The Peruvian department of Madre de Dios (MDD) is a flashpoint in the current tension between expanding frontier development in the South American Neotropics, global demand for gold, and conservation of floodplain rainforest ecosystems. Located at the base of the eastern slope of the Peruvian Andes and stretching over 85,000 km2 to the borders of Brazil and Bolivia, MDD contains a diverse variety of forests harboring high levels of endemism which have been a focal point for biodiversity research and conservation for nearly half a century. In addition, MDD harbors large populations of indigenous people in isolation and living traditional lifestyles within communal reserves and protected areas.
However, recent trends across the region threaten both ecological and human communities. Deforestation due to expansive artisanal and small scale gold mining has emerged as a significant and growing driver across the Amazon Basin. In MDD alone, such mining has consumed over 100,000 hectares of forest. Furthermore, mining has led to the expansion of unplanned roads, illegal timber harvest, and widespread contamination of watersheds with high levels of mercury-laden sediment, compromising the integrity of protected areas, indigenous community reserves, and viability of recently-contacted native peoples. Efforts to regulate the mining sector have largely failed, and although recent intervention efforts using the Peruvian military have temporarily displaced operations, long-term outcomes are uncertain. Moderate-resolution remotely sensed data provides basic patterns of deforestation, yet underestimates forest degradation, small-scale disturbance, and dynamics of aquatic sediments. Unfortunately, this information is critical for designing effective remediation strategies and tractable mining regulation.
Our proposed cross-institutional project brings together interdisciplinary expertise to construct a multi-source remotely sensed fusion workflow using a number of novel computer vision, change detection, deep learning, and feature-level and score-level techniques for more effective land cover monitoring in neotropical systems. To reach our goals, we will employ an array of remotely sensed data including satellite and UAV-based optical and radar datasets. Additionally, we will gauge the effectiveness of recent Peruvian federal policy directives to constrain gold mining, thereby providing significant insight into a major international effort to address land cover change from this emergent deforestation driver. Our main objetives are to:
1.Construct an accurate, scalable methodology capable of identifying gradients of intensive land-use change from deforestation and mining through information fusion of moderate and VHR optical, C-band radar backscatter, and UAV RGB data products.
2.Assess the impact of recent Peruvian national policies designed to control illegal mining and reform the artisanal mining sector via our fusion products and techniques, and thereby provide guidance for post-mining mitigation and restoration efforts, demonstrating relevance and applicability across the wider Amazon Basin.
This proposal addresses several requirements emphasized by the solicitation by focusing on a hotspot area, MDD, where deforestation has large areas of rainforest. We develop methods for multi-data fusion for quantifying LCLUC over large areas of change by incorporating sources of moderate optical, synthetic aperture radar, VHR optical, and HSR optical data into our workflow. By investigating deforestation from ASGM, we highlight disturbance that has a significant environmental impact. Finally, in conjunction with our Peruvian partners, CINCIA and the regional government of MDD, we construct a remote sensing-based policy analysis to characterize recent trends in intervention in the mining sector and associated environmental terrestrial and aquatic impacts, thereby ensuring that our project has a significant impact and policy relevance.