium to another. LCA has been used to study the environmental impact of energy technologies such as electric vehicles, solar photovoltaics, wind turbines, etc. (Samaras and Meisterling, 2008; Martinez et al., 2009; Tremeac and Meunier, 2009; Garrett and Rønde, 2013; Lizin et al., 2013). ISO 14040 and 14044 developed by the International Standards Organization provide principles and guidelines for evaluating environmental impacts of products using LCA (Suh and Huppes, 2005; Finkbeiner et al., 2006; ISO, 2006; Ness et al., 2007). According to ISO 14040, a typical LCA can be carried out in four interdependent steps: goal and scope definition, life-cycle inventory (LCI), life-cycle impact assessment (LCIA), and interpretation of results (Figure 1) (Curran, 2006; Lizin et al., 2013; Weidema et al., 2013). It should be noted that several iterations are generally needed when conducting an LCA (shown as the bi-directional arrows in Figure 1).

As shown in Section “Major REE Production Pathways,” there is significant amount of literature describing the material and energy inputs of process steps involved in representative REE production pathways. However, it should be noted that the data and information reported for the Mountain Pass pathway are over 20 years old. On the other hand, the description of production pathways used in China reflects more recent practices. Unfortunately, emissions and environmental damages associated with these pathways are rarely quantified or at best discussed in an anecdotal manner. For all pathways, only aggregated environmental emissions are reported, i.e., at facility level instead of process level. All of these make it challenging to conduct an LCA study. As a result, to date only a handful of LCA studies on REE production have been conducted. In the following paragraphs, these LCA studies will be critically reviewed.

The datasets in the Ecoinvent database (Ecoinvent v2.0 and subsequent v2.2 and v3.0) represent probably the first LCA study related to REEs (Althaus et al., 2007), which examines the environmental impacts of the Bayan Obo pathway. As shown in Figure 9, the study covers mining, physical beneficiation, cracking, and solvent extraction while leaving out REM refining and purification. Correspondingly, six unit processes/datasets are included: rare-earth concentrate, 70% REO, from bastnasite, at beneficiation/CN; Cerium concentrate, 60% Cerium oxide, at plant/CN; neodymium oxide, at plant/CN; samarium–europium–gadolinium concentrate, 94% rare-earth oxide, at plant/CN; lanthanum oxide, at plant/CN; and Praseodymium oxide, at plant/CN. The first unit process includes mining and physical beneficiation, while the remaining five unit processes cover the cracking and the multi-stage solvent extraction process. Here, the geographical code “CN” represents production in China.

The Ecoinvent datasets were developed by aggregating data from literature and stoichiometric estimations. Energy use, chemical consumption, transportation, and infrastructure are all considered when developing inventory data. Due to the quality and availability of data, the study has the following limitations:

Ore composition. The REO composition used in the study is slightly different from the average of bastnasite at Bayan Obo. It seems that the composition is based on annual production weighted average of Bayan Obo and Mountain Pass. In addition, it should be noted that although it is claimed that the study investigates the Bayan Obo pathway, the ores processed at Bayan obo are actually mixture of bastnasite and monazite.

Although the LCI datasets in Ecoinvent database carry many limitations, the study provides a good starting point for further LCA efforts. Nuss and Eckelman (2014) conducted an LCA of 63 metals, including all of the 15 lanthanide elements. The default Ecoinvent inventory data was used but the energy/material inputs and emissions were re-allocated to all the 15 REOs based on Bayan Obo bastnasite composition provided in Chinese Rare-Earth Yearbook 2010 and 2006–2010 REO prices (Nuss and Eckelman, 2014). It was concluded that for REOs the cracking/solvent extraction steps carry larger environmental impacts than mining/physical beneficiation. It is interesting to point out that the study develops LCI datasets associated with a pathway of co-producing thorium and REEs from monazites. The data was largely based on environmental assessment report for a planned Lynas project in Malaysia, which will process REO concentrates derived from monazites mined at Mount Weld, Australia (Schmidt, 2013).

There have been efforts on addressing some of the limitations associated with Ecoinvent datasets. Koltun and Tharumarajah (2014) proposed a two-step allocation procedure to deal with the co-mining issue at Bayan Obo. In the first step, environmental burdens of mining and physical beneficiation are allocated to hematite (FeO), columbite (niobium ore), and REO bearing ore (i.e., bastnasite and monazite) based on mass fractions and prices. In the second step, the allocation among all the 17 REOs was done based on economic value, assuming recovery rate for each REE oxide is the same. Different from Ecoinvent, the study considers the REE ores as mixture of bastnasite and monazite with 3:1 ratio. In a separate study, Koltun and Tharumarajah (2010) expanded the system boundary to include reduction of REOs. These two publications, however, lack critical details on how life inventory data was compiled and do not include a complete LCI. It was implied that the inventory was developed based on Ecoinvent dataset but the lack of transparency makes it extremely difficult for other researchers to repeat their studies.

In a recent study, Sprecher et al. (2014) conducted comprehensive updates on Ecoinvent datasets. Seven major REOs are included: cerium oxide, lanthanum oxide, neodymium oxide, praseodymium oxide, europium oxide, gadolinium oxide, and samarium oxide. Since the study aims at assessing life-cycle environmental impacts of NdFeB magnets, particular attention was put on neodymium. Two Ecoinvent unit processes, i.e., rare-earth concentrate, 70% REO, from bastnasite, at beneficiation/CN and neodymium oxide, at plant/CN were split into five processes: ore removal from mine, beneficiation of REE containing ore, acid roasting, leaching, and solvent extraction. For the mining process, it argues that the REE ore mining in Bayan Obo is equivalent to regular open pit iron ore mining, and the default Ecoinvent process of iron ore mining was adopted. The REO concentration in the ores is assumed to be 4.1%, which is more realistic than the 6.0% used in Ecoinvent. The same approach was used for REE ore beneficiation because beneficiation techniques used in the Bayan Obo mine can be assumed to be the same as in a normal iron ore mine. Radioactive emissions were also updated based on thorium and uranium content of monazites mined in Bayan Obo. In addition, allocation was done between iron ore and REE ore using economic value. For acid roasting, leaching, and solvent extraction, the study adopted amounts of chemical consumption (e.g., sulfuric acid, hydrochloric acid, ammonium bicarbonate, and sodium hydroxide) reported in several Chinese publications. Given the focus on neodymium, allocation for solvent extraction process was done only between neodymium and praseodymium oxides and other REOs.

Sprecher et al. (2014) further expanded Ecoinvent datasets to include production of pure neodymium metal via molten salt electrolysis, which is similar to the Hall–Heroult process used for aluminum production. Therefore, the production of metallic neodymium was modeled based on Ecoinvent process “Aluminum, primary, liquid, at plant/RER” while adjusting energy and material inputs based on properties of alumina and neodymium oxide. It is interesting to point out that this study also explores two hard disk drives magnets recycling/remanufacturing routes: one via manual dismantling and the other via shredding.

To date, the most comprehensive LCI datasets on REEs are probably those in the GaBi databases (Figure 10) developed by PE International in 2011 and 2012 (PE International, 2012). The databases include 20 datasets, i.e., cerium and its oxides, lanthanum and its oxides, praseodymium and its oxides, neodymium oxides, samarium and its oxides, europium and its oxides, gadolinium, terbium, holmium, erbium, thulium, ytterbium, lutetium, and yttrium. Production of pure metals via molten salt electrolysis is also included. According to the database document, the main data source is the Chinese Rare-Earth Industry Report 2009 and additional oral communication with experts. Although having extensive coverage, the datasets have many issues. Overall, it seems that all the LCI data is highly aggregated. From the documentation, it is difficult to tell whether or not the solvent extraction process was included. In addition, two different pathways are selected for the production of pure metal and the production of REOs. For the production of pure metal, it was assumed that mining, beneficiation, roasting, and production of rare-earth chlorides are all conducted in Sichuan, China. The REE ores available in Sichuan are mainly bastnasites, and the annual processing capacity of facilities in Sichuan only account for less than 20% of the total capacity in China (Ministry of Environmental Protection (MEP) of the People’s Republic of China, 2011). For REOs (except neodymium oxides), the production in Bayan Obo was modeled. Although allocation was performed on co-mining and co-processing of iron ore and REE ore, no details were provided regarding cracking and solvent extraction. In both pathways, the database-documentation suggests that cracking is done via either direct acid leaching or acid leaching followed by calcination. These technologies seem to be only used by facilities in Sichuan while processing concentrated ores from bastnasite (please note Bayan Obo facility uses high temperature sulfuric acid roasting for cracking) (Huang et al., 2011). Moreover, due to some reasons the database-documentation mentions that monazites are predominant in China, which is simply not true (the three leading REE-bearing ores mined in China are Bayan Obo bastnasite–monazite mixture, ion-adsorption clay in southern provinces, and bastnasites in Sichuan). The dataset of neodymium oxide was developed using Mountain Pass production data from 1995, which is almost 20 years old. To make it worse, even though the Mountain Pass facility has restarted operation, its processing capacity is much smaller than that of China. All of these make the representativeness of the inventory datasets questionable.

In summary, very limited LCA studies have been conducted on REEs. All of the studies are based on Mountain Pass production data from the 1990s or information from Chinese literature or a combination. In general, the data quality is low. For many process steps, no information is available and surrogates are commonly used. This is a common practice in LCA efforts and reasonable in some cases. For example, mining of REE-bearing ores can be approximated by open pit mining of iron ores. However, using vegetable oil extraction as an approximate to REO separation to estimate electricity consumption may not be a good choice. In addition, the current LCI databases do not have entries corresponding to the solvents (e.g., P204 and P507) used in REO separation. Generic organic chemical is usually used as surrogate, which may introduce large errors.

Both mass based and revenue based allocation have been adopted to deal with the multi-products issues associated with several REE processing steps. Although ISO 14040 recommends system expansion over allocation, this is not feasible in the case of REEs. On the other hand, both mass based and revenue based allocation have their own shortcomings. Mass-based allocation tends to put most environmental burdens to metals of high abundance, e.g., cerium and lanthanum, while revenue based allocation suffers from extremely volatile REE market. In addition, the prices of REEs vary significantly from metal to metal (e.g., scandium is currently priced at $17,500/kg while cerium at $12/kg, July 30, 2014, www.mineralprices.com).

In addition to data accuracy, completeness, representativeness issues, and allocation method selection, published LCA studies carry some discrepancy with regard to system boundaries, cutoff criteria, and assumptions made. As a result, the LCA results likely carry large uncertainties. For example, the carbon footprint of 1 kg neodymium oxide reported in the studies ranges from 12 kg CO₂-Eq (Sprecher et al., 2014) to 66 kgCO₂-Eq (Koltun and Tharumarajah, 2014).

Due to the limitations associated with current LCA studies, further efforts are clearly needed. The Mountain Pass facility has been upgraded for reduced environmental impacts and restarted operation in 2013, justifying an updated LCA. For the Bayan Obo pathway, a critical analysis of the current studies and a closer look into Chinese literature are desired in order to harmonize the results so the LCA can provide better support for process development and policy making. The DOE’s National Renewable Energy Laboratory (NREL) has done work on LCA harmonization of solar and wind energy (Heath and Mann, 2012) and the methodology could be adopted. In addition, no LCA has been attempted for the production of REEs from ion-adsorption clay, although ion-adsorption clay has been an important source of HREEs. Moreover, there are specific chemicals used in REE processing for which LCI should be developed.

There have been certain sites designated for REE extraction outside of China. In the U.S., these include mines in Diamond Creek, ID, USA, Bear Lodge Mountains, WY, USA, and Pea Ridge, MO, USA (Paul and Campbell, 2011; Long et al., 2012). There are also reserves in Australia, Brazil, Canada, and India (Kanazawa and Kamitani, 2006; Hatch, 2011; Schüler et al., 2011). Japanese scientists have begun to explore the Pacific Ocean for REEs and have shown substantial reserve potential (Kato et al., 2011). Each of these sites has their associated ways of extraction and different processing pathways may be adopted. This will add to the complexity of assessing the environmental impacts of REEs using LCA. In addition, with the advent of new processing technology, new life inventory will need to be developed. Depending on the nature of the new technology, this could be done by updating current inventory, although this will not be easy given the fact that most of the current inventory lacks the needed transparency.

Furthermore, recycling of REEs will be an important topic for future LCAs. Since many products only contain small amounts of REEs, recycling is costly and energy intensive. Estimates of recycling rates are around 1% (Buchert et al., 2009; Chakhmouradian and Wall, 2012; Reck and Graedel, 2012; Binnemans et al., 2013; Bogue, 2014). Given the uncertainties in price, quality, quantity, and dependability of the materials, the research is increasing for REE recycling and this percentage will certainly rise. Companies such as Toyota, Honda, Hitachi, and Mitsubishi have announced REE recycling initiatives targeting as much as 10% of their REE consumption via recycled materials (Akahori et al., 2014; Golev et al., 2014). Further, there have been attempts to look at the recycling and remanufacturing of products incorporating REEs (Dent, 2012; Nagai and Uzawa, 2014; Tan et al., 2014). For some REEs, such as the Nd and Dy in NdFeB permanent magnets, the in-use stock can be four times that of the yearly extraction amount (Du and Graedel, 2011b). A typical approach for recycling includes: collection, dismantling, separation, and processing (Buchert et al., 2009; Sonich-Mullin, 2012). Recycling would further complicate the REE life cycle. The methods and processes for recycling would have to be evaluated, and a couple of LCA studies have been conducted on different routes of recovering Nd and Dy from magnets (Akahori et al., 2014; Sprecher et al., 2014).

As REEs gain increasing importance in many new energy technologies and systems, with the drive to reduce GHG emissions the demand for REEs will keep increasing. Although REEs contribute significantly to energy sustainability, the production of REEs is energy/material intensive and heavily polluting. With the increasing demand of REEs, a comprehensive understanding on the environmental impacts of REEs production is needed. LCA is the most widely used methodology for evaluating environmental sustainability. However, very limited LCA studies have been conducted on the production of REEs.

A critical review of LCA studies on REEs (as well as major REE production pathways) reveals some severe limitations. The LCI was mostly developed based on data from Chinese literature and/or Mountain Pass production data from 1990s. Data consistency, accuracy, completeness, and representativeness are of significant concerns and the LCA results are expected to carry large uncertainties. The studies to date usually draw system boundaries, define cutoff criteria, select allocation methods, and make assumptions differently. Therefore, a harmonization effort is needed. In addition, all the studies are for the processing bastnasite or bastnasite–monazite mixture, while no LCA has been conducted on the processing of ion-adsorption clay, which is an important source of HREEs.

Worldwide, there are sites being explored for REE ore extraction to meet the increasing demand and to reduce supply risks. Depending on the composition of the ores, different processing pathways may be adopted. There are also efforts undergoing to improve REE processing technologies, and to recycle REEs. All of these have impacts on REE life cycle, and LCI has to be developed. Due to the proprietary nature of these new developments, close collaboration with industry is critical. It is envisioned that a comprehensive and transparent LCA database will provide better support to efforts on greening the REE life cycle and to policymaking.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.