Sampling and blending in geoenvironmental campaigns - Amulsar

Sampling and blending in geoenvironmental campaigns – current practice and future opportunities A Parbhakar-Fox1 and S C Dominy2,3 ABSTRACT The prediction of acid and metalliferous drainage (AMD) from mine waste materials is critically important during all stages of the mine value chain. However, to determine such geoenvironmental characteristics static and kinetic testing of individual waste units must be performed on representative samples. The importance of sample selection cannot be underestimated; it is the single most critical aspect of any geoenvironmental investigation. Poor sampling techniques and inadequate sample selection will contribute to excessive variance, difficulties in interpretation and incorrect assessment. By undertaking mesotextural characterisation based on hyperspectral mineralogical data (with sulfur assay data if available), opportunities to improve sampling strategies are presented. Such data will inform the selection of samples for low-cost total sulfur, paste pH and geoenvironmental logging, which collectively will allow for accurate AMD forecasting and improved sampling practices to be achieved. Once sampled, robust quality assurance/quality control (QA/QC) protocols must be developed where the use of international certified reference standards become obligatory. This is currently not practised, resulting in inter-laboratory discrepancies when undertaking waste classification. Further improvements in geoenvironmental characterisation testing lies in the adoption of blended static testing protocols, whereby blends of waste materials can be efficiently tested so as to give an early forecast as to whether they will produce acidic or metal-laden drainage. Such information is integral in developing a waste schedule that will allow for a non-acid forming waste landform to be constructed right from the start of the operational phase. Ultimately, improvements in this discipline lie in technological innovations with regards to mineralogical characterisation, and by offering geoscience and minerals engineering students a competent education in the discipline. Collectively, this will empower a step-change in AMD prediction and waste handling, and will make the realisation of the United Nations Sustainable Development Goals a reality, rather than a dream.

INTRODUCTION The scrutiny faced by the mining industry with regards to their environmental footprint is ever increasing, with several articles published on a monthly basis focused on water and airborne contamination related to their activities (Noble, Lottermoser and Townsend, 2016; Kristensen, Taylor and Flegal, 2017). Indeed, in the current socio-economic climate, environmental management practices are of significant concern to the industry, with the ‘social licence to operate’ (ie neglecting mine rehabilitation obligations which are impacting on the sector’s image) listed as the fourth biggest risk in 2016–2017 for any new mining project or operations (Figure 1). Rightly so, there is no masking the vast quantity of waste generated by the mining industry. For example, in 2010 it was estimated that approximately 14 Bt of tailings were produced globally, whilst up to 30 Gt of solid mine waste are produced annually (Mudd, 2013; Adiansyah et al, 2015). If these

mine waste materials are sulfidic, they will oxidise to produce acid and metalliferous drainage (AMD) the management of which is perceived to be the greatest environmental challenge facing the mining industry (Lottermoser, 2017). Once oxidation processes have begun, acid generation can continue for hundreds to thousands of years after mining has ceased Alakangas, Andersson and Mueller (2013). In Australia alone, there are at least 50 000 sites with historic or legacy mine wastes, the majority of which require rehabilitation (Mudd, 2008). To fund rehabilitation work at abandoned mines (Unger et al, 2015), supplementary funding from Government bodies is likely required, a topical issue drawing much media attention in Australia (Roache and Judd, 2016). Ultimately, such funds are sourced from taxes, therefore compromising the allocation of funding towards perceived priorities, ie improved transport infrastructure and upgrading community services.

1. MAusIMM, Research Fellow, ARC (Australian Research Council) Industrial Transformation Hub for Transforming the Mining Value Chain, University of Tasmania, Sandy Bay, Tas.

Email: [email protected] 2. FAusIMM(CP), Visiting Associate Professor, Camborne School of Mines, University of Exeter, Penryn Cornwall TR10 9FE, UK. Email [email protected] 3. Adjunct Professor, Department of Mining and Metallurgical Engineering, Western Australian School of Mines, Curtin University, Bentley WA; Adjunct Professor, Resources Engineering,

Department of Civil Engineering, Monash University, Clayton Vic 3800. EIGHTH WORLD CONFERENCE ON SAMPLING AND BLENDING / PERTH, WA, 9–11 MAY 2017

45

A PARBHAKAR-FOX AND S C DOMINY

The approach adopted in additive rehabilitation strategies is to use raw materials such as calcium carbonate or calcium hydroxide, both of which are costly (up to A$200 per tonne for water treatment materials). Instead, comprehensive characterisation of geoenvironmental properties of a deposit at the early stages of mine life and correctly planning for mine closure is a far better socio-economic approach to safe guard the environment of mining. However, to achieve this, an adequate number of samples must be characterised. Undertaking effective environmental ore characterisation at the scoping, prefeasibility and feasibility stages are essential for both efficient mine operations and reducing environmental impacts post-closure. Environmental parameters requiring characterisation, include the propensity of a rock unit to generate acid, mapping deleterious element deportment, and characterising the release of toxic dusts (Parbhakar-Fox and Lottermoser, 2015). Geoenvironmental characterisation and modelling is a key input into the geometallurgical approach to mine value chain optimisation (Dominy and O’Connor, 2016). Geometallurgy attempts to resolve ore and waste variability by the use of quick and inexpensive proxies, avoiding the use of many samples and tests (Dominy and O’Connor, 2016). As technology continues to advance, techniques such as handheld analytical tools and automated core scanning allow for faster, less expensive in situ testing to supplement more expensive laboratory testing. This paper evaluates current recommendations for geoenvironmental sampling across the mine value chain and compares this against industry practice (based on publicly available data), with recommendations for improvements given. In the second part of the paper, the focus shifts to reviewing the accuracy of the commonly used geoenvironmental testing suite and presents a case study in which a blended static testing strategy was adopted as a means of better replicating real waste landform evolution. It concludes with how can geoenvironmental characterisation can be improved and keep the mining industry on track to achieve United Nations Sustainable Development Goals by 2030 (UNSDG, 2015).

GEOENVIRONMENTAL SAMPLING FOR ACID AND METALLIFEROUS DRAINAGE Theory of Sampling The mining industry routinely collects samples to assist with decision-making, whether for exploration, resource estimation, grade control and geoenvironmental characterisation, or plant design and balances. Poorly designed sampling protocols can result in elevated project risk by increasing variability. Critically, variability produces both financial and intangible losses. Samples should be collected and prepared within the framework of the Theory of Sampling (TOS; Gy, 1982; Pitard, 1993), along with an appropriate QA/QC system (Vallée, 1999). TOS defines key sampling errors and provides guidelines for reducing them (Gy, 1982; Pitard, 1993). The results of QA/QC programs will ultimately provide evidence for representivity through the application of duplicate field and laboratory samples. The precision of sampling protocols relates to the constitution heterogeneity (CH) of the material in question and leads to the fundamental sampling error (FSE). Poor precision in samples generates type misclassification. The FSE can be estimated via the FSE equation as part of an optimisation program (Gy, 1982). Out of all sampling errors, the FSE does not cancel 46

out and remains even after a sampling operation is perfect. It is controlled via the optimisation of sample mass and size reduction process. The distribution heterogeneity (DH) and related grouping and segregation error (GSE), also contribute to sample precision. DH represents the difference in average composition of the lot from one place to the next in the lot; it is responsible for the irregular distribution of grade and values in groups of fragments of broken ore. The DH can be influenced by large differences in density and fragment composition. DH of a given lot is controlled by CH, the spatial distribution of fragments and the lot shape thus DH leads to GSE. From a practical perspective GSE cannot be measured, but may have a material effect on the total sample error. It is controlled by accumulating many small increments to form a composite sample. Segregation can theoretically be reduced by homogenisation, though in the presence of dense liberated metallics (eg gold) and/or sulfide particles can be a futile exercise that promotes further segregation. The other sampling errors (delimitation, extraction, weighting, preparation and analytical errors) arise as a consequence of the physical interaction between the material being sampled and the technology employed to extract the sample. They result in bias, which can be reduced by the correct application of sampling methods and procedures.

Defining a representative sample A sample can be described as being representative when it results in acceptable levels of bias and precision (Pitard, 2013). Whilst precision (reproducibility) can be determined, bias is difficult to estimate without generally impractical and costly experimental efforts. Danish Standard (DS3077, 2013) recommends that the total sampling variability can be quantified by the relative sampling variance (RSV): the percentage coefficient of variance for repeat sample values. A generalised value of ±20 per cent or less is recommended, where higher values signify poor precision, indicating that the sampling procedure requires improvement. The RSV encompasses all sampling and analytical errors in the entire field-to-aliquot process. RSV measures the total empirical sampling variance influenced by the heterogeneity of the lot being sampled under the current sampling procedure (DS3077, 2013). The RSV comprises all stages of the sampling protocol, including all errors incurred by mass reduction as well as analytical error. The accepted value of RSV is up to the practitioner and based upon the nature of the mineralisation in question, the data quality objectives and what is cost-effective and practical. In minerals sampling, achieving an RSV of ±20 per cent is generally impossible due to the geological (or in situ) nugget effect. However, great effort can be made to minimise the RSV through TOS and QA/QC application. Pitard (2013) states that the total variance for example, resource and grade control sampling should not be more than ±32 per cent, with the FSE component not more than ±16 per cent (Pitard, 2013). All sampling variances are cumulative and contribute to the total, which in turn contributes to the sampling nugget effect. In reality, the FSE and GSE are likely to contribute up to 90 per cent, with delimitation, extraction and analytical errors up to 25 per cent of the total (Pitard, 1993). For an individual field sample to be representative may be impractical. A theoretical field sample mass of tonnes could in rare cases be required to achieve an acceptable RSV. A reasonable strategy is to collect multiple samples across a given domain. Each sample may be locally unrepresentative,

EIGHTH WORLD CONFERENCE ON SAMPLING AND BLENDING / PERTH, WA, 9–11 MAY 2017

SAMPLING AND BLENDING IN GEOENVIRONMENTAL CAMPAIGNS – CURRENT PRACTICE AND FUTURE OPPORTUNITIES

but appropriately spaced samples informing an optimised model will provide a more robust estimate.

Sampling methods Various sample types can be used for geoenvironmental sampling. These range from in situ (eg punctual and linear samples) to grab samples from broken rock piles, and the use of drilling (Table 1). To achieve quality samples, techniques employed should minimise FSE, GSE, delimitation, extraction and weighting errors. A substantive warning is given to the collection of grab samples from stockpiles or waste dumps (including tailings). The method is prone to very high sampling errors, thus making them generally unrepresentative (Dominy, 2010a). Samplers should be well-trained and motivated, and supported by clear written protocols.

Early life-of-mine Current practice Unrepresentative samples lead to poor deposit characterisation and will likely result in inadequate mine planning and potentially financial disaster relating to having to pay longterm liability costs in geoenvironmental cases. Representative samples require consideration of: •• deposit geology (eg heterogeneity and mineralogy) •• the sampling protocols. Rightly or wrongly, the ultimate force driving sampling is project budget. Simple methods for AMD characterisation include the use of sulfur assay data with an assigned cut-off value to determine what is potentially acid forming (PAF), and what is non-acid forming (NAF). However, this value makes the assumption that all sulfur is present as sulfide-sulfur, and further, that all sulfide phases are acid forming in a surficial environment (Dold, 2017). Considering this, is using sulfur assay data from a project database accurate enough to produce an AMD block model (presuming data exists for overburden and waste zones) and initiate the process of waste management planning?

TABLE 1 Sampling types and methods (after Dominy, 2010b). Source In situ rock

Broken rock

Sample type

Sampling method

Punctual

Point chip

Linear

Chip-channel; channel panel

Drill hole

Diamond core; reverse circulation; percussion

Grab

Grab

Optimal split

1

Microbulk/bulk2

1. Via linear splitter; 2. microbulk 1 t in mass.

Often the question is posed by non-ARD (acid rock drainage) practitioners ‘which one test can I do to predict AMD’? The field of AMD prediction is by no means a one-test discipline, therefore the answer is that at least two or more tests must be performed to ensure confidence in the waste classification. However, the financial demand for performing at least two tests can result in fewer samples being selected for testing. For example, the current cost for static geochemical net acid generation testing (NAG) and net acid producing potential (NAPP) is in the order of A$50–$100 per sample, with further testing (ie mineralogical testing) required if ambiguous results are returned (ie the sample falls into an uncertain field; Dold, 2017). Essentially, an AMD appraisal involves sampling through five phases (eg Australian guidelines: Table 2). During the exploration phase, up to ten samples from each key lithology/alteration type are recommended for testing, through to several hundred samples during a prefeasibility study. However, one must question if this is truly enough. For example, Downing and Giroux (1993) collected over 1700 samples to create an AMD waste block model for the Windy Craggy base metal project in Canada. The findings from which resulted in its termination, as management of sulfidic waste that would have been generated made the project uneconomic. Considering the cost of AMD testing, a project of this scale is rarely undertaken, however, if not performed then costly remediation and ongoing management will likely result. Downing (2014) correctly argues that sampling costs should not predetermine the number of samples taken and analysed, but should be dependent on the amount necessary to increase confidence in the data and stage along the mine value chain. Despite knowing the value of Downing’s (1999) advice, in practice, sample numbers more akin to the Australian guidelines are collected. Consequently AMD remains a challenge for many modern operations (Lottermoser, 2017). Another approach to determining sample numbers is the use of a hypothetical sample curve which recommends how many samples is appropriate to characterise each visible geological unit (Figure 1). The curve was applied to recommend how many samples would be required to characterise various deposits for which sampling numbers have been made publicly available (ie given in annual environmental or sustainability reports). For these projects, the estimated number of samples falls short (by between 33 and 345; Table 3). The reason for this related to cost. For example, using the recommended sampling number for the >1000 Mt epithermal-porphyry, the static testing costs are approximately A$140 000 (without considering the cost of sample preparation, QA/QC testing, consultant fees, laboratory infrastructure charges or reporting). Ultimately, committing to a robust AMD prediction program is financially challenging during early life-of-mine stages using current tests. Common sampling problems include (Downing, 2014): •• inappropriate sample type

TABLE 2 Suggested initial numbers of samples and test work (adapted from Australian Government Department of Industry, Tourism and Resources, 2007 in Price, 2009). Phase

Description

Exploration – prospect testing

At least three to five samples should be tested for each key lithology/alteration type.

Exploration – resource definition

At least five to ten samples should be tested for each key lithology/alteration type.

Prefeasibility

Several hundred representative samples of high- and low-grade ore, waste rock and tailings should be collected for geochemical work, sufficient samples to populate a block model with reliable distribution of static test data on ore, waste and wall rock. Kinetic tests should be established for at least one to two representative samples for each key lithology/alteration type.

Feasibility

Continue to refine block model if necessary and conduct sufficient mineralogical test work to cross-check data for key lithologies. If there are insufficient data to assess drainage chemistry and provide a convincing management plan for approval, additional sampling, test work and refinement of block models will be required.


47

A PARBHAKAR-FOX AND S C DOMINY

By performing low-cost tests at, and prior to, stage-one of the GMT approach means that more data can be acquired and analysed and first-pass AMD models improved. Moving onto stage-two, non-acid forming and non-metalliferous samples are not analysed so no further expenses are spent on their characterisation. However, the GMT approach states that the number of samples to be selected should be identical to that used at the site for assaying for resource development (eg one in every 2 or 5 m). In contrast, Price (2009) recommended that for drill core, point samples should be collected at least every 50 m laterally and vertically in order to create a statistically significant waste rock model. Both cases represent either end of the sampling spectrum; collecting and characterising a sample as frequently as every 2 m is potentially overkill, whereas one sample every 50 m may oversimplify the AMD characteristics of heterogeneous mineralisation.

FIG 1 – Hypothetical curve to determine the number of samples required to characterise geological units (Steffen, Robertson and Kirsten, 1989). TABLE 3 Examples of sample numbers used for acid rock drainage testing for various mineral deposit types with best practice values (calculated from the hypothetical sample curve in Steffen, Robertson and Kirsten, 1989, Figure 1). Deposit type, location and year of sampling

Estimated resource (Mt)

Actual sampling

Estimated number of samples required

Epithermal-porphyry, Australia (2008)

200

217

250

Iron oxide copper-gold, Australia (2000)

500

118

300

Porphyry Au, South America (2010)

750

96

350

Porphyry Au-Cu, Australia (2010)

850

188

400

>1000

155

500

Epithermal-porphyry, Asia Pacific (2010)

•• insufficient samples taken to properly characterise the waste material with regard to sample variability •• inappropriate field sample spacing and size for waste material characterisation •• representative subsamples are not obtained due to laboratory sampling procedures that reduce field sample mass at too coarse a particle size •• representative subsample analyses are not obtained because the analytical sample size is too small.

One obvious step to enhance sampling is to use hyperspectral mineralogy and chemical data (eg sulfur) collected from intact drill core or pulps (Fox et al, 2017; Jackson et al, 2017). In practice, this has the potential to enable more accurate and cost-effective mesotextural geoenvironmental domaining, thereby guiding the number of samples to be collected during later stages of the GMT approach. The provision of quality drill core is critical, where high core recoveries are optimal (Annels and Dominy, 2003). There will always be inherent sources of error resulting from intact core analysis, however adopting robust QA/QC is the best approach to minimising these. For a new project in exploration phase, in addition to drill core materials there are several other sample sources including blasthole cuttings, trenches, exploration adits and bulk samples for metallurgical testing (Downing, 2014; Dominy, O’Connor and Xie, 2016; Dominy et al, 2017). Indeed, recognising the value of existing data (or that generated by other mining disciplines) is vital in maximising deposit knowledge. For example, AMD can in theory be calculated from assay if high-quality data exists (Berry et al, 2015; Figure 2). Moving into exploration stages, drilling focuses on orebody definition, therefore additional drilling may be required for detailed waste rock characterisation. Hyperspectral scanning (and sulfur analyses) will assist with geoenvironmental domaining and preliminary sampling. For existing projects looking to extending the life-of-mine via

Sampling for new projects Many emerging technologies have been discussed in the past decade focused on improving mineralogical characterisation and can be applied for geoenvironmental characterisation to improve AMD forecasting. Motivated by this, a three-stage integrated geochemistry-mineralogy-texture (GMT) approach to AMD prediction was proposed by Parbhakar-Fox et al (2011). At the earliest life-of-mine stages, mesotextural grouping is based on visual observations. In future, this may be enhanced by portable technologies such as pXRF (Gazley and Fisher, 2014) and laser-induced breakdown spectroscopy, followed by measurements of total sulfur, paste pH and geoenvironmental logging to domain materials based on their AMD properties. 48

FIG 2 – Proposed framework for geoenvironmental acid and metalliferous drainage (AMD) sampling during early life-of-mine stages. (Notes: 1 – sample mass should be 1 kg; ARDI – acid rock drainage index; NAF – non-acid forming; NM – non-metalliferous.)


SAMPLING AND BLENDING IN GEOENVIRONMENTAL CAMPAIGNS – CURRENT PRACTICE AND FUTURE OPPORTUNITIES

expansion, samples can be obtained from drilling, open pit or underground workings, stockpiles or waste piles.

examine grain size characteristics and mineral associations with acid forming phases.

A nominal recommendation is that a geoenvironmental sample, taken at any point during the mine value chain, should be a minimum of 1 kg. However, the actual required mass is highly dependent upon the heterogeneity of the mineralisation in question. Samples should be collected using hand tools (eg geological pick, shovel, scoop or hand driven auger) with due care given to minimising crosscontamination. Reverse circulation drill chippings should be treated with care. The drilling process liberates gold and/or sulfide particles that are likely to display overgrinding, thus resulting in biased test results.

Sampling of the processed mine waste should also be mandated, with collection of samples from the process plant recommended. Characterisation of tailings produced after beneficiation can be undertaken with a focus on static geochemical testing so as to produce an AMD model of the repository, and in situ mineral chemistry on any remnant sulfide phases so as to ascertain if, for example, economic grade of other metals (ie not that primarily sought) remain. No formal guidelines exist for how many samples should be collected, but it is recommended that at least every time a new zone of the deposit is mined, or the processing methodology changed, several should be obtained. Based on the tails characteristics, the FSE equation can be applied to optimise the required sample mass based on given precision and confidence limits (Gy, 1982; Pitard, 1993).

Drill core or chip samples should be placed into plastic sample bags, however for highly reactive samples (eg framboidal pyrite-bearing material) they should be stored in a cold container/refrigerator to minimise oxidation prior to testing. It is recommended that all post-reaction pulps are stored in a cool-dry place in case testing is required later. Photographs of samples should be collected and an AMD database established with drill hole and sample identifier, sample description, hyperspectral and chemical data (ie calculated AMD from assay) recorded. Sample preparation is critical in a sampling campaign as contamination can occur. It can be performed in the field or laboratory, but either way, must be fully documented. If composite samples are to be used then they should be produced as follows (Downing, 2014): •• Pulps – those used for resource development (or metallurgical testing) assaying can also be used for geoenvironmental analyses, if waste sectors were also assayed with those mineralised. An equal, weighed, amount of each pulp should be split from each sample envelope and then combined forming a composite sample over some specified interval (as determined from the hyperspectral data analysis). From this composite, subsamples can be taken for testing. •• Rejects – samples from assay sample preparation can also be used for geoenvironmental analyses and kinetic and humidity cell test work. Composite samples should represent a given domain and comprise samples from intersections within that domain. If samples require compositing and/or subsampling, then each should be crushed and fed to a rotary sample divider. Samples can be fed individually into the rotary sample divider or in an interleaved fashion (thereby enabling blending). By definition, composites smooth variability and may not show the true picture. Variability samples are individual samples taken across a domain to investigate variability. Where possible, variability samples are preferable. Many static geoenvironmental tests require only a small sampling volume (eg 2.5 g for NAG testing). Therefore, the necessity for repeat triplicate samples and a QA/QC program using a geoenvironmental standard is significant, and must be fully discussed with the analytical laboratory performing the test work. However, a review of current laboratory practices demonstrates that no universal inter-laboratory certified reference materials (CRMs) are used (a handful of laboratories use standards including CANMETs KZK-1 and NBM-1). This results in discrepancies which affect the waste classification. Development and implementation of such CRMs is the next challenge which must be addressed in geoenvironmental test work programs. However, the best validation tool currently available is to perform mineralogical analysis as recommended in the GMT approach (Parbhakar-Fox et al, 2011) either as a bulk analysis (ie X-ray diffractometry) or detailed petrography and scanning electron microscopy to

Mine closure and beyond Rigorous geoenvironmental characterisation is not restricted to early life-of-mine stages, but also applies at the end of mine life and beyond. As stated, there are approximately 50 000 abandoned mine waste sites in Australia, many of which are historic, and relatively small in terms of their footprint. However, rehabilitation (or further economic assessment) of these sites is generally required, but in order to create a prioritisation list, cost-effective sampling strategies for their characterisation are required. Such screening procedures should be designed to capture the average behaviour and environmental impacts of the abandoned site (Smith, Ramsey and Hageman, 2000). Theoretically, the most effective sampling campaigns are those based on rock volume per lithology, with recommended sample numbers given in Table 4. However, often a limitation is that for historic piles, these values (ie rock volume and number of lithologies) are unknown. This limitation manifests in the use of relatively low numbers of samples when performing waste rock characterisation at historic mine sites (Munroe, McLemore and Kyle, 1999; Hammarstrom et al, 2003; Harris, Lottermoser and Duchesne, 2003; Ashley et al, 2004; Akabzaa, Benoeng-Yakubo and Seyire, 2007; Marescotti et al, 2008; Changul et al, 2010). Sampling strategies are not detailed in these examples, with no reference to published guidelines made, suggesting their general absence. However, some guidelines do exist. First, the USEPA (1994) outlined two strategies, one of which (proposed by an unnamed consultancy) stated that eight to 12 samples should be collected from each significant rock type (whereby a significant rock type is presumed to be one to two per cent of the total mine rock volume). The second strategy recommended that one sample (1.5 kg) should be collected per 20 000 t of waste rock, or approximately 50 samples per 1 Mt (USDA Forest Service, 1992). The British Columbia AMD taskforce (Steffen, Robertson and Kirsten, 1989) recommend 25 samples as a minimum should be

TABLE 4 Minimum number of samples collected from each rock/overburden type during sampling (Price, 2009). Mass of each separate rock type (t)

Minimum number of samples
