Reunion Island - 2017, reference spatial database
The reference spatial database for 2017 is composed of 6256 plots. We use it to calculate a land use map from satellite images.It is organized according to a nomenclature offering 3 levels of precision. We randomly selected 20% of the plots in each class to build a validation database while the remaining 80% is used for learning (5002 polygons for learning and 1254 for validation). <br> The following is a brief description of the sources and techniques used to develop it according to land use types : <ul> <li> <b> For agricultural areas </b>, we have a land use database based on farmers' declarations to apply for EU subsidies. This is the Registre Parcellaire Graphique (RPG) published in France by the French Institute of Geography (IGN). The description of this data is available here: http://professionnels.ign.fr/doc/DC_DL_RPG-2-0.pdf . These vector data are accurate and can be used as a model to locate crops. The release times imply that we use the RPG for year N -1. It is therefore necessary to check the correct consistency of the data by photo-interpretation of the VHR image. The RPG provides limited information on orchards. For these classes we called on colleagues specialised in mango, lychee and citrus fruit cultivation technicians who are familiar with their sector and can locate plots in the VHR image. Field surveys were conducted using GPS for market gardening crops. The plots of the "greenhouse or shade cultivation" class are derived from the "industrial building" layer of the IGN's "BD Topo" product of IGN. A random selection of 20% of the polygons in the height field of the IGN layer allows to keep a diversity of greenhouse types. Each of the polygons was verified by photo-interpretation of the Pleiades image. If the greenhouse or shade was not visible in the image, the polygon was deleted. <li> <b> For natural areas</b>, there is no regularly updated mapping, but the main classes can be recognized from the GIS layers of the State services that manage these areas (ONF and DEAL). Two specific classes have been added (identified by photo-interpretation) to address the problems of satellite images: a class of shadows due to the island's steep terrain (areas not visible because of the shadow cast) and a class of vegetation located on steep slopes facing the morning sun called "savannah on cliffs". <li> <b> For wet areas</b>, the "marsh", "water" and "hillside retention" classes were obtained by photo-interpretation of the 2017 Pleiades image. These classes are easily recognizable on this type of image. <li> <b> For urban areas</b> we randomly selected polygons from the IGN's BD Topo layer. For the housing type building, 4 building height classes have previously been created (depending on the height of the layer field) in order to preserve a good diversity of the types of buildings present on the island. A random selection of polygons from each class was then made. The "built" layer was completed by a random selection of industrial buildings from the "industrial built" layer of the IGN's TOPO database. This selection was made in the "nature" field of the layer (i.e. the following types: silo, industrial and livestock). The "photovoltaic panel" class was obtained by photo-interpretation of the polygons on the 2017 Pleiades image.</li> </ul> <br>
| Main Author: | |
|---|---|
| Format: | Observational data biblioteca |
| Language: | French |
| Published: |
CIRAD Dataverse
2018
|
| Subjects: | Agricultural Sciences, Computer and Information Science, Earth and Environmental Sciences, Social Sciences, remote sensing, Télédétection, spatial database, Base de données spatiale, Land cover, Occupation des terres, Forêt primaire, primary forests, |
| Online Access: | http://dx.doi.org/10.18167/DVN1/TOARDN |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The reference spatial database for 2017 is composed of 6256 plots. We use it to calculate a land use map from satellite images.It is organized according to a nomenclature offering 3 levels of precision. We randomly selected 20% of the plots in each class to build a validation database while the remaining 80% is used for learning (5002 polygons for learning and 1254 for validation).
<br>
The following is a brief description of the sources and techniques used to develop it according to land use types :
<ul>
<li> <b> For agricultural areas </b>, we have a land use database based on farmers' declarations to apply for EU subsidies. This is the Registre Parcellaire Graphique (RPG) published in France by the French Institute of Geography (IGN). The description of this data is available here: http://professionnels.ign.fr/doc/DC_DL_RPG-2-0.pdf . These vector data are accurate and can be used as a model to locate crops. The release times imply that we use the RPG for year N -1. It is therefore necessary to check the correct consistency of the data by photo-interpretation of the VHR image. The RPG provides limited information on orchards. For these classes we called on colleagues specialised in mango, lychee and citrus fruit cultivation technicians who are familiar with their sector and can locate plots in the VHR image. Field surveys were conducted using GPS for market gardening crops. The plots of the "greenhouse or shade cultivation" class are derived from the "industrial building" layer of the IGN's "BD Topo" product of IGN. A random selection of 20% of the polygons in the height field of the IGN layer allows to keep a diversity of greenhouse types. Each of the polygons was verified by photo-interpretation of the Pleiades image. If the greenhouse or shade was not visible in the image, the polygon was deleted.
<li> <b> For natural areas</b>, there is no regularly updated mapping, but the main classes can be recognized from the GIS layers of the State services that manage these areas (ONF and DEAL). Two specific classes have been added (identified by photo-interpretation) to address the problems of satellite images: a class of shadows due to the island's steep terrain (areas not visible because of the shadow cast) and a class of vegetation located on steep slopes facing the morning sun called "savannah on cliffs".
<li> <b> For wet areas</b>, the "marsh", "water" and "hillside retention" classes were obtained by photo-interpretation of the 2017 Pleiades image. These classes are easily recognizable on this type of image.
<li> <b> For urban areas</b> we randomly selected polygons from the IGN's BD Topo layer. For the housing type building, 4 building height classes have previously been created (depending on the height of the layer field) in order to preserve a good diversity of the types of buildings present on the island. A random selection of polygons from each class was then made. The "built" layer was completed by a random selection of industrial buildings from the "industrial built" layer of the IGN's TOPO database. This selection was made in the "nature" field of the layer (i.e. the following types: silo, industrial and livestock). The "photovoltaic panel" class was obtained by photo-interpretation of the polygons on the 2017 Pleiades image.</li>
</ul>
<br> |
|---|