Detection of Vegetation Coverage in Urban Agglomeration of Brussels by NDVI Indicator Using eCognition Software and Remote Sensing Measurements
Résumé
Introduction. The study area encompasses selected regions of the Brussels municipality, Belgium. In the past years the city of Brussels is experiencing intensification of the density of building structures. Unlike in some other European cities, where the most evident problem is urbanization and expansion of the city margins to the suburbia, the urban structure Brussels is the intensification of the buildings density in the city centre and the existing dwelling districts. Thus, the city structure tends to become more intense and dense, due to the process of filling the empty spaces in the urban patterns and high level housing. Another example of urban processes in Brussels is reorganisation of the industrial areas. At the same time, monitoring vegetation areas is essential for environmental sustainability of the capital city. The lack of the green spaces may cause ecological instability and increase atmospheric pollution. For studies of the specific problems of the Brussels city the remote sensing data (raster image) was used together with NDVI function, in order to detect areas covered by city parks.
Methods and Results
Step I. The panchromatic image was loaded into project and processed.
Step II. The image was segmented using the principle of Multiresolution Segmentation. This operation consists in the splitting of an image into segments in order to simplify complexity of the whole initial image. It is performed by the machine embedded logic based on mathematic algorithms and simplifying models. The general rule divides the area into regions according to the principle “neighbour pixels have similar parameters” (spectral reflectance value, texture, form, shape).
Step III. All vegetation coverage has been detected and separated from other objects (impervious structures) using natural characteristics of the vegetation spectral reflectance. For this, first the arithmetic expression for Normalized Difference Vegetation Index (NDVI) was created in eCognition operators using formula for spectral reflectance in visible (VIS) and near infrared (NIR) bands. This enables to detect pixels of vegetation: NDVI=(NIR-VIS)/(NIR+VIS). After this operation was created, it was added to the conditions of objects processing.
Step IV. After this, the operation of extraction of vegetation was performed. The logical condition for vegetation detection is that we assign all objects which have values of NDVI more than 0.3 to vegetation. This is based on the properties of vegetation: dense tree canopy usually have positive values of NDVI (0.3 to 0.8). On the contrary, other objects, which do not belong to vegetation, have low NDVI values. For example, water bodies have low reflectance in both spectral bands (band 3 and band 4). Therefore, they have very low positive and sometimes slightly negative NDVI values (depending on local hydro-chemical conditions, depth, etc). Bare soils usually also have small positive NDVI values (0.1 to 0.2), since their spectral reflectance in near-infrared bands is larger than in red ones. So, using this knowledge, the NDVI formula was applied, and green areas within the city of Brussels were distinguished. The objects with NDVI values more than 0.3 were assigned to the “vegetation” class.
Conclusions. The urban landscapes have complex environmental and socio-economic function and serve as habitat and agricultural surface in the surroundings. Land cover studies supported by satellite image contribute to the development of urban management system. Using object-oriented approach together with GIS techniques applied to remote sensing data enables to perform geospatial analysis with focus on urban landscapes.
Origine : Fichiers produits par l'(les) auteur(s)