Banca de DEFESA: MARCONE CORREIA DE OLIVEIRA LIMA FILHO

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : MARCONE CORREIA DE OLIVEIRA LIMA FILHO
DATE: 27/04/2022
TIME: 14:00
LOCAL: Google Meet
TITLE:

Performance evaluation of semi-empirical models for remote estimation of CDOM concentration in a tropical productive estuary

KEY WORDS:

CDOM; REMOTE SENSING; ESTUARIAN SYSTEMS

PAGES: 39
BIG AREA: Engenharias
AREA: Engenharia Sanitária
SUBÁREA: Recursos Hídricos
SPECIALTY: Tecnologia e Problemas Sanitários de Irrigação
SUMMARY:

Colored dissolved organic matter (CDOM) is largely responsible for the absorption of sunlight and affects the photobiological and ecological processes of aquatic systems. Estuarine systems are susceptible to high concentrations due to terrestrial carbon coming from the watershed. Thus, there is a need for regular monitoring of this parameter in these aquatic systems aiming at an adequate ecosystem operation. The use of remote sensing is an excellent alternative due to its ability to monitor water bodies at a low cost, in wide areas and adequate time intervals. However, there is still a need for more robust algorithms for estimating CDOM in optically complex systems such as estuarine systems. In this work, we sought to develop semi-empirical models for remote estimation of CDOM in a productive tropical estuarine-lagoon system. For this, data collected in situ of spectral reflectance and measurements of CDOM concentration were used. From this dataset, semi-empirical models based on the ratio of two, three and four bands were developed. The search for bands was performed automatically by a genetic algorithm. Simple linear regression, Support Vector Machine (SVM) and the RANSAC robust linear regression method were evaluated. Several band ratios were tested and the best fit was obtained using a two-band model, R(702)/R(539), and three bands, [R(539)-1-R(699)-1]xR(716 ), with R2 of 0.917 and 0.923, respectively. It was observed that the choice of bands plays a critical role in the development of good models for estimating CDOM, and the use of a genetic algorithm allows automating the task. Furthermore, the use of two properly selected bands is sufficient to obtain good estimates. Finally, the application of robust linear regression proved to deal well with data from real sensors, as it is more tolerant of the presence of outliers.

BANKING MEMBERS:
Presidente - 1718531 - CARLOS RUBERTO FRAGOSO JUNIOR
Interna - 1272239 - DANIELE VITAL VICH
Externa ao Programa - 3242909 - REGINA CAMARA LINS
Externa à Instituição - JOSICLEDA DOMICIANO GALVINCIO - UFPE