Application of Object Based Image Analysis (OBIA) in detecting and quantifying forest loss caused by artisanal gold mining activities in Upper Mazaruni River Basin, Guyana

Abstract:

Artisanal gold mining in Guyana is mostly done in forest covered areas, causing forest loss and land degradation. Data from the Guyana Geology and Mining commission show there has been an increase of 2074% between 1986 and 2009. Our analysis of Landsat data between 1986 and 2013 for a part of the Upper Mazaruni area shows an increase from 34.2 to 278.4 hectares, which amounts to more than 800%. While the frequent coverage of Landsat data is useful for multitemporal studies, the lower resolution may not be adequate for accurate detection of mining sites. Therefore, RapidEye imagery from 2011 at a resolution of 5m was used to detect gold mining activity and to compare the results with the Landsat study. Processing was conducted in eCognition, an object-based image analysis (OBIA) software. OBIA is an image processing technique that has proven to be advantageous over traditional pixel based image processing techniques, with the primary advantage being the ability of the approach in combining both the spatial and spectral information. The satellite image was subjected to segmentation at multiple scales and classified using fuzzy sets of membership functions. Classification explicitly incorporated the different scales in order to accommodate different sizes of real-world objects and spatial relationships were utilized to establish connections between related objects. For example the presence or absence of water in pits, or the existence of sediments in the river may serve as additional indicators of mining sites besides the spectral components. Preliminary results show that OBIA approach was able to successfully detect and quantify small scale mining activities in the basin, and that the Landsat data were giving an acceptable estimate of mining sites over time.

 

Keywords:

Object Based Image Analysis, Gold Mining, Remote Sensing, Guyana