Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Allemagne → Improving the Utilization of Remote Sensing Data for Land Cover Characterization and Vegetation Dynamics Modelling

Ludwig-Maximilians-Universität München (2016)

Improving the Utilization of Remote Sensing Data for Land Cover Characterization and Vegetation Dynamics Modelling

Liya Sun

Titre : Improving the Utilization of Remote Sensing Data for Land Cover Characterization and Vegetation Dynamics Modelling

Auteur : Liya Sun

Université de soutenance : Ludwig-Maximilians-Universität München

Grade : Doctoral thesis 2016

Résumé
The land surface is strongly controlling the energy and water fluxes between the above-surface and subsurface systems. The complexity of the land surface system exhibits with large spatial heterogeneity of the land surface properties or high temporal variability of land surface processes. As the essential parts of the land surface system, land cover patterns and dynamic changes are strongly required in the land surface modelling across the temporal and spatial scales. Therefore, in the joint CAOS project (“From Catchement as Organised Systems to Models based on Dynamic Functional Units”), one primary objective is to improve the retrieval of land surface characteristics in a meso-scale catchment. Specifically, standing at the point of view by using remote sensing techniques, great efforts are made in this thesis to derive the spatially distributed land cover information and quantify the vegetation dynamics. Remote sensing techniques provide multi-spatial and multi-temporal land cover information, which have been successfully applied in a variety of land surface studies. Current land cover mapping studies have been focusing on developing the classification methods by using the visible or near-infrared data (VIS/NIR). However, very limited studies have considered the effectiveness of the thermal infrared (TIR) data. TIR information has been proved to be tightly related to the energy and water fluxes in the land surface system. The land surface temperature (LST) is frequently used as an important parameter for the modelling of land surface energy balance, or the evaluation of surface moisture and evapotranspiration. Moreover, the development of satellite instruments have promoted the availability of TIR data. The valuable TIR data captured by the current satellite sensors should be fully exploited. Therefore, one of our objectives is to investigate the usefulness of the TIR data in the land cover classification A comprehensive study of the TIR and VIS/NIR bands from the Landsat images was conducted. Contrary to previous studies with tremendous efforts on developing the classification algorithms, the essential characteristics of the Landsat data are paid more attention in this work. Therefore, the simple k-fold nearest neighborhood algorithm and advanced random forest method were selected as the classification algorithms. In the aspect of the data features, different variants were derived from the Landsat images ranging from two bands to seven bands composition. From the temporal scales, both single-date and multi temporal Landsat images were evaluated. Furthermore, the classification results were analyzed by the pixel-based and polygon-based cross validation (CV) methods for uncertainty assessment

Présentation

Version intégrale (5 Mb)

Page publiée le 28 septembre 2017