GRSG Conference 2020 Presentation

Title: The application of synthetic aperture radar for the detection and mapping of small-scale mining in Ghana

Author: Gabrielle Denner

Artisanal and small-scale mining (SSM) is a cause of major environmental concern in developing countries. In Ghana, SSM is a mixture of legal and illegal operations where illegal mining is referred to as “galamsey”.

Earth observation techniques can assist local governments in regulating SSM activities by providing specific spatial information on the whereabouts of SSM mines.

The tropical climate in Ghana, however, hinders the regular flow of useful optical imagery due to a high percentage of cloud cover for most parts of the year. Synthetic aperture radar (SAR) can overcome this limitation.

The study area includes a portion of the Ofin River near the mining town of Obuasi, Ghana. The area is tropical in climate, rural and dominated by forests.

This study aims to assess the accuracy and reliability of applying SAR for the detection and mapping of small-scale mining in Ghana with classification and change detection analysis. A literature review on remote sensing and image processing literature was conducted.

The satellite imagery collected for the study included single-date C-band Sentinel-1, a time series of Sentinel-1 and a single-date X-band KompSAT-5 image for the SAR analysis with Sentinel-2 and Landsat-8 imagery as ground truth datasets.

Classification analysis was conducted in two experiments which included the analysis of two classification schemes, i.e. multi-class- and a binary-water classification scheme. The first experiment assessed the accuracy of random forest classification applied to single-date Sentinel-1, KompSAT-5 and multi-temporally filtered Sentinel-1 databases.

The second experiment was a comparison of five machine learning supervised classification methods applied to the multi-temporally filtered Sentinel-1 database. The potential of change detection on Sentinel-1 time series data was analysed in the third experiment for the detection of SSM. Image differencing was applied and two threshold methods were tested for producing the most accurate change maps.

The classification with the object-based image analysis approach was successful in classifying water bodies associated with SSM. The multi-temporally filtered Sentinel-1 dataset was the most reliable with kappa coefficients at 0.65 and 0.82 for the multi-class classification scheme and binary-water classification scheme respectively.

The single-date Sentinel-1 dataset has the highest overall accuracy at 90.93% for the binary water classification scheme.

The KompSAT-5 dataset only achieved the lowest accuracy at an overall accuracy of 80.61% and a kappa coefficient of 0.61 for the binary-water classification scheme.

The results of the change detection analysis indicated that the Sentinel-1 imagery was able to detect and map SSM. The change detection analysis also showed the potential of discerning active from abandoned mines, but this has to be further investigated.

In conclusion, SAR can detect illegal mining activity in tropical areas such as Ghana when focussing on the SSM activities surrounding rivers and the use of high-resolution commercial imagery is not necessary.

The change detection analysis detected SSM where the classification methods only detected water bodies associated with SSM. Further research includes exploring the use of thresholding for the binary-water classification analysis, refining the change detection technique by applying segmentation and machine learning to create the change maps.