Abstract

The Amazon rainforest, as a biodiversity hotspot and regulator of the earths climate, is one of the most important ecosystems on earth, but has been facing extensive deforestation for decades due to urban growth, agricultural expansion, logging and mining. Mining (and the use of remote sensing methods to detect it) has been relatively understudied in the Amazon compared to the other drivers up until a decade ago, highlighting the importance of current research. The objectives of this study are: To quantify the increase in industrial and artisanal mining and its impact on forest cover in the northern Amazonian country of Suriname between 1997 and 2019; Evaluate the impact of this expansion on the structure (fragmentation) and health (phenology) of the forest; and improve existing remote sensing techniques for mining detection through the development of a pioneer method based on cloud processing and semi-automated mining reclassification. The cloud processing software known as Google Earth Engine (GEE) was used for the initial land use land cover classification of the study area. Landsat 5 and 8 images and the classification and regression trees (C.A.R.T) algorithm were used in this step. The resulting classified maps were fed into the semi-automated re-classification model developed for this study, producing final re-classified output maps, which were used to analyse the expansion of mining and its associated impacts on forest fragmentation and phenology. The proposed method is the first documented method which combines cloud processing with a semi-automated re-classification model, providing a technologically advanced approach capable of rapid and efficient detection of mines. This approach resulted in an 89.5% accuracy of mining detection, and the combination of speed, efficiency, and highly accurate detection outperformed many of the other currently documented methods for mining detection in the Amazon. The results highlighted that mining increased from 69.4km² in 1997 to 431.6km² in 2019, an increase of 522% over 22 years. This growth led directly to 351.9km² of forest loss, 83% of which was due to artisanal mining. This loss of forest led to a 122.8km² reduction in the effective mesh size for the artisanal mine sub-area, compared to a decrease of 83km² for the Industrial mine sub-area. Mining also caused a decrease in the health of the surrounding forest, with the decrease in peak greenness being more pronounced for artisanal mining compared to industrial mining. Recommendations for future research include exploring the use of higher resolution imagery such as Sentinel for better results, as well as the use of microwave data in the classification to combat the issue of extensive cloud cover in the Amazon. The issue of overclassification present in the proposed method can potentially be combated by exploring combinations of different classification algorithms with the reclassification model.