Background

Since the 1970s, forests in the Amazon region have been increasingly converted to pasturelands, human settlement areas, and natural resource extraction regions. By 2006, almost 95% of all deforestation in the Brazilian Amazon occurred within 5.5 km of roadways or less than 1 km from navigable rivers with a number of consequences for freshwater ecosystems, including changes in runoff characteristics. One measurable impact on the riparian system is greater discharge and increased sediment flux in the channels of the main rivers, which can be used as a proxy for ecosystem degradation (Coe et al., 2011; Costa et al.; 2003; Latrubesse et al.; 2009).

One means of understanding the impact of deforestation or mitigation through reforestation is by monitoring the sediment flux over time. This understanding of the dynamics of hydro-sedimentological processes in river basins provides critical data for decision-making and supports management planning for the rational use of natural resources. However, due to the logistical difficulties in measuring sediment flux in a distributed way through time and space, there is a significant opportunity to increase monitoring frequency of water bodies through satellite remote sensing.

New artificial intelligence techniques such as convolutional neural networks (CNN) have shown promising performance in the application of object detection and continuous forest structure estimation when combined with remote sensing imagery (Chang et al. 2019). The ability of CNNs to use contextual and other structural information found in imagery can often outperform traditional remote sensing approaches that are limited to spectral analysis alone. CNNs and other machine learning techniques present an opportunity to improve predictions of suspended sediment in river systems using satellite imagery.