Abstract Heavy minerals are typically rare but important components of siliciclastic sediments and rocks. Their abundance, proportions, and variability carry valuable information on source rocks, climatic, environmental and transport conditions between source to sink, and diagenetic processes. They are important for practical purposes such as prospecting for mineral resources or the correlation and interpretation of geologic reservoirs. Despite the extensive use of heavy mineral analysis in sedimentary petrography and quite diverse methods for quantifying heavy mineral assemblages, there has never been a systematic comparison of results obtained by different methods and/or operators. This study provides the first interlaboratory test of heavy mineral analysis. Two synthetic heavy mineral samples were prepared with considerably contrasting compositions intended to resemble natural samples. The contributors were requested to provide (i) metadata describing methods, measurement conditions and experience of the operators and (ii) results tables with mineral species and grain counts. One hundred thirty analyses of the two samples were performed by 67 contributors, encompassing both classical microscopic analyses and data obtained by emerging automated techniques based on electron-beam chemical analysis or Raman spectroscopy. Because relatively low numbers of mineral counts (N) are typical for optical analyses while automated techniques allow for high N, the results vary considerably with respect to the Poisson uncertainty of the counting statistics. Therefore standard methods used in evaluation of round robin tests are not feasible. In our case the ‘true’ compositions of the test samples are not known. Three methods have been applied to determine possible reference values: (i) the initially measured weight percentages, (ii) calculation of grain percentages using estimates of grain volumes and densities, and (iii) the best-match average calculated from the most reliable analyses following multiple pragmatic and robust criteria. The range of these three values is taken as best approximation of the ‘true’ composition. The reported grain percentages were evaluated according to (i) their overall scatter relative to the most likely composition, (ii) the number of identified components that were part of the test samples, (iii) the total amount of mistakenly identified mineral grains that were actually not added to the samples, and (iv) the number of major components, which match the reference values with 95% confidence. Results indicate that the overall comparability of the analyses is reasonable. However, there are several issues with respect to methods and/or operators. Optical methods yield the poorest results with respect to the scatter of the data. This, however, is not considered inherent to the method as demonstrated by a significant number of optical analyses fulfilling the criteria for the best-match average. Training of the operators is thus considered paramount for optical analyses. Electron-beam methods yield satisfactory results, but problems in the identification of polymorphs and the discrimination of chain silicates are evident. Labs refining their electron-beam results by optical analysis practically tackle this issue. Raman methods yield the best results as indicated by the highest number of major components correctly quantified with 95% confidence and the fact that all laboratories and operators fulfil the criteria for the best-match average. However, a number of problems must be solved before the full potential of the automated high-throughput techniques in heavy mineral analysis can be achieved.

中文翻译：

---

重矿物数据的可比性——首次实验室间循环测试

摘要 重矿物通常是硅质碎屑沉积物和岩石中罕见但重要的成分。它们的丰度、比例和变异性携带着有关烃源岩、气候、环境和源汇之间的运输条件以及成岩过程的宝贵信息。它们对于实际用途很重要，例如矿产资源勘探或地质储层的对比和解释。尽管在沉积岩相学中广泛使用重矿物分析，并且量化重矿物组合的方法多种多样，但从未对不同方法和/或操作员获得的结果进行系统比较。这项研究提供了重矿物分析的首次实验室间测试。制备了两种合成重矿物样品，其中的成分对比强烈，旨在模拟天然样品。要求贡献者提供 (i) 描述方法、测量条件和操作员经验的元数据，以及 (ii) 包含矿物种类和颗粒计数的结果表。67 位贡献者对这两个样品进行了 130 次分析，包括经典的显微分析和通过基于电子束化学分析或拉曼光谱的新兴自动化技术获得的数据。由于相对较低数量的矿物计数 (N) 是光学分析的典型特征，而自动化技术允许使用高 N，因此在计数统计的泊松不确定性方面，结果差异很大。因此，用于评估循环测试的标准方法是不可行的。在我们的例子中，测试样品的“真实”成分是未知的。已应用三种方法来确定可能的参考值：(i) 最初测量的重量百分比，(ii) 使用谷物体积和密度的估计值计算谷物百分比，以及 (iii) 根据最可靠的分析计算出的最佳匹配平均值遵循多项务实和稳健的标准。这三个值的范围被视为“真实”成分的最佳近似值。报告的谷物百分比是根据 (i) 它们相对于最可能的成分的整体散布，(ii) 作为测试样品一部分的已识别成分的数量进行评估的，(iii) 实际未添加到样品中的错误识别的矿物颗粒的总量，以及 (iv) 主要成分的数量，与参考值相匹配，置信度为 95%。结果表明分析的总体可比性是合理的。但是，在方法和/或运算符方面存在几个问题。光学方法在数据分散方面产生最差的结果。然而，这不被认为是该方法固有的，如满足最佳匹配平均值标准的大量光学分析所证明的那样。因此，操作员的培训被认为是光学分析的重中之重。电子束方法产生了令人满意的结果，但在多晶型物的鉴定和链状硅酸盐的鉴别方面存在明显问题。实验室通过光学分析改进他们的电子束结果实际上解决了这个问题。拉曼方法产生了最好的结果，如以 95% 的置信度正确量化的主要成分数量最多，以及所有实验室和操作员都满足最佳匹配平均值的标准这一事实表明。然而，在重矿物分析中实现自动化高通量技术的全部潜力之前，必须解决许多问题。