Criação e avaliação de modelos de prognóstico futuro de linhas de costa, utilizando regressão estatística e redes neurais artificiais, a partir das séries temporais de imagens de satélite

• Ano de defesa: 2022
• Autor(a) principal: Sousa, Willamys Rangel Nunes de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Geoprocessamento; Sensoriamento Remoto; Erosão costeira
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/69728

Coastal erosion is the process of material removal from the shoreline that results in the loss of land as it recedes towards the territory. This process causes loss of properties, infrastructure, biodiversity, in addition to generating, annually, great economic impacts. It has become a global problem and therefore the analysis and monitoring of such impacts is an issue that deserves adequate attention. For this purpose, Remote Sensing data have been widely used in various studies that assess spatial and temporal change in land use and land cover. Furthermore, the use of time series of satellite images applied to the investigation of land cover change and its spatio- temporal pattern has been proven to be an extremely efficient approach for studies on coastal monitoring. In this context, the main objective of this work was to create and evaluate prognostic models to generate future scenarios, based on the analysis of spatio-temporal changes of coastlines extracted from orbital images in an automated way, from 1985 to 2018. To achieve this goal, two models were proposed, the first using linear regression techniques and the second applying artificial neural networks to create a predictive model of future scenarios, implemented in Python language, using the Numpy, OpenCV, Sklearn and Keras libraries. In addition to these, Digital Image Processing techniques and the extraction of the Modified Normalized Difference Water Index (MNDWI) were used in Remote Sensing images. As a result, an algorithm for automatic extraction of shorelines from MNDWI images was implemented. In addition, a coastal erosion prognosis model was generated for the year 2021, based on the time series from 1985 to 2015. Finally, a hybrid architecture of artificial neural networks was proposed, composed of a convolutional neural network layer and another Long Short-Term Memory layer, to estimate future scenarios of sectors of the coastline, in the city of Icapuí-CE. For this, a time series from 1988 to 2018 was used and MNDWIs were generated for the years 2003, 2008, 2013 and 2018, in addition to the generation of the prognosis for the year 2023. In addition, a comparison was made between the lines automatically generated coastlines and the lines extracted by a photo interpreter, using the Geographic Information System