基于机器学习的黑云母成分判别花岗岩成因类型方法研究

doi:10.19657/j.geoscience.1000-8527.2025.001

摘要/Abstract

摘要：

花岗岩成因分类方案的合理性和可操作性仍然备受争议。黑云母作为花岗质岩浆结晶分异过程中重要的镁铁质矿物相,其化学成分能够反映岩浆的物理化学条件和演化历程,这为建立花岗岩成因判别指标提供了关键矿物学依据。本文对来自I型、S型和A型花岗岩的1455个黑云母的主量元素数据进行了统计分析,结果表明:I型花岗岩的黑云母SiO₂、TiO₂、MgO和MnO的含量较高,S型花岗岩的黑云母则较为富含Al₂O₃和Na₂O,A型花岗岩的黑云母FeO^T的含量较高。通过黑云母成分估算表明,I型花岗岩形成于相对高温、低压和高氧逸度环境,而S型花岗岩通常形成于相对高压环境,氧逸度和温度均低于I型花岗岩。I型、S型和A型花岗岩中黑云母的F和Cl含量和富集程度也有明显的区别,其中A型花岗岩中的黑云母F和Cl的含量相对最高。黑云母成分具有判别花岗岩成因类型的巨大潜力,但已有的基于黑云母成分的分类模型仍存在较大不确定性。为了进一步提高黑云母成分对于花岗岩类型的识别可靠程度,本研究对比了基于黑云母离子占位数据应用了PCA、t-SNE、UMAP降维机器学习算法和基于决策树的随机森林(RF)、极端随机树(ERT)、梯度提升决策树(GBDT)、极端梯度提升(XGBoost)、轻量梯度提升(LightGBM)和CatBoost机器学习模型算法,对I型、S型和A型花岗岩成因类型进行判别。结果显示,PCA、t-SNE、UMAP降维算法的判别准确率较低,而基于决策树的机器学习模型能够以94.5%以上的准确率对花岗岩类型进行判别。其中,T.Al、T.Fe³⁺、M.Al、M.Mg和M.Mn等5个阳离子占位是影响机器学习模型分类结果的关键预测因子。基于黑云母成分构建的花岗岩成因类型判别标尺,为解析岩浆源区性质与演化过程提供了定量矿物地球化学约束。

关键词: 黑云母, 机器学习, I型花岗岩, S型花岗岩, A型花岗岩

Abstract:

The chemical composition of biotite can reflect the physical and chemical conditions and evolution process of magma. This study collected the EPMA data of 1455 biotite samples derived from I-type,S-type,and A-type granites to explore their differences and intrinsic association with granite type. The results show that the biotites from I-type granite are relatively rich in SiO₂,TiO₂,MgO,and MnO,while the biotites from S-type granite are relatively rich in Al₂O₃ and Na₂O,and the biotites from A-type granite are relatively rich in FeO^T. Based on the chemistry of biotites from I-type,S-type,and A-type granites,The I-type granite forms in a relatively high temperature,low pressure,and high oxygen fugacity environment. In contrast,the S-type granite usually forms in a relatively high-pressure environment,and the oxygen fugacity and temperature are lower than those of the I-type granite. The F and Cl contents and enrichment extent of biotite from different granite types are also significantly different,with relatively highest F and Cl contents in the biotites from A-type granites. Previous studies have shown that biotite composition has excellent potential to distinguish the genetic types of granites,but the existing classification models based on biotite composition still have significant uncertainty. PCA,t-SNE,UMAP dimensionality reduction methods and decision tree-based random forest (RF),extreme random tree (ERT),gradient boosting decision tree (GBDT),extreme gradient boosting (XGBoost),lightweight gradient boosting (LightGBM)and CatBoost machine learning model algorithms were applied to identify I-type,S-type and A-type granites based on calculated molar proportions of cation assignment in the biotite formula. The results show that PCA,t-SNE,and UMAP dimensionality reduction methods are not ineffective in distinguishing different granite types. In contrast,machine learning models based on decision trees can effectively identify granite types with more than 94.5% accuracy. In the biotite formula,T.Al,T.Fe³⁺,M.Al,M.Mg,and M.Mn are the five key cation assignments that affect the classification of machine-learning models.

Key words: biotite, machine learning, I-type granite, S-type granite, A-type granite

中图分类号:

P571
P581

孙玉洁, 李晓彦, 张超. 基于机器学习的黑云母成分判别花岗岩成因类型方法研究[J]. 现代地质, 2025, 39(03): 523-540.

SUN Yujie, LI Xiaoyan, ZHANG Chao. Machine-learning Based Discrimination of Granite Type Using Biotite Composition[J]. Geoscience, 2025, 39(03): 523-540.

图/表 12

图1 I型、S型和A型花岗岩中黑云母的主量元素(a)和微量元素(b)数据箱形图(图中空心方块代表数据组的平均值)

Fig.1 Box plots of major (a)and minor (b)elements of biotite in I-type,S-type,and A-type granites

图2 云母分类命名图(据Foster等[42])

Fig.2 Classification and nomenclature diagram of mica(According to Foster[42])

图3 基于黑云母成分的花岗岩成因类型判别图 (a)基于黑云母MgO-Al2O3的花岗岩构造成因类型判别图,A: 非造山带碱性花岗岩系;C:造山带钙碱性花岗岩系;P:过铝质花岗岩系(底图据Abdel-Rahman[39]);(b)基于黑云母FeOT-MgO-Al2O3的花岗岩成因类型判别图。虚线为Abdel-Rahman等[39]提出的A型,C型和P型划分线;实线为Gion等[13]提出的I型,S型,和A型花岗岩划分线

Fig.3 Discrimination diagram of granite genetic classification based on biotite composition

图4 I型、S型、A型花岗岩黑云母温度压力分布图 (a)温度压力分布散点图;(b)温度分布图;(c)压力分布图

Fig.4 Temperature and pressure distribution of biotite in I-type,S-type,and A-type granites

图5 黑云母所记录的I型、S型和A型花岗岩岩浆氧逸度条件分布图 (a)氧逸度数据T-log f O 2散点图,其中缓冲剂曲线设定为5 kbar(根据黑云母成分计算得到的平均压力)条件下,NNO据Hueber和Sato[52],MH据Frost[53],FMQ据Myers和Eugster[49]; (b)氧逸度频率分布图; (c)黑云母Fe3+-Fe2+-Mg三角图以及表氧逸度条件范围(底图据Wones 和 Eugster[20])

Fig.5 Distribution diagram of oxygen fugacity conditions for I-type,S-type,and A-type granite magmas recorded by biotite

图6 I型、S型和A型花岗岩黑云母卤素截距值 (a)及逸度(b)分布箱形图(图中星号和空心圆圈为数据点,空心方块代表数据组的平均值)

Fig.6 Box plots displaying the biotite halogen intercept values (a)and halogen fugacity (b)for I-type,S-type,and A-type granites

图7 I型、S型和A型花岗岩黑云母F、Cl含量分布图(a)和三类花岗岩的 l o g ( f H 2 O / f H C l )分布图(b)

Fig.7 Distribution diagram showing the content of F and Cl in biotite from I-type,S-type,and A-type granites (a)and a frequency distribution diagram of l o g ( f H 2 O / f H C l ) for I-type,S-type,and A-type granites (b)

图8 I型、S型和A型花岗岩中黑云母晶格内离子之间的相关关系图解

Fig.8 Correlation diagrams between ions of biotite in I-type granite (a),S-type granite (b),and A-type granite (c)

图9 PCA、UMAP和t-SNE降维可视化图

Fig.9 Visualization of PCA,UMAP,and t-SNE dimensionality reduction techniques

图10 黑云母晶格位置离子数在各种分类模型中的特征权重图

Fig.10 Relative importance of ions of biotite in the classification models

图11 各个机器学习模型在测试集上预测结果的混淆矩阵

Fig.11 Confusion matrices for the prediction results of each machine learning model on the test set

图12 各个分类模型的评价指标比较图

Fig.12 Comparison of four evaluation metrics for each classification models

参考文献 67

[1]	吴福元, 李献华, 杨进辉, 等. 花岗岩成因研究的若干问题[J]. 岩石学报, 2007, 23(6): 1217-1238.
[2]	CHAPPELL B, WHITE A J. Two contrasting granite types[J]. Pac Geol, 1974, 8: 173-174.
[3]	LOISELLE M C, WONES D R. Characteristics and origin of anorogenic granites[M]. Geological society of America, 1979, 11: 468.
[4]	WHITE A J R. Sources of granite magmas[M]. Geological Society of America. 1979, 11: 539.
[5]	BARBARIN B. Granitoids: Main petrogenetic classifications in relation to origin and tectonic setting[J]. Geological Journal, 1990, 25(3/4): 227-238.
[6]	ISHIHARA S. The magnetite-series and ilmenite-series granitic rocks[J]. MiningGeology, 1977, 27(145): 293-305.
[7]	ISHIHARA K, YAMAZAKI A. Analysis of wave-induced liquefaction in seabed deposits of sand[J]. Soils and Foundations, 1984, 24(3): 85-100.
[8]	张旗, 翟明国, 魏春景, 等. 一个新的花岗岩成因分类: 基于变质岩深熔作用理论与大数据的证据[J]. 地学前缘, 2022, 29(4): 319-329. DOI
[9]	CHAPPELL B W. Aluminium saturation in I- and S-type granites and the characterization of fractionated haplogranites[J]. Lithos, 1999, 46(3): 535-551.
[10]	GAO P, ZHENG Y F, ZHAO Z F. Distinction between S-type and peraluminous I-type granites: Zircon versus whole-rock geochemistry[J]. Lithos, 2016, 258: 77-91.
[11]	莫静, 夏小平, 徐健. 锆石中的水:S型和I型花岗岩的新判别标志[C]// 广州: 中国地球科学联合学术年会, 2021.
[12]	ZHONG S H, LIU Y, LI S Z, et al. A machine learning method for distinguishing detrital zircon provenance[J]. Contributions to Mineralogy and Petrology, 2023, 178: 35.
[13]	GION A M, PICCOLI P M, CANDELA P A. Characterization of biotite and amphibole compositions in granites[J]. Contributions to Mineralogy and Petrology, 2022, 177(4): 43.
[14]	LI X Y, ZHANG C. Machine learning thermobarometry for biotite-bearing magmas[J]. Journal of Geophysical Research: Solid Earth, 2022, 127(9): e2022JB024137.
[15]	HENRY D J, DAIGLE N M. Chlorine incorporation into amphibole and biotite in high-grade iron-formations: Interplay between crystallography and metamorphic fluids[J]. American Mineralogist, 2018, 103(1): 55-68.
[16]	UCHIDA E, ENDO S, MAKINO M. Relationship between solidification depth of granitic rocks and formation of hydrothermal ore deposits[J]. Resource Geology, 2007, 57(1): 47-56.
[17]	WONES D. Significance of the assemblage titanite+magnetite+quartz in granitic rocks[J]. American Mineralogist, 1989, 74: 744-749.
[18]	MUNOZ J L, LUDINGTON S D. Fluoride-hydroxyl exchange in biotite[J]. American Journal of Science, 1974, 274(4): 396-413.
[19]	ZHANG C, LI X Y, BEHRENS H, et al. Partitioning of OH-F-Cl between biotite and silicate melt: Experiments and an empirical model[J]. Geochimica et Cosmochimica Acta, 2022, 317: 155-179.
[20]	WONES R D, EUGSTER P H. Stability of biotite: experiment,theory,and application1[J]. American Mineralogist, 1965, 50(9): 1228-1272.
[21]	周永章, 王俊, 左仁广, 等. 地质领域机器学习、深度学习及实现语言[J]. 岩石学报, 2018, 34(11): 3173-3178.
[22]	李晓彦, 张超. 大数据和机器学习在矿物学研究中的应用[J]. 矿物岩石地球化学通报, 2023, 42(2): 253-266,252.
[23]	LI X Y, ZHANG C, BEHRENS H, et al. Calculating biotite formula from electron microprobe analysis data using a machine learning method based on principal components regression[J]. Lithos, 2020, 356: 105371.
[24]	WANG L, SU C, WANG L Q, et al. Refined estimation of Li in mica by a machine learning method[J]. American Mineralogist, 2022, 107(6): 1034-1044.
[25]	HYSTAD G, DOWNS R T, GREW E S, et al. Statistical analysis of mineral diversity and distribution: Earth’s mineralogy is unique[J]. Earth and Planetary Science Letters, 2015, 426: 154-157.
[26]	LI X Y, ZHOU Y Z, WANG J, et al. Contrasting granite metallogeny through the zircon REEcomposition: Perspective from data mining[J]. Applied Geochemistry, 2020, 122: 104758.
[27]	王语, 周永章, 肖凡, 等. 基于成矿条件数值模拟和支持向量机算法的深部成矿预测: 以粤北凡口铅锌矿为例[J]. 大地构造与成矿学, 2020, 44(2): 222-230.
[28]	焦守涛, 周永章, 张旗, 等. 基于GEOROC数据库的全球辉长岩大数据的大地构造环境智能判别研究[J]. 岩石学报, 2018, 34(11): 3189-3194.
[29]	SAHA R, UPADHYAY D, MISHRA B. Discriminating tectonic setting of igneous rocks using biotite major element chemistry-A machine learning approach[J]. Geochemistry,Geophysics,Geosystems, 2021, 22: e2021GC010053.
[30]	SAMADI R, TORABI G, KAWABATA H, et al. Biotite as a petrogenetic discriminator: Chemical insights from igneous,meta-igneous and meta-sedimentary rocks in Iran[J]. Lithos, 2021, 386: 106016.
[31]	LI X Y, ZHANG C, BEHRENS H, et al. On the improvement of calculating biotite formula from EPMA data: Reexamination of the methods of Dymek (1983),Yavuz and Öztas (1997),Li et al. (2020)and reply to the discussion of Baidya and Das[J]. Lithos, 2022, 412: 106403.
[32]	MUSHKIN A, NAVON O, HALICZ L, et al. The petrogenesis of A-type magmas from the amram massif,southern Israel[J]. Journal of Petrology, 2003, 44(5): 815-832.
[33]	HUANG H Q, LI X H, LI W X, et al. Formation of high 18O fayalite-bearing A-type granite by high-temperature melting of granulitic metasedimentary rocks,Southern China[J]. Geology, 2011, 39(10): 903-906.
[34]	ZHAO K D, JIANG S Y, CHEN W F, et al. Zircon U-Pb chronology and elemental and Sr-Nd-Hf isotope geochemistry of two Triassic A-type granites in South China: Implication for petrogenesis and Indosinian transtensional tectonism[J]. Lithos, 2013, 160: 292-306.
[35]	MILLER C F. Are strongly peraluminous magmas derived from pelitic sedimentary sources?[J]. The Journal of Geology, 1985, 93(6): 673-689.
[36]	STEVENS G, VILLAROS A, MOYEN J F. Selective peritectic garnet entrainment as the origin of geochemical diversity in S-type granites[J]. Geology, 2007, 35(1): 9.
[37]	WHALEN J B, CURRIE K L, CHAPPELL B W. A-type granites: Geochemical characteristics,discrimination and petrogenesis[J]. Contributions to Mineralogy and Petrology, 1987, 95(4): 407-419.
[38]	CHAPPELL B W, WHITE A J R. I- and S-type granites in the Lachlan fold belt[J]. Earth and Environmental Science Transactions of the Royal Society of Edinburgh, 1992, 83(1/2): 1-26.
[39]	ABDEL-RAHMAN A M. Nature of biotites from alkaline,calc-alkaline,and peraluminous magmas[J]. Journal of Petrology, 1994, 35(2): 525-541.
[40]	FODEN J, SOSSI P A, WAWRYK C M. Fe isotopes and the contrasting petrogenesis of A-,I- and S-type granite[J]. Lithos, 2015, 212: 32-44.
[41]	周宝全, 孙金凤, 杨进辉. 花岗岩类岩石成因的磷灰石微区地球化学和同位素示踪[J]. 岩石学报, 2022, 38(12): 3853-3867.
[42]	FOSTER M D, NOLAN T B. Interpretation of the composition of trioctahedral micas[J]. Geological Survey Professional Paper, 1960, 354(B): 1-49.
[43]	HENRY D J. The Ti-saturation surface for low-to-medium pressure metapelitic biotites: Implications for geothermometry and Ti-substitution mechanisms[J]. American Mineralogist, 2005, 90(2/3): 316-328.
[44]	SHAO T B, XIA Y, DING X, et al. Zircon saturation model in silicate melts: A review and update[J]. Acta Geochimica, 2020, 39(3): 387-403.
[45]	徐鸿雪, 汪洋. 黑云母温度计能否用于估计花岗质侵入岩的结晶温度?[J]. 岩石矿物学杂志, 2024, 43(1): 37-46.
[46]	PATIÑO DOUCE A E. Generation of metaluminous A-type granites by low-pressure melting of calc-alkaline granitoids[J]. Geology, 1997, 25(8): 743.
[47]	李小伟, 莫宣学, 赵志丹, 等. 关于A型花岗岩判别过程中若干问题的讨论[J]. 地质通报, 2010, 29(增): 278-285.
[48]	WONES D R. Stability of biotite: A Reply[J]. American Mineralogist, 1972, 57(1/2): 316-317.
[49]	MYERS J, EUGSTER H P. The system Fe-Si-O: Oxygen buffer calibrations to 1,500K[J]. Contributions to Mineralogy and Petrology, 1983, 82(1): 75-90.
[50]	CASTRO A, STEPHENS W E. Amphibole-rich polycrystalline clots in calc-alkaline granitic rocks and their enclaves[J]. The Canadian Mineralogist, 1992, 30(4): 1093-1112.
[51]	WINTER J D. An introduction to igneous and metamorphic petrology[M]. Upper Saddle River,N.J.: Prentice Hall, 2001.
[52]	HUEBNER J, SATO M. The oxygen fugacity-temperature relationships of manganese oxide and nickel oxide buffers[J]. American Mineralogist, 1970, 55: 934-952.
[53]	FROST B R. Chapter 1.introduction to oxygen fugacity and its petrologic importance[M]// Oxide Minerals. Lindsley,Berlin: De Gruyter, 1991.
[54]	MUNOZ J L. F-OH and Cl-OH exchange in micas with applications to hydrothermal ore deposits[J]. Reviews in Mineralogy and Geochemistry, 1984, 13(1): 469-493.
[55]	MUNOZ J L. Calculation of HF and HCl fugacities from biotite compositions: Revised equations[J]. Geological Society of America Abstract with Program, 1992, 24: A221.
[56]	PEDREGOSA F, VAROQUAUX G, GRAMFORT A, et al. Scikit-learn: Machine learning in Python[J]. Journal of Machine Learning Research, 2011, 12: 2825-2830.
[57]	MAATEN L V D, HINTON G E. Visualizing Data using t-SNE[J]. Machine Learning Research, 2008, 9: 2579-2605.
[58]	GHOJOGH B, CROWLEY M, KARRAY F, et al. Uniform manifold approximation and projection (UMAP)[M]// Elements of Dimensionality Reduction and Manifold Learning. Cham: Springer International Publishing, 2023.
[59]	BREIMAN L. Bagging predictors[J]. Machine Learning, 1996, 24(2): 123-140.
[60]	ZHANG W S, YU T Z. Research on Boosting theory and its applications[J]. Journal of University of Science and Technology of China, 2016, 46(3): 222-230.
[61]	BREIMAN L. Random Forests[J]. Machine Learning, 2001, 45(1): 5-32.
[62]	GEURTS P, ERNST D, WEHENKEL L. Extremely randomized trees[J]. Machine Learning, 2006, 63(1): 3-42.
[63]	FRIEDMAN J H. Stochastic gradient boosting[J]. Computational Statistics & Data Analysis, 2002, 38(4): 367-378.
[64]	CHEN T Q, GUESTRIN C, CHEN T Q, et al. XGBoost[C]// San Francisco,California,USA. ACM:Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 785-794.
[65]	KE G, MENG Q, FINLEY T, et al. LightGBM: A Highly Efficient Gradient Boosting Decision Tree[C]// State of Ohio: Neural Information Processing Systems, 2017.
[66]	PROKHORENKOVA L, GUSEV G, VOROBEV A, et al. CatBoost: unbiased boosting with categorical features[C]. 32nd Conference on Neural Information Processing Systems (NIPS), 2018.
[67]	ÖZGÜR A, ÖZGÜR L, GÜNGÖR T. Text categorization with class-based and corpus-based keyword selection[M]// Computer and Information Sciences-ISCIS 2005. Berlin,Heidelberg: Springer Berlin Heidelberg, 2005.