基于LightGBM的盐湖锂浓度遥感反演研究——以西藏扎布耶盐湖北湖为例

青藏高原盐湖中的锂、硼等矿产资源具有极高的经济价值.锂、硼等矿产资源的含量及其空间分布是盐湖观测的重点,对于盐湖资源的开发利用具有重要的指导意义.利用遥感技术开展盐湖观测可以克服传统观测站观测空间上数据不连续、费时费力等缺点,而机器学习等人工智能算法可以快速高效地挖掘遥感数据信息,因此本文基于Landsat-8遥感影像数据,利用LightGBM算法开展西藏扎布耶盐湖北湖锂浓度遥感反演.经过采样点波谱数据的获取、锂浓度LightGBM回归模型的构建、盐湖锂浓度反演一系列的实验步骤,最终的模型评价结果显示平均相对误差为0.053530925,均方根误差为10.2869,卡方为0.867,模型与实测数据拟合程度较高.反演结果表明:整个北湖中,锂浓度最高的是东南部的水域,最低的是中西部的水域;河流和秋里南木泉水群的汇入使得附近的水域锂浓度降低;对于本身位于中部锂浓度低值区的钙华岛来说,岛内泉水的汇入使得附近的水域锂浓度有所升高,这一结果与实际情况较为吻合.通过本文的反演研究,证明了LightGBM机器学习算法用于快速反演盐湖锂浓度的可行性和精确性,同时为其它盐湖矿产资源遥感反演提供了技术启示,也为后续的盐湖资源量评估奠定了基础.

Remote sensing inversion of lithium concentration in salt lake using LightGBM: a case study of northern Zabuye salt lake in Tibet

The mineral resources such as lithium and boron in the saline lakes of Qinghai Tibet Plateau are of great economic value. The content and spatial distribution of lithium, boron and other mineral resources are the focus of salt lake observation, which has important significance for the development and utilization of saline lake resources. Using remote sensing technology to carry out saline lake observation can overcome the shortcomings of traditional observation station, such as discontinuous data, time and labor intensive, whilemachine learning algorithms can quickly and efficiently mine remote sensing data information. Therefore, based on landsat- 8 remote sensing image data, this paper uses LightGBM algorithm to carry out remote sensing inversion of lithium concentration in north lake of Zabuye salt lake in Tibet. After a series of experimental steps, such as the acquisition of spectral data of sampling points, the construction of LightGBM regression model of lithium concentration, and the inversion of lithium concentration in salt lake, the final model evaluation results show that the average relative error is 0. 053530925, the root mean square error is 10. 2869, and the chi square is 0. 867. The model has a high degree of fitting with the measured data. The inversion results show that: in the whole of north lake the highest lithium concentration is in the southeast waters, and the lowest is in the midwest waters; the confluence of rivers and the Qiulinanmu spring group lowers the lithium concentration in the nearby waters; for travertine island, which is located in the low lithium concentration area in the middle, the confluence of springs in the island makes the lithium concentration in the nearby waters higher, which is consistent with the actual situation. Through the inversion research in this paper, the feasibility and accuracy of LightGBM machine learning algorithm for fast inversion of saline lake lithium concentration are proved. At the same time, it provides technical explanation for remote sensing inversion of other salt lake mineral resources, and lays the foundation for subsequent salt lake resource assessment.