Algorithm development for characterizing coastal and inland aquatic environments using satellite remote sensing

Coastal and inland aquatic environments are among the most ecologically rich and socioeconomically vital systems on Earth, playing a crucial role in sustaining ecological balance, supporting biodiversity, and enabling socio-economic activities. However, they are increasingly vulnerable to the compounded effects of climate change and anthropogenic pressures, which alter water quality, disrupt habitat integrity, and threaten public health. These transformations occur across broad spatial and temporal scales, demanding integrated, scalable, and timely monitoring systems. Satellite remote sensing has emerged as a promising tool for large-scale observation, yet its operational use is limited by challenges in algorithm robustness and transferability, processing scalability, and the ability to monitor complex aquatic indicators in real time. This thesis contributes to the advancement of Earth Observation (EO) technologies for aquatic monitoring by developing advanced remote sensing algorithms, integrating empirical, machine-learning/deep learning approaches, and implementing automated and scalable processing workflows. Conducted under the SIMBAD (Sentinel Imagery Multiband Analysis and Dissemination) R&D initiative at Quasar Science Resources, the research presented in this dissertation focuses on enhancing the application and operational capacity of Sentinel-2 imagery for monitoring turbidity, microbial contamination, and benthic habitats. The core of the thesis lies in four chapters (Chapters 2-5), each addressing a key challenge in aquatic system monitoring and management. Chapter 2 presents the development of a regional empirical model specifically designed for the Guadalquivir Estuary, one of Spain's most dynamic and turbid estuarine environments. This multi-conditional turbidity model addresses the limitations of existing approaches in highly variable systems, offering robust performance across a wide turbidity range (0-600 FNU) with a strong correlation (r = 0.97) and low error metrics (RMSE = 15.93 FNU, MAE = 13.82 FNU, and Bias = 13.34 FNU), thereby offering a practical solution for local environmental monitoring. Complementing this site-specific approach, the second part of the chapter introduces a globally adaptable turbidity model based on a gradient boosting machine-learning (ML) approach, trained on extensive datasets from 17 countries encompassing lakes, rivers, estuaries, and coastal waters. This ML model offers strong predictive capacity across diverse aquatic environments, with an overall r-value of 0.95, MAE of 43. 24 FNU, and minimal bias (1.32 FNU) across a turbidity range of 0-2,200 FNU. While the site-specific turbidity model achieves superior accuracy for local applications, the global model also provides high accuracy and a robust, scalable solution, especially at extreme turbidity levels (> 1,000 FNU), filling a critical gap in remote sensing models for extreme events. Chapter 3 explores an innovative use of Sentinel-2 data to monitor faecal indicator bacteria (FIB), specifically E. coli and Enterococcus, in recreational coastal waters. This is the first study to demonstrate the feasibility of Sentinel-2 bands for microbial pollution monitoring, offering an independent and complementary tool to conventional water quality indicators such as turbidity, chlorophyll-a, and coloured dissolved organic matter. With strong correlations for E. coli (r = 0.94) and Enterococcus (r = 0.96), the research highlights the potential of integrating satellite-based methods with existing regulatory frameworks to improve waterborne contamination management. Chapter 4 focuses on mapping Posidonia oceanica, an endemic and endangered seagrass species crucial to Mediterranean ecosystems. Using deep-learning algorithms and high-resolution Sentinel-2 imagery, a robust methodology was developed for scalable habitat mapping and change detection across the Balearic and Maltese Islands (74-92% accuracy). The model's effectiveness, even in the data-scarce regions, makes it a valuable tool for assessing this vital seagrass species across the Mediterranean basin. Chapter 5 addresses the technical challenges of automating and scaling EO-based monitoring. By modularizing and dockerizing the developed algorithms within a Scientific Exploitation Platform, the thesis delivers scientific and operational processing pipelines that ensure reproducibility, computational efficiency, and real-time applicability. This infrastructure supports the integration of satellite data with in-situ observations for developing tailored EO solutions and facilitates large-scale, automated water quality and habitat monitoring for environmental assessments. The final chapters of the thesis provide a general discussion of the findings, positioning them within the broader context of aquatic remote sensing, and introduce ongoing and future research initiatives. Together, these efforts reinforce the thesis's contribution to scalable, interdisciplinary, and actionable solutions for aquatic environmental monitoring in support of both scientific advancement and operational management.

Author dc:creator

• Chowdhury, Masuma

Advisors dc:contributor.advisor

• Laiz Alonso, Irene María
• Perez de la Calle, Ignacio

Statement dc:rights

• Attribution-NonCommercial-NoDerivatives 4.0 Internacional

Licence dc:rights.uri

• http://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/10498/38179

OAI identifier oai:identifier

oai:rodin.uca.es:10498/38179