Using remote sensing data for geological mapping in semi-arid environment: a machine learning approach

The geological map encapsulates basic information that can be crucial in a multitude of fields such as landslide risk assessment, engineering projects, as well as petroleum and mineral resources studies. In addition, it is difficult, expensive and time-consuming to achieve it in complex and inaccessible lands. However, remote sensing data linking and the application of Machine Learning Algorithms (MLAs) can be interesting for geological mapping of large areas, especially in arid and semi-arid regions, where remote sensing provides a diversified and detailed spatial database and MLAs offer the possibility of effective and efficient classification of remotely sensed images. This article highlights the use of Aster spectral data in a comparative approach of the performance of six (MLAs) to better produce the geological map of a portion of the Aït Ahmane region. The results indicated an overall Accuracy and a kappa coefficient that exceeded 60% for the different models. Prioritizing the Regularized Discriminant Analysis (RDA) (Kappa = 83.5%) and Support Vector Machines (SVM) (Kappa = 81%) algorithms, they managed to classify the lithology on Aster images of the region. However, the classification of lithology using the RDA was slightly more accurate than the one obtained by SVM with 2.3%. From the results shown, we can conclude that the ability of RDA as a learning algorithm is the best for the geological mapping of our study site.