Unmixing the patterns within big data: discriminating forest in French Guiana using modis time-series - Université de Montpellier Accéder directement au contenu
Poster De Conférence Année : 2015

Unmixing the patterns within big data: discriminating forest in French Guiana using modis time-series


Despite having access to over 40 years of remotely sensed satellite data, from multiple sources (e.g. Landsat, SPOT), and multiple types of sensors (e.g. optical, radar, thermal), tropical forests are still poorly understood in terms of their composition and functioning. This is particularly true for French Guiana, the sparsely populated and highly forested French overseas Department which comprises the French segment of the biologically diverse Guiana Shield ecoregion. While French Guiana’s forests have been mapped under successive global land cover assessment initiatives such as the Global Land Cover Characterization (GLCC), Global Land Cover 2000 (GLC2000), and GlobCover 2009, among others, French Guiana’s forests are repeatedly depicted as one large ‘green carpet,’ despite evidence from the ground that French Guiana possesses a range of diverse forest ecosystems. In 2011, the need to go beyond the ‘green carpet’ effect led researchers to publish a new map of French Guiana’s “forest landscape types” which was based on data from the Vegetation instrument onboard the SPOT-4 satellite, and differentiating the ‘green carpet’ into five main groups of forest. Nevertheless, in 2015, beyond having access to a larger archive of satellite data from an ever growing array of sources, the range of techniques available for processing “big data” has also grown. For instance, the 2011 effort to map French Guiana’s forests utilized a variant on the classical technique for classifying forest types. In the classical approach, reflectance data from a single date is classified – usually using a supervised or unsupervised classification technique – in order to translate reflectance data into forest classes. In the 2011 effort, an entire year’s worth of SPOT Vegetation data was used to generate a single image mosaic which was then classified using an unsupervised clustering technique. The single-date processing approach has been used by remote sensing researchers the world over. Nevertheless, since the advent in the late 1990s / early 2000s of sensors such as SPOT Vegetation and MODIS, the ‘big data’ revolution in remote sensing has also had the consequence of areas of the world being imaged on an almost daily basis over the past ~16 years, compared to the 18 day revisit time of the first Landsat satellite, and the 26 day revisit time of the first SPOT satellite. In the context almost daily remotely sensed observations for the past decade and a half, this paper re-examines the issue of mapping of forest types in French Guiana, using different techniques and inputs from those previously used. The overall objective of this study was to determine the extent to which forest types in French Guiana could be discriminated via unmixing of multitemporal, multispectral reflectance data.
Fichier principal
Vignette du fichier
cherrington_multitemp_poster1.pdf (15.14 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-02318236 , version 1 (16-10-2019)


  • HAL Id : hal-02318236 , version 1


Emil Alexander Cherrington, Grégoire Vincent, Nicolas Barbier, Daniel Sabatier, Raphaël Pélissier. Unmixing the patterns within big data: discriminating forest in French Guiana using modis time-series. Multitemp 2015, 8th International Workshop on the Analysis of Multitemporal Remote Sensing Images, Jul 2015, Annecy, France. 13, pp.357 - 367, 2011. ⟨hal-02318236⟩
115 Consultations
21 Téléchargements


Gmail Mastodon Facebook X LinkedIn More