Banca de DEFESA: MATEUS RIBEIRO DA SILVA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : MATEUS RIBEIRO DA SILVA
DATE: 27/02/2026
TIME: 14:00
LOCAL: Auditório do CCET
TITLE:

Comparative evaluation between artificial neural networks and linear regression in the estimation of satellite-derived bathymetry: implications for the morphological characterization of the shallow continental shelf adjacent to Caiçara do Norte/RN and São Bento do Norte/RN

KEY WORDS:

Satellite-derived bathymetry; Artificial Neural Networks (ANN); Coastal monitoring

PAGES: 97
BIG AREA: Ciências Exatas e da Terra
AREA: Geociências
SUMMARY:

The high complexity and elevated cost of traditional shallow-water hydrographic survey methods represent a significant challenge for the acquisition of bathymetric data. In this context, this study aimed to compare the performance of Artificial Neural Networks (ANNs) and linear regression, assess the influence of different orbital sensors, and analyze the relationship between water depth and estimation accuracy within a satellite-derived bathymetry (SDB) framework. To this end, imagery from the Landsat 8 and Sentinel-2A sensors was processed, including geometric and radiometric corrections. The ANN models employed Multilayer Perceptron architectures with three hidden layers, while linear regression applied logarithmic transformations to spectral bands. The results demonstrated the superiority of ANNs, which achieved a root mean square error (RMSE) of 4.10 m for Landsat 8, compared to 9.71 m obtained using linear regression. Landsat 8 outperformed Sentinel-2A, contrary to expectations regarding the advantage of higher spatial resolution. Depth-stratified analysis revealed consistent patterns, with minimum errors in very shallow waters (0–25 m) and an exponential increase in errors beyond 40 m, indicating limitations related to light penetration within the water column. These findings are consistent with international SDB studies and highlight challenges in environments characterized by high turbidity or complex benthic substrates. The results indicate that ANNs constitute a viable tool for coastal bathymetric mapping and environmental monitoring, particularly in shallow-water environments, achieving sub-meter accuracy in very shallow depths. In accordance with International Hydrographic Organization (IHO) guidelines, satellite derived bathymetry is not intended to replace conventional hydrographic surveying methods but rather to complement them, especially in regions with limited in situ data availability. Owing to its spectral resolution, Landsat 8 emerges as a priority sensor for SDB applications in regions such as the Potiguar continental shelf, supporting environmental monitoring and offshore wind farm planning. The integration of multisensor data and physics-based adjustments may further reduce uncertainty at greater depths, expanding the applicability of SDB for regional and global mapping and monitoring initiatives

COMMITTEE MEMBERS:
Presidente - 2218779 - HELENICE VITAL
Externo ao Programa - 1858120 - DAVID MENDES - UFRNExterno à Instituição - THIAGO AUGUSTO BEZERRA FERREIRA