On the nature of models in remote sensing
Remote Sensing of Environment
Introductory Digital Image Processing: A Remote Sensing Perspective
Introductory Digital Image Processing: A Remote Sensing Perspective
| Hi-index | 0.00 |
The majority of the vast literature on remote sensing of urban landscapes has adopted a 'hard classification' approach, in which each image pixel is assigned a single land use and land cover category. Owing to the nature of urban landscapes, the confusion between land use and land cover definitions and the constraints of widely applied medium spatial resolution satellite images, high classification accuracy has been difficult to achieve with the conventional 'hard' classifiers. The prevalence of the mixed pixel problem in urban landscapes indicates a crucial need for an alternative approach to urban analyses. Identification, description and quantification, rather than classification, may provide a better understanding of the compositions and processes of heterogeneous landscapes such as urban areas. This study applied the Vegetation-Impervious Surface-Soil (V-I-S) model for characterizing urban landscapes and analysing their dynamics in Indianapolis, USA, between 1991 and 2000. To extract these landscape components from three dates of Landsat Thematic Mapper/Enhanced Thematic Mapper Plus (TM/ETM+) images in 1991 1995 and 2000, we used the technique of linear spectral mixture analysis (LSMA). These components were further classified into urban thematic classes, and used for analysis of the landscape patterns and dynamics. The results indicate that LSMA provides a suitable technique for detecting and mapping urban materials and V-I-S component surfaces in repetitive and consistent ways, and for solving the spectral mixing of medium spatial resolution satellite imagery. The reconciliation between the V-I-S model with LSMA for Landsat imagery allowed this continuum landscape model to be an alternative, effective approach to characterizing and quantifying the spatial and temporal changes of the urban landscape compositions in Indianapolis from 1991 to 2000. It is suggested that the model developed in this study offers a more realistic and robust representation of the true nature of urban landscapes, as compared with the conventional method based on 'hard classification' of satellite imagery. The general applicability of this continuum model, especially its spectral, spatial and temporal variability, is discussed.