Journal
JOURNAL OF GEOCHEMICAL EXPLORATION
Volume 232, Issue -, Pages -Publisher
ELSEVIER
DOI: 10.1016/j.gexplo.2021.106888
Keywords
Lithological mapping; Stream sediment geochemical data; Random forest; Support vector machine
Categories
Funding
- National Key Research and Development Program of China [2018YFE0204204]
- National Natural Science Foundation of China [41702075, 42050103]
- Chinese Geological Survey project [DD20190459]
- Fundamental Research Funds for the Central Universities [2652018132]
Ask authors/readers for more resources
This study successfully applied random forest and support vector machine methods to delineate basalt in the Jining region of Inner Mongolia, China, by collecting field lithological points and integrating stream sediment geochemical data. The evaluation showed that random forest outperformed support vector machine. Using lithological points as data labels and trace-element stream sediment data as a training dataset showed promising results for lithological mapping in covered areas.
Multidisciplinary exploration data have been widely and successfully applied when using machine learning methods to conduct geological mapping. However, in covered areas such as Jining, Inner Mongolia, China, where remote sensing and geophysical data are unavailable or difficult to obtain, geochemical data become more important. In addition, previous studies have often selected data labels based on geological maps, which are generally obtained by interpolation or extrapolation of field lithological points and so are inherently uncertain. This study collected seven types of 2341 field lithological points and evaluated the errors of each lithological unit, based on a confusion matrix. Using these field lithological points, we applied the random forest (RF) and support vector machine (SVM) methods to delineate basalt in the Jining region by integrating 1:50,000 stream sediment geochemical data. The evaluation indexes of accuracy, precision, recall, and the receiver operating characteristic curve (ROC) all indicated that RF outperformed SVM. Based on the predictions of RF, five types of target areas were generated, which were further verified using Sentinel-2 images. This research highlights that using lithological points as data labels and trace-element stream sediment data as a training dataset can provide encouraging results when conducting lithological mapping in covered areas.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available