Potential of Pléiades mono- and tri-stereoscopic images for the agricultural mapping in Mayotte
In Mayotte, the agricultural land use is only partially known, due to very little GPS plot sampling, while an exhaustive mapping would be needed for decision support. Farms are very small (<0,5ha) and based on associated crops, orchards, and agroforestry. More than 40% of their production is for local livelihood. Pléiades' VHSR images thus appear as a relevant tool for the characterization of these production systems, provided that these images are acquired at different dates (dry season/rainy season) to better discriminate crops thanks to the agricultural calendar. Besides, the complexity of Mayotte's landscapes, dominated by the presence of various trees, requires developing a methodology which takes into account both the multi-temporal radiometry and the texture, but also the vegetation height. We therefore propose to test the capabilities of Pleiades' tri-stereoscopic mode to produce a reliable DSM usable in this context. The objective of this study is to assess the suitability of this type of data to a mapping base regularly updated. No image acquisition attempt during the rainy season was successful because of the heavy cloud cover. Only two images were acquired, respectively in July 2012 and April 2013. This latest, acquired in tri- stereoscopic mode, was issued as raw data in August 2013 due to production problems. The relevance of these two dates is weak because April and July correspond to the same phenological period of the vegetation; so there is little information related to the cycle differentiation between species. Finally, one year apart does not represent sufficient development in terms of land cover nor land use dynamics. Moreover, the data late provision did not allow us to achieve all the processes and get any result. The prospects of this study are thus to implement: The assessment of the capacity of deriving the DSM from the tri- stereoscopic images and the validity of this product. The production of a DEM derived from Pléiades-DSM, followed by the analysis of the coherence of the measured heights with those of the LiDAR-DSM acquired in 2008, and the suitability of this DEM to help producing the land use map. The texture analysis characterizing Mayotte's large tree cover variety in terms of composition, structure, height or density of vegetation, and discriminating different wooded patterns on the image (e.g. natural forest, agroforestry plot, monospecific tree grove). We will thus integrate indices derived from the texture co-occurrence matrix, calculated on different spectral bands with various neighborhood sizes. We will analyze the choice of the relevant panchromatic/spectral bands for derivation, the most discriminating sizes, but also the most appropriate indices. However, anomalies due to equalization residues at Pleiades sensor strips produce vertical stripes in the texture images, which limit the use of Pleiades imagery for this type of process. These limits will be evaluated. The object-oriented analysis (in eCognition Developer) to create a map of the agricultural land use by combining all the data described above.
| Main Authors: | , , |
|---|---|
| Format: | conference_item biblioteca |
| Language: | eng |
| Published: |
s.n.
|
| Subjects: | E90 - Structure agraire, U30 - Méthodes de recherche, P31 - Levés et cartographie des sols, K01 - Foresterie - Considérations générales, |
| Online Access: | http://agritrop.cirad.fr/573333/ http://agritrop.cirad.fr/573333/1/document_573333.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In Mayotte, the agricultural land use is only partially known, due to very little GPS plot sampling, while an exhaustive mapping would be needed for decision support. Farms are very small (<0,5ha) and based on associated crops, orchards, and agroforestry. More than 40% of their production is for local livelihood. Pléiades' VHSR images thus appear as a relevant tool for the characterization of these production systems, provided that these images are acquired at different dates (dry season/rainy season) to better discriminate crops thanks to the agricultural calendar. Besides, the complexity of Mayotte's landscapes, dominated by the presence of various trees, requires developing a methodology which takes into account both the multi-temporal radiometry and the texture, but also the vegetation height. We therefore propose to test the capabilities of Pleiades' tri-stereoscopic mode to produce a reliable DSM usable in this context. The objective of this study is to assess the suitability of this type of data to a mapping base regularly updated. No image acquisition attempt during the rainy season was successful because of the heavy cloud cover. Only two images were acquired, respectively in July 2012 and April 2013. This latest, acquired in tri- stereoscopic mode, was issued as raw data in August 2013 due to production problems. The relevance of these two dates is weak because April and July correspond to the same phenological period of the vegetation; so there is little information related to the cycle differentiation between species. Finally, one year apart does not represent sufficient development in terms of land cover nor land use dynamics. Moreover, the data late provision did not allow us to achieve all the processes and get any result. The prospects of this study are thus to implement: The assessment of the capacity of deriving the DSM from the tri- stereoscopic images and the validity of this product. The production of a DEM derived from Pléiades-DSM, followed by the analysis of the coherence of the measured heights with those of the LiDAR-DSM acquired in 2008, and the suitability of this DEM to help producing the land use map. The texture analysis characterizing Mayotte's large tree cover variety in terms of composition, structure, height or density of vegetation, and discriminating different wooded patterns on the image (e.g. natural forest, agroforestry plot, monospecific tree grove). We will thus integrate indices derived from the texture co-occurrence matrix, calculated on different spectral bands with various neighborhood sizes. We will analyze the choice of the relevant panchromatic/spectral bands for derivation, the most discriminating sizes, but also the most appropriate indices. However, anomalies due to equalization residues at Pleiades sensor strips produce vertical stripes in the texture images, which limit the use of Pleiades imagery for this type of process. These limits will be evaluated. The object-oriented analysis (in eCognition Developer) to create a map of the agricultural land use by combining all the data described above. |
|---|