Developing aeolian sand spectral index using multispectral imagery and machine learning models: A representative case study in Iraq

Abstract

Aeolian and drifting sand are major environmental problems in arid and semi-arid regions such as Iraq, as they pose an increasing risk of land degradation and desertification due to climate change and human activities. A spectral index was proposed and studies to detect aeolian sand and monitor changes in the Najaf-Samawah Field in Iraq region. The research evaluated several machine learning (ML) models performed on a binary classification of Landsat OLI data. Different libraries, including LibSVM, LibLINEAR, and generalized linear model (GLM), were used to develop an accurate spectral aeolian and drifting sand index (DSI) in two forms with a complete equation (DSI-C) and the simplified and reduced equation (DSI-R). The Landsat OLI reflectance values generated 15 Normalized Differences (ND), and nine ML models were implemented. The most highly weighted NDs were selected from the 15 trained NDs. The accuracy assessments with other spectral sand indices showed that the DSI-C and DSI-R have significant accuracy and results in detecting aeolian sand. The study presented a simple methodology product for mapping aeolian sand. Quantitatively, the proposed DSI-R attained an overall accuracy, average Kappa, and average F-score of 93.617 %, 87.233 %, and 93.331 %, respectively.