Sand Dunes Spectral Index Determination Using Machine Learning Model: Case study of Baiji Sand Dunes Field Northern Iraq

Ehsan Ali Al-Zubaidi; Ahmed H. Al-Sulttani; Furkan Rabee

doi:10.46717/igj.55.1F.9Ms-2022-06-24

Ehsan Ali Al-Zubaidi Department of Computer Science, Faculty of Computer Science and Mathematics, University of Kufa, Iraq: Department of Environmental Planning, Faculty of Physical Planning, University of Kufa, Iraq.
Ahmed H. Al-Sulttani Department of Environmental Planning, Faculty of Physical Planning, University of Kufa, Iraq. https://orcid.org/0000-0001-5251-7755
Furkan Rabee Department of Computer Science, Faculty of Computer Science and Mathematics, University of Kufa, Iraq.

DOI: https://doi.org/10.46717/igj.55.1F.9Ms-2022-06-24

Keywords: Spectral index, Sand dunes, SVM, DSI, Remote sensing, Multispectral

Abstract

Monitoring the propagation of dunes is essential for natural hazard management. Accurate dunes mapping is critical in this situation. Landscape elements such as vegetation, water, dunes, and built-up are commonly separated using spectral indices. The discovery of dune features using a spectral index is one of the most significant achievements in earth observation. In this research, it was suggested Drifting Sand Index (DSI) is a newly created index that can be used to extract the land of dunes. The DSI is calculated using the normalized difference between six Landsat-8 bands (B, R, G, NIR, SWIR-1, SWIR-2). The linear SVM algorithm was implemented using library (LibLINEAR) in the R software to calculate a new linear equation. Two versions of the index have been proposed, the first is the complete version (DSI-C), and the second is the reduced version (DSI-R). The suggested indices results were compared to four previously proposed spectral indices (NDSI-1, NDSI-2, CI, and NDSLI). The acquired results demonstrated that the DSI-C and DSI-R had a high ability to distinguish between sand and different land covers, such as vegetation, water bodies, and various soil types. The average overall accuracy for all levels of the DSI-R, DSI-C, NDSI-1, NDSI-2, CI, and NDSLI was 88.59%, 83.43%, 78.030%, 68.52%, 65.98%, and 56.490%, respectively. The average Kappa Coefficient for DSI-R, DSI-C NDSI-1, NDSI-2, CI, and NDSLI was 77.20%, 66.87%, 56.076%, 37.073%, 31.978%, and 13.011%, respectively.