Lithium-bearing pegmatite identification, based on spectral analysis and machine learning: a case study of the Chaibeiyuan area, NW China

Lithium (Li)-bearing minerals are common in granitic pegmatite deposits in remote western China, where geological surveys are challenging because of complex terrains and extreme environments. Hyperspectral remote sensing is desirable for providing real-time detailed spectral information to monitor such pegmatite deposits. This study focuses on the northern margin of the Qaidam Basin in Qinghai, China. An ASD spectrometer was used to obtain the spectral characteristics of granitic pegmatite samples, including hand samples, powder samples, and country rocks. The spectral characteristics of granitic pegmatite were analysed, and sensitive bands were selected. When GF-2 and GF-5 images were fused on the basis of sensitive bands, the random forest (RF) method outperformed the support vector machine (SVM) and spectral angle mapper (SAM) methods in identifying pegmatites in three typical deposits: Shaliuquan, Chakabeishan, and Shinaihai. The 23 sensitive bands selected by the competitive adaptive reweighted sampling (CARS) method from GF-5 were more effective for model inversion. Therefore, a three-band spectral index (TBI) was established via these 23 bands. When fused images are uesed as the data source, partial least squares regression (PLS) outperforms the random forest (RF) and support vector machine (SVM) methods in modelling applications. This study provides an effective methodology for rapid monitoring of the Li content in granitic pegmatites, which is crucial for resource exploration in remote and challenging terrains.