Banca de DEFESA: JONAS FELIPE SANTOS DE SOUZA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : JONAS FELIPE SANTOS DE SOUZA
DATE: 21/02/2024
TIME: 14:00
LOCAL: Pós-Graduação Engenharia Civil
TITLE:

Application Of Optical and Radar Sensor Products for Monitoring Reservoirs in Pernambuco

KEY WORDS:

Sentinel-1. Sentinel-2. Water detection. Artificial Neural Network.

PAGES: 80
BIG AREA: Engenharias
AREA: Engenharia Civil
SUMMARY:

This work analyzes the application of remote sensing using radar and optical images from the Sentinel-1 and Sentinel-2 satellites, respectively, for mapping and monitoring the extent of surface water in reservoirs in the state of Pernambuco. Furthermore, a methodology for bias correction of water area data obtained by satellite using Artificial Neural Networks was also proposed and applied. In the first approach, using Sentinel-1 products, six reservoirs were selected in Mata Norte and the Metropolitan Region of Recife for the application and evaluation of a water detection algorithm based on Random Forest and three thresholding methods, these being the predefined threshold method, Otsu method and Kittler-Illingworth method. In the second approach, using Sentinel-2 products, three reservoirs were selected in the Sertão region for the application and evaluation of an unsupervised and non-parametric automatic water detection algorithm. After generating the water masks and calculating the surface water areas in both approaches, the results were compared with two datasets: in situ monitoring and MapBiomas. For the methodology adopted in the first approach, the algorithm based on Random Forest presented the best results, with the area values obtained satisfactorily reflecting the trends of the historical series of in situ monitoring, but with limitations in water detection, with underestimation maximum area values and problems in complex environments. For the methodology adopted in the second approach, the applied algorithm did not achieve satisfactory results in water detection, with the calculated area values underestimating the values obtained from in situ observations. Furthermore, several water masks generated by the algorithm presented gaps in the pixel classification, compromising the final result. The use of MapBiomas as a reference database presented limitations regarding the temporal scale, the classification of water bodies and the underestimation of minimum values of surface water area. Finally, the applied bias correction method proved to be efficient for situations with a sufficient test sample size for training and calibration the Artificial Neural Network model.

COMMITTEE MEMBERS:
Externo à Instituição - GERALD NORBERT SOUZA DA SILVA - UFPB
Presidente - 2130612 - JOSE ROBERTO GONCALVES DE AZEVEDO
Externo à Instituição - WENDSON DE OLIVEIRA SOUZA - UFPI