A carregar...
Publicação
The contribution of multitemporal information from multispectral satellite images for automatic land cover classification at the national scale
Autores
Resumo(s)
Imaging and sensing technologies are constantly evolving so that, now, the latest
generations of satellites commonly provide with Earth’s surface snapshots at very short
sampling periods (i.e. daily images). It is unquestionable that this tendency towards
continuous time observation will broaden up the scope of remotely sensed activities.
Inevitable also, such increasing amount of information will prompt methodological
approaches that combine digital image processing techniques with time series analysis for
the characterization of land cover distribution and monitoring of its dynamics on a frequent
basis. Nonetheless, quantitative analyses that convey the proficiency of three-dimensional
satellite images data sets (i.e. spatial, spectral and temporal) for the automatic mapping of
land cover and land cover time evolution have not been thoroughly explored. In this
dissertation, we investigate the usefulness of multispectral time series sets of medium spatial
resolution satellite images for the regular land cover characterization at the national scale.
This study is carried out on the territory of Continental Portugal and exploits satellite
images acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) and
MEdium Resolution Imaging Spectrometer (MERIS). In detail, we first focus on the analysis
of the contribution of multitemporal information from multispectral satellite images for the
automatic land cover classes’ discrimination. The outcomes show that multispectral
information contributes more significantly than multitemporal information for the automatic
classification of land cover types. In the sequence, we review some of the most important
steps that constitute a standard protocol for the automatic land cover mapping from satellite
images. Moreover, we delineate a methodological approach for the production and
assessment of land cover maps from multitemporal satellite images that guides us in the
production of a land cover map with high thematic accuracy for the study area. Finally, we
develop a nonlinear harmonic model for fitting multispectral reflectances and vegetation
indices time series from satellite images for numerous land cover classes. The simplified
multitemporal information retrieved with the model proves adequate to describe the main
land cover classes’ characteristics and to predict the time evolution of land cover classes’individuals.
Descrição
Thesis submitted to the Instituto Superior de Estatística e Gestão de
Informação da Universidade Nova de Lisboa in partial fulfillment of the
requirements for the Degree of Doctor of Philosophy in Information Management – Geographic Information Systems
Palavras-chave
Remote sensing Time series Multispectral images Land cover Separability analysis Classification Prediction Reference data Accuracy assessment Detecção remota Séries temporais Imagens multi-espectrais Ocupação do solo Análise de separabilidade Classificação Predição Dados de referência Avaliação da exactidão.