Interprétation d'images radar sur l'île d'Algodoal (Para - Brésil) en vue d'une cartographie des milieux naturels du littoral amazonien
Résumé
Interpretation of radar images of the Algodoal Island (Pará, Brazil) in order to achieve a mapping of the tropical environment of amazonian coastal zones
Synthetic aperture radar images are a powerful potential source of data about tropical zones, since microwaves are not hindered by clouds in the wavelengths used by these radars. However, the "speckle" - i.e. a coherent noise that is linked with the nature of the instrument - make them uneasy to interpret. We cannot solely rely on the intensity of a given pixel in order to identify its nature. In order to make a good interpretation, one must also consider a "texture", that is a peculiar arrangement of many close pixels of different intensities. Our study is a comparison between the interpretations made on two images, one produced by the SAR of the European satellite ERS-1 and the other by the SAR of the Japanese satellite JERS-1.
The images are first reduced from 16 to 8 bits by pixel and then treated with a speckle removal filter. These processings were made by using various softwares. After some testing, we decided that speckle removal filter for our purpose was an adaptative median . We then introduce the images in a Geographical Information System (namely Mapinfo). We use the coordinates given in the header of the ERS-1 SAR image in order to geoprocess this image. We then rectify the JERS-1 SAR image by taking Control Points on the ERS-1 SAR image. Thus, even if the absolute precision of the different covers is not perfect, the two images are completely compatible (However field studies lead us to think that the accuracy of these header coordinates is about 200 m for ERS-1).
The images are then interpreted by mere photointerpretation and several layers are produced, each containing a kind of geographical objects : mangrove areas, forest areas, dune areas, etc. The map obtained from the ERS-1 SAR image shows that this instrument is especially good in the cartography of coastal geomorphological forms like dunes, sand beaches or ripplemarks zones. It does also give a clear view of the importance and localization of the mangrove areas which are present on the main part of the coast.
The JERS-1 SAR shows the human activities better, like clearings in the forest remnants or the small villages, which could not be detected in the other image. The mangrove areas, a critical topic in this context, are well detected in this image also, despite some differences in the evaluation of the area covered. On this topic, the comparison with an HRV XS image of SPOT enhanced by field studies proves that the evaluation given by the ERS-1 SAR is closer to reality than JERS-1 SAR's.
The map produced from each image is not as precise and complete as a topographic map, but it is a good instrument to study the different vegetative areas and the coastline. Thus this study tends to demonstrate that it is possible to produce maps at the scale of 1 : 100 000 of this kind of zones, which could be of great help for land planning and environment monitoring in Brazil.
Synthetic aperture radar images are a powerful potential source of data about tropical zones, since microwaves are not hindered by clouds in the wavelengths used by these radars. However, the "speckle" - i.e. a coherent noise that is linked with the nature of the instrument - make them uneasy to interpret. We cannot solely rely on the intensity of a given pixel in order to identify its nature. In order to make a good interpretation, one must also consider a "texture", that is a peculiar arrangement of many close pixels of different intensities. Our study is a comparison between the interpretations made on two images, one produced by the SAR of the European satellite ERS-1 and the other by the SAR of the Japanese satellite JERS-1.
The images are first reduced from 16 to 8 bits by pixel and then treated with a speckle removal filter. These processings were made by using various softwares. After some testing, we decided that speckle removal filter for our purpose was an adaptative median . We then introduce the images in a Geographical Information System (namely Mapinfo). We use the coordinates given in the header of the ERS-1 SAR image in order to geoprocess this image. We then rectify the JERS-1 SAR image by taking Control Points on the ERS-1 SAR image. Thus, even if the absolute precision of the different covers is not perfect, the two images are completely compatible (However field studies lead us to think that the accuracy of these header coordinates is about 200 m for ERS-1).
The images are then interpreted by mere photointerpretation and several layers are produced, each containing a kind of geographical objects : mangrove areas, forest areas, dune areas, etc. The map obtained from the ERS-1 SAR image shows that this instrument is especially good in the cartography of coastal geomorphological forms like dunes, sand beaches or ripplemarks zones. It does also give a clear view of the importance and localization of the mangrove areas which are present on the main part of the coast.
The JERS-1 SAR shows the human activities better, like clearings in the forest remnants or the small villages, which could not be detected in the other image. The mangrove areas, a critical topic in this context, are well detected in this image also, despite some differences in the evaluation of the area covered. On this topic, the comparison with an HRV XS image of SPOT enhanced by field studies proves that the evaluation given by the ERS-1 SAR is closer to reality than JERS-1 SAR's.
The map produced from each image is not as precise and complete as a topographic map, but it is a good instrument to study the different vegetative areas and the coastline. Thus this study tends to demonstrate that it is possible to produce maps at the scale of 1 : 100 000 of this kind of zones, which could be of great help for land planning and environment monitoring in Brazil.
Nous comparons ici les interprétations que l'on peut faire d'images des RSO des satellites ERS-1 et JERS-1 de l'île d'Algodoal au Brésil. Ces images ont été préalablement converties de 16 bits en 8 bits, et traitées de manière à atténuer le chatoiement. On remarque que le capteur RSO de ERS-1 est particulièrement performant pour la discrimination des milieux naturels littoraux, et qu'il nous donne de nombreux éléments pour la compréhension de la dynamique littorale. Le capteur RSO de JERS-1 est plus sensible aux activités humaines, il nous permet de cartographier les défrichements et les villages. Malgré ces différences, les secteurs de mangrove, le milieu le plus significatif de l'île, sont bien repérés dans les deux images, même si la surface totale reconnue varie significativement. Les résultats obtenus permettent d'envisager une cartographie au 1 : 100 000 à partir d'interprétations visuelles d'images de ces deux capteurs.