Abstract
We present an optimal integration of multi-sensor datasets, including Advanced Spaceborne Thermal and Reflection Radiometer (ASTER), Phased Array type L-band Synthetic Aperture Radar (PALSAR), Sentinel-1, and digital elevation model for lithological classification using Machine Learning Models (MLMs). Different input features such as spectral, spectral and transformed spectral, spectral and morphological, spectral and textural, and optimum hybrid features were derived and evaluated to accurately classify different rock types found in the Chhatarpur district (Madhya Pradesh), India using the Support Vector Machine (SVM) and Random Forest (RF). The SVM achieves better classification accuracy and shows less sensitivity to the number of samples used in model development. The optimum hybrid features outperform other input features with an overall accuracy and κ coefficient of 77.78% and 0.74, which is around 15% higher as obtained using ASTER spectral data alone. Thus, the proposed multi-sensor optimal integration approach is recommended for successful lithological classification using MLMs.
Acknowledgement
We are thankful to reviewers for their constructive comments and suggestions to enrich the manuscript.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data sharing agreement
The datasets and R code can be available by communicating with the corresponding author of the manuscript.
Table 3. A brief summary of lithological units, areal extent, number of training, and testing datasets used in lithological classification using ML models.
Table 5. Textural features derived from optimum window size using DEM.
Table 8. The accuracy statistics of lithological classification derived from integrated spectral and morphological input dataset, and integrated spectral and textural input dataset using SVM and RF models.