Abstract: Aquamarine is aniron-bearing blue beryl, which is the second most important gemstones in the beryl family except for emeralds. In recent years, aquamarine has been increasingly favored by more and more jewelers and consumers, which has become one of the gemstones with the highest price increase, showing significant economic and ornamental value. The origins of gemstones is one of the important factors affecting its market value. Previous studies on aquamarine have predominantly focused on the assessment of colour, gemmological properties, and inclusion compositions from the single source^1-8. However, the comparative researches on the identification features of aquamarines from different origins are still lacking, with only a few non-comprehensive comparisons available for individual sources^9-12. Therefore, the origin identification system for aquamarine is still in the exploratory stage, necessitating more systematic data. The chemical composition is a crucial criterion for origin identification. The genesis of minerals, metallogenic mechanisms, and changesin the formation environment all influence trace elements and their contents. Therefore, the analysis of trace elements of aquamarines from different localities will assist in reconstructing the mineralization process and provide more accurate data for origin determination.In this study, aquamarine (blue beryl) samples from 12 localities were collected, such as Afghanistan, Brazil, Madagascar, Mozambique, Nigeria, Namibia (the Erongo Volcanic Complex), Pakistan (Skardu), Vietnam (NghêAn and Thanh Hoa), and three regions in China: Xuebaoding in Sichuan Province, Yuanyang in Yunnan Province, and Cuonadong in Xizang(Fig. 1). The selected aquamarine samples range in colour from light blue to deep blue, with some showing a faint green hue. Among them, samples from Vietnam and Brazil exhibit high colour saturation and moderate brightness, while those from Sichuan Province, China display only a slight blue tint. Samples from Mozambique have a more greenish-blue colour. The transparency of the aquamarine samples vary from opaque to transparent, mainly semi-transparent to transparent. Notably, all samples from Sichuan Province, China, Mozambique, Brazil, and Vietnam are basically transparent, although those of samples from Mozambique are poor. The crystal size is between 1 cm to 6 cm, with a variety of morphologies, including long columnar, short columnar, plate-like, and compact massive morphologies. The aquamarine samples from different origins show variations in crystal habits.For example, the samples from Vietnam are long columnar; samples from Nigeria, Brazil, Madagascar, Xizang, and Yunnan in China are mainly short columnar; the samples from Sichuan, China are predominantly thick tabular, while Afghanistan samples are tabular, and Mozambique samples are mostly compact massive in shape.The aquamarine samples from different origins are rich in various fluid inclusions.The jagged three-phase fluid inclusions (Fig. 2a) are common in samples from Sichuan Province, China, while short needle-like three-phase fluid inclusions(Fig. 2b) are present in samples from Yunnan Province, China. The thin fluid inclusions with rainbow interference colour (Fig. 2c) can be seen in samples from Xizang, China. The composition of fluid inclusions was analyzed using Raman spectrometer. The liquid phase of the fluid inclusions predominantly consists of H₂O, with CO₂ and minor N₂ in the gaseous part CO₂, being the highest valence state of carbon, indicates that the CO₂ and H₂O components in the fluid inclusions are the final oxidation products. Given that the tested fluid inclusions are morphologically primary, it can be inferred that aquamarine likely forms in a fluid-rich oxidizing environment¹³. There is a large number of tourmaline co-existence in the samples from Afghanistan (Fig. 2d), indicating a high concentration of boron in the mineralizing fluids; the presence of fluorite inclusions in the Namibian samples indicate an enrichment of fluorine in the fluids (Fig. 2e), while the samples from Nigeria contain abundant primary pyrrhotite inclusions, corresponding to a fluid enriched in sulfur (Fig. 2f). These three elements, boron, fluorine and sulfur, are all showing high fluid mobility, further corroborating the formation of aquamarine in a fluid-rich environment.From the characteristics of appearance and inclusions, aquamarine samples from various origins are very similar, and cannot be well distinguished. However, there are certain differences in the trace elements and their contents among different localities.Influenced by the mineralization environment, isomorphous substitution of various elements occurs during the crystallization process of beryl crystals. EPMA and LA-ICP-MS are used to conduct major and trace element analysis on samples from various origins. In the crystal lattice of aquamarine, the Y-site of the Al-O octahedron is unsaturated with Al³⁺. The primary substituting element is Fe, and the charge transfer between divalent and trivalent iron is the main reason for the green-blue hue of beryl. Aquamarines from Thoung Xuan and NghêAn in Vietnam have the highest Fe contents, and their colour exhibit relatively high blue saturation. There is a negative correlation between Be and Li, indicating that Be²⁺ in the Be-O tetrahedron at the T2 site is partially replaced by Li⁺, which is known as "tetrahedral" substitution. The highest Li content is found in aquamarines from Sichuan Province, China, while Namibian Erongo aquamarines have the lowest average Li content, with Li content varying by up to two orders of magnitude between samples. Alkali metal ions (Rf⁺: Na⁺, K⁺, Li⁺, Rb⁺, Cs⁺) are present in the channels of aquamarine to balance the charge. Samples from most localities such as Namibia's Erongo, Madagascar, and Nigeria generally have low alkali metal contents, categorizing them as "low-alkali beryl". In contrast, Rb content in samples from Mozambique is significantly higher than that of other localities. The average alkali metal content in Sichuan Province, China, Afghanistan and Mozambique is far exceeding the total alkali metal content of other localities.Based on the elemental content test data from various origins, we introduced the XGBoost (Extreme Gradient Boosting) algorithm framework to attempt data modeling and training. Through iterative optimization, XGBoost can more accurately capture the complex relationships between trace elements and their geographic origins. During the simulation training, the model achieved a final accuracy of 0.93 and identified previously underappreciated trace elements, such as Yb, Sc, and Mn, as having potential contributions to the provenance identification of aquamarine. Machine learning has the capability to uncover hidden features that traditional statistical analysis might miss, thereby further enhancing the accuracy of aquamarine provenance identification.