Machine learning approaches for mapping silica sand deposits using spaceborne remote sensing

The Alappuzha district in Kerala, India, has large deposits of high-grade silica sand mainly used for glass manufacturing. In the present study, these mineral deposits were effectively mapped using multispectral satellite data and machine learning algorithms (MLAs). The samples collected from Cherthala show a silica sand content of 93.93-97.94%. Detailed geochemical characterisation using the Energy Dispersive X-ray Fluorescence (ED-XRF) technique reveals a SiO₂ content of 96.93-99.13%. The X-ray diffraction (XRD) analysis reveals diffraction peaks characteristic of quartz, confirming the silica sand's composition. The spectral signatures of silica sand were captured using an ASD Fieldspec® 3 spectroradiometer (spectral range of 400–2500 nm) and compiled as a reference spectrum for mapping using Landsat and ASTER remote sensing datasets. The reference spectra of silica sand were used to generate the potential targets of silica sand occurrences, followed by a comparison of four widely used MLAs, which shows that the Support vector machine (SVM) outperforms other algorithms, such as the Random Forest Classifier (RFC), maximum likelihood classification (MLC), and artificial neural network (ANN), with an overall accuracy of 97.82% and Kappa coefficient of 0.96. The results derived from the satellite data show a good correlation with ground truth verification and laboratory analysis. Integrating remote sensing techniques, mineral characterization, and field data facilitates eco-friendly and sustainable mining of these strategic minerals.