GRSG 35th Conference 2024 Abstract

Title: Multispectral Imagery for Bathymetric Derivation in Monitoring Sand Mining Activities

Author: Raul Adriaensen

Organisation: Imperial

Sand mining along riverbeds is a prevalent and legally regulated practice. However, operators sometimes exceed their licensed extraction volumes and hide additional profits, leading to significant economic and environmental consequences. Accurate monitoring of extraction volumes is thus essential, not just for environmental protection but also to ensure fair and appropriate legal repercussions for those who violate their concession volumes.

This study focusses on sand mining activities in India. The first modern methods for estimating mined volumes utilized drone-generated Digital Elevation Models. However, miners quickly caught on and adapted by infilling the pits with water to obscure their illicit activities. This challenge opened an opportunity to apply optical imagery for deriving bathymetry, a technique that estimated water depth based on light penetration.

My Ph.D. research concentrates on leveraging remote sensing data to enhance numerical flow models, particularly for low-lying islands like atolls. A dominant source of error in these models is the lack of accurate bathymetric data, especially in shallow waters. While investigating Satellite Derived Bathymetry (SDB) methods, it became apparent that two significant gaps exist in the field: the absence of an industry benchmark to fairly assess algorithm performance and a shortage of data to test innovative, data-intensive methods. To address these challenges, part of my research has involved developing a toolbox to generate synthetic optical imagery, helping to overcome these hurdles.

Applying this background to sand mining case study aligns well with the conference theme of geological remote sensing and showcases a novel application. The data for this study includes bathymetric measurements from various mining pits, very high resolution drone imagery, and high resolutions satellite imagery from Planet Labs. This dataset enables us to assess the capability of each data source and evaluate the impact of temporal gaps between datasets caused by fluctuations in river elevation over time.

The methodology begins with creating a various training datasets based on different image resolutions/sources and varying preprocessing steps. This might include down sampling the drone imagery to better understand accuracy as a function of resolution. This data is then tested with a variety of bathymetry derivation algorithms allowing us to determine the accuracy associated with each method. Additionally, we study the effect of training data volumes on accuracy to understand the minimum data requirements and identify the point at which additional data yields diminishing returns.

Preliminary results indicate a clear capability to predict bathymetry using high-resolution satellite data. However, it becomes evident that data from different mining sites cannot be combined due to variability in water column properties, which affects optical characteristics and, consequently, the accuracy of bathymetric estimations.

The conclusions from this study will highlight the impact of image resolution on accuracy and the effects of temporal gaps between optical imagery and bathymetry data due to river level fluctuations. We will identify key preprocessing steps that can improve accuracy and determine the minimum and maximum data point requirements for each method, optimizing data acquisition efforts.

Potential extensions of this work include testing data-independent inversion algorithms and validating them against bathymetric data. This is particularly valuable because collecting bathymetric data in certain regions is not only time-consuming but also dangerous, often requiring military escorts due to the presence of organized crime groups involved in illegal mining activities. Additional research could focus on developing algorithms to automatically identify mining locations along riverbeds, enhancing the detection of illegal sand mining activities and making monitoring efforts more comprehensive and efficient.

In summary, this research demonstrates the viability of using multispectral optical imagery for bathymetric estimation in sand mining regions. It addresses critical gaps in remote sensing applications for environmental monitoring and offers practical solutions for safer, more efficient, and accurate data acquisition. Hopefully these findings will motivate future practices and have an impact on regulatory compliance for this precious resource.