The Influence of Database Characteristics on the Internal Consistency of Predictive Models of Trace Element Partitioning for Clinopyroxene, Garnet, and Amphibole
The Influence of Database Characteristics on the Internal Consistency of Predictive Models of Trace Element Partitioning for Clinopyroxene, Garnet, and Amphibole
复制标题
DOI：
10.1029/2023gc010876
复制
发表时间：
2023-06
期刊：
影响因子：
3.7
通讯作者：
Erica Cung;G. Ustunisik;A. Wolf;R. Nielsen
中科院分区：
文献类型：
--
作者：
Erica Cung;G. Ustunisik;A. Wolf;R. Nielsen
要理解火成岩材料中产生主要元素和微量元素特征的过程，就需要建立在各种自然条件下微量元素行为的定量模型。这种预测模型是基于通过回归分析校准表达的实验室实验结果。这些表达式的预测准确性取决于测量特定元素的实验次数和测量的分析精度/准确性，以及适应已知依赖关系的模型。在这种模型中很少考虑的一个因素是实验数据中有关成分、压力和温度范围的“覆盖范围”。本研究的目的是评估斜辉石、角闪石和石榴石的分配系数（Di）如何与具有不同替代机制的矿物的各种强度和组成变量相关。我们的结果表明，实验测定的次数，即使是在一组具有系统行为的元素中（例如，REE），也可能变化多达五倍。此外，每种元素的实验数据库的平均组成、温度和压力存在显著差异。此外，数据库差异和每个元素的分析精度的结合导致控制参数的大小存在系统差异。所有这些因素都会影响我们所依赖的回归的预测能力，并可能在预测行为中产生偏差，这可能与分析误差或实验的平均组成有关。
Understanding the processes that produce the major and trace element signature of igneous materials requires quantitative models of the behavior of trace elements under the full range of natural conditions. Such predictive models are based on the results of laboratory experiments used to calibrate expressions via regression analysis. The predictive accuracy of those expressions depends on the number of experiments where a specific element was measured and the analytical precision/accuracy of the measurements, together with models that accommodate the known dependencies. A factor that has rarely been considered in such models is the “coverage” with respect to the range of composition, pressure, and temperature in the experimental data. The goal of this research is to evaluate how partition coefficients (Di) for clinopyroxene, amphibole, and garnet correlate with a variety of intensive and compositional variables for minerals with different substitution mechanisms. Our results show that the number of experimental determinations, even within a group of elements that behave systematically (e.g., REE), may vary by as much as a factor of five. Further, there are significant differences in the average composition, temperature, and pressure of the experimental database for each element. In addition, the combination of database differences and analytical precision for each element result in systematic differences in the magnitude of the controlling parameters. All of these factors impact the predictive power of the regressions on which we rely and can produce a bias in the predicted behavior that may be correlated with analytical error or average composition of the experiments.