Automated mapping of regolith units with support vector machine and artificial neural network using data from Landsat-8 OLI, ALOS PALSAR DEM, and Sentinel-1A radar images: the case of the Sissingué Gold Project, Northern Côte d’Ivoire

ABSTRACT

The Sissingué Gold Project is in Northern Côte d’Ivoire and part of the southern extension, a large peneplain of the landscape in South Mali. The area has experienced extensive weathering and erosion over a long time, resulting in a widespread and complex regolith cover. This diverse and vast regolith cover poses considerable difficulties for exploration. To overcome this problem, we adopted mapping of regolith units using machine learning (ML) with support vector machine (SVM) and artificial neural network (ANN) algorithms. The main objectives were to (1) make the predictive regolith map of the study region using these algorithms with data obtained from Landsat-8 operational land imager, Advanced Land Observing Satellite Phased Array Type L-Band Synthetic Aperture Radar digital elevation model and Sentinel-1A; (2) show the method of pre-processing and processing the required data and (3) develop the regolith landform unit (RLU) map from the predictive regolith map and collected data based on the Relict – Erosional – Depositional model. Tests using the SVM and ANN algorithms showed that both ML tools could accurately map regolith landforms. An innovative method of optimizing data using parameters is presented. The results showed that ANN outperformed SVM with an overall accuracy of 87.01% and a kappa coefficient of 0.84, whereas the corresponding values for SVM were 86.69% and 0.84, respectively. However, the validation data obtained from SVM-based prediction exhibited a better score than that of validation data obtained from ANN-based prediction. The Sentinel-1A radar band combined with Landsat-8 data reduced the vegetation-masking effect and improved the classification results. The RLUs of this area are composed primarily of relicts at 24.91%, including lateritic residuum and soil, depositional material at 36.39% with exotic sediments and ferrierite and the remaining 38.7% of erosional material made of saprolite, colluvial fragments, and mottled zones on flanks.

KEYWORDS:

• Support vector machines
• neural networks
• remote sensing

Acknowledgements

This work was supported by the support of CURAT (University Centre for Research and Application in Remote Sensing). This study benefited from the Support of SEMS exploration and three anonymous reviewers.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author, H.A.D., upon reasonable request.