Supplementary Tables for "A machine learning approach to single garnet geothermometry and application to tracing the fingerprint of superdeep diamonds"

Supporting Information forA machine learning approach to single garnet geothermometry and application to tracing the fingerprint of superdeep diamonds

Table S1 Major and minor element compositions of olivine-orthopyroxene-clinopyroxene-garnet from well-equilibrated peridotites (N = 1308).

Table S2 Calculation spreadsheet for Mn-in-garnet thermometer calibrated in this study, assuming a fixed olivine composition. The example data are the garnets from the well-equilibrated peridotite dataset.

Table S3 Key hyperparameters used for ML model determined by Bayesian optimisation approach.

Table S4 Results of 200 times 5-fold cross validation of the XGBoost model using a training set with TTA98 ranging from 700 to 1450 °C and with TTA98 ranging from 900 to 1400 °C.

Table S5 Normalised feature importance after 1000 runs of the XGBoost model using a training set with TTA98 ranging from 700 to 1450 °C and with TTA98 ranging from 900 to 1400 °C.

Table S6 Comparison of temperatures calculated by the Ni-in-garnet thermometer (TNM24, Nimis et al. (2024)), the Mn-in-garnet thermometer, and the ML-based garnet thermometer.

Table S7 Percentage of garnet types from 25 kimberlites across the Slave Craton and Kaapvaal Craton.

Table S8 Major and minor element compositions and temperatures calculated by ML-based thermometer of garnet xenocrysts from the Slave and Kaapvaal Craton

本研究配套支撑材料：基于单颗石榴石地质温标的机器学习方法及其在示踪超深金刚石指纹特征中的应用
表S1 平衡良好橄榄岩中橄榄石-斜方辉石-单斜辉石-石榴石的主量与微量成分（样本量N=1308）
表S2 本研究标定的石榴石含锰地质温标的计算表格，假设橄榄石成分固定。示例数据取自上述平衡良好的橄榄岩数据集内的石榴石样本
表S3 经贝叶斯优化方法确定的机器学习（Machine Learning）模型所用关键超参数
表S4 当训练集的TTA98范围分别为700~1450℃与900~1400℃时，XGBoost模型经200次5折交叉验证所得的结果
表S5 当训练集的TTA98范围分别为700~1450℃与900~1400℃时，XGBoost模型经1000次迭代后的归一化特征重要性
表S6 分别基于含镍石榴石地质温标（TNM24，引自Nimis等（2024））、含锰石榴石地质温标以及基于机器学习的石榴石地质温标所计算得到的温度对比结果
表S7 斯莱特克拉通与卡普瓦尔克拉通共25处金伯利岩的石榴石类型占比统计
表S8 取自斯莱特克拉通与卡普瓦尔克拉通的石榴石捕虏晶的主量与微量成分，以及基于机器学习温标计算所得的温度数据