Land use and land cover monitoring using remote sensing image time series

• Ano de defesa: 2016
• Autor(a) principal: Victor Wegner Maus
• Orientador(a): Gilberto Câmara, Fernando Manoel Ramos
• Banca de defesa: Dalton de Morisson Valeriano, Maria Isabel Sobral Escada, Daniel de Castro Victoria, Márcio Pupin de Mello
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Instituto Nacional de Pesquisas Espaciais (INPE)
• Programa de Pós-Graduação: Programa de Pós-Graduação do INPE em Ciência do Sistema Terrestre
• País: BR
• Link de acesso: http://urlib.net/sid.inpe.br/mtc-m21b/2016/06.01.14.07

Land system change has a wide range of impacts on Earth system components. Tropical forests in particular have been identified as crucial ecosystems for climate regulation, global biodiversity, and hydrological cycling. The Brazilian Amazon has experienced a high rate of deforestation in the last decade and it is the main source of Brazils anthropogenic CO$_{2}$ emissions. The growing global population will further increase the demand for food and therefore increase the pressure on agricultural systems. High quality, fine resolution, and near-real time land use and land cover monitoring systems play a crucial role in generating information to advance our understanding of human impact on land cover. Earth Observation satellites are the only source that provides a continuous and consistent set of information about the Earth${'}$s land. The current large-scale classification systems such as MODIS Land Cover and GLC 2000 have limitations and their accuracy is not sufficient for land change modeling. Therefore, new techniques for improving land system products are urgently needed. The contribution of this thesis to Earth System Science is threefold. Firstly, the thesis presents a new method for analysis of remote-sensed image time series that improves spatio-temporal land cover data sets and has a substantial potential for contributing to land system change modeling. The developed Time- Weighted Dynamic Time Warping (TWDTW) method is a time-constraint variation of the well-known Dynamic Time Warping (DTW) method, which has in the extensive literature proved to be a robust time series data mining. Secondly, this thesis contributed to open and reproducible science by making the algorithms available for larger audience. TWDTW is implemented in an open source R package called dtwSat available in the Comprehensive R Archive Network (CRAN). Thirdly, this thesis presents an analysis of land cover changes in the Amazon, focusing on the Brazilian state of Mato Grosso that has gone through high rate of deforestation and cropland expansion in the last decade. This study identified and estimated the land cover change using MODIS image time series, contributing to better understand the land dynamics in the Brazilian Amazon. In the study area the pasture is the dominant land use after deforestation, whereas most of the single cropping area comes from pasture, and the cropping system is undergoing intensification from single to double cropping. Moreover, the regenerative secondary forest comes mainly from pasture. The study showed the potential of the TWDTW method for large-scale remote sensing data analysis, which could be extended to other Brazilian biomes to help understand land change in the whole Brazilian territory.