Content uploaded by Vincent Thierion
Author content
All content in this area was uploaded by Vincent Thierion on Nov 12, 2014
Content may be subject to copyright.
A preview of the PDF is not available
The potential of pléiades imagery for vegetation mapping: A case study of plain and mountainous open environments
Abstract and Figures
Nowadays the use of remote sensing for vegetation mapping over large areas is becoming progressively common, with the increase of satellites providing a good trade-off between metric spatial resolution and large swath (e.g. Spot 5, RapidEye). Infra-metric imagery of Pléiades constellation offer valuable insights on vegetation structure. In the framework of the French national project CarHAB, this research aims at exploring the potential of this imagery and associated texture features (Haralick et SFS) in order to improve the discrimination of woody and herbaceous habitats and vegetation associated to screes. The work was tested in both, plain and mountainous environments in the French Alps (Isere Department). Promising results suggested that texture features derived from Pléiades imagery have a great potential discriminating vegetation structure. In all, the approach developed opens innovative ways towards a replicable rule-based classification scheme for vegetation mapping over open environments.
Figures - uploaded by Vincent Thierion
Author content
All figure content in this area was uploaded by Vincent Thierion
Content may be subject to copyright.
ResearchGate has not been able to resolve any citations for this publication.
SYKE is performing new CORINE 2006-classification for Finland. One of the aims is to make CORINE classification in Northern Finland, meaning that classes like Natural grasslands and Moors and heathlands should be classified with higher accuracy. Also, some specific classes need to be interpreted for national purposes like mountain birch forests. This paper documents the first experiments made using decision tree classifier, optical and microwave remote sensing data as well as DEM and soil information. Classes are pine, spruce, deciduous tree forests, two classes of mountain birch, open bog, grasslands, heathlands and open rocks. The best overall accuracies were about 73%, when the overall accuracy of Maximum Likelihood Classification was about 58%.
The detection of forest types and habitats is of major importance for silvicultural management as well as for the monitoring of biodiversity in the context of NATURA 2000. For these purposes, the presented study applies an object-based classification method using VHR QuickBird data at a test site in the pre-alpine area of Bavaria (southern Germany). Additional geo-data and derived parameters, such as altitude, aspect, slope, or soil type, are combined with information about forest development and integrated into the classification using a fuzzy knowledgebase. Natural site conditions and silvicultural site conditions are considered in this rule-base. The results of the presented approach show higher classification accuracy for the classification of forest types using ancillary information than can be reached without additional data. Moreover, for forest types with very distinctly defined ecological niches (e. g. alluvial types of forest), a better characterisation and integration of rules is possible than for habitats with very wide ecological niches. Hence, classification accuracies are significantly higher when these rules are applied. In a second step NATURA 2000 habitat types and selected habitat qualities are derived from the classified forest types. However, the share of habitat qualities varies with an altering scale. This difficulty should be addressed in further research of NATURA 2000 monitoring.
In this survey we review the image processing literature on the various approaches and models investigators have used for texture. These include statistical approaches of autocorrelation functions, optical transforms, digital transforms, textural edgeness, structural element, gray tone co-occurrence, run lengths, and autoregressive models. We discuss and generalize some structural approaches to texture based on more complex primitives than gray tone. We conclude with some structural-statistical generalizations which apply the statistical techniques to the structural primitives. -Author
The analysis of forest structure and species composition with high spatial resolution (≤ 1 m) multispectral digital imagery is described in an experiment using spatial co-occurrence texture analysis and maximum-likelihood classification. The objective was to determine if higher forest species composition classification accuracies would result in comparison to the use of spectral response patterns alone. Increased accuracy was obtained when using texture at all levels of a classification hierarchry. At the stand level, accuracies were on the order of 75 percent in agreement with field surveys, an improvement of 21 percent over the accuracy obtained using spectral data alone; in stands grouped according to species dominance/co-dominance, the accuracy improved still further to 80 percent. The overall classification accuracy in a highly generalized lifeform classification was 100 percent. This represented a 33 percent increase in accuracy over that which could be obtained, in a classic spectral "signature" classification approach, using spectral response patterns alone.
This study investigated the use of a geographic object-based image analysis (GEOBIA) approach with the incorporation of object-specific grey-level co-occurrence matrix (GLCM) texture measures from a multispectral Ikonos image for delineation of deciduous, evergreen, and mixed forest types in Guilford Courthouse National Military Park, North Carolina. A series of automated segmentations was produced at a range of scales, each resulting in an associated range of number and size of objects (or segments). Prior to classification, the spatial autocorrelation of each segmentation was evaluated by calculating Moran’s I using the average image digital numbers (DNs) per segment. An initial assumption was made that the optimal segmentation scales would have the lowest spatial autocorrelation, and conversely, that over- and under-segmentation would result in higher autocorrelation between segments. At these optimal segmentation scales, the automated segmentation was found to yield information comparable to manually interpreted stand-level forest maps in terms of the size and classification of images identified from a spatial autocorre-lation analysis to have an optimal segmentation, compared to those determined to have over- and under-segmentation. An overall accuracy of 79 percent with a Kappa of 0.65 was obtained at the optimal segmentation scale of 19. The addition of object-specific GLCM multiple texture analysis improved classification accuracies up to a value of 83 percent overall accuracy and a Kappa of 0.71 by reducing the confusion between evergreen and mixed forest types. Although some misclassification still remained because of local segmentation quality, a visual assessment of the texture-enhanced GEOBIA classification generally agreeable with manually interpreted forest types.
Grassland is a land cover in the area of conflict between agriculture and conservation, where intensification of land use is a major threat to grassland biodiversity. Grassland use intensity is a key factor for the conservation value of grassland, and detailed spatial data on grassland use intensity is needed to improve strategies for biodiversity conservation. A new remote sensing-based approach using multi-temporal high resolution RapidEye satellite data was developed in the present study that makes a large-scale assessment of grassland use intensity possible. RapidEye is a constellation of five satellites with 6.5 m spatial resolution, which allows frequent and timely image acquisition targeted at specific growing seasons. Semi-natural grassland, extensively used grassland, intensively used grassland and tilled grassland could be reliably differentiated at the management plot level in a study area in southern Germany. Various combinations of images from different observation dates have been tested as classification input and their overall classification accuracies were validated by field data. Best results were achieved using a combination of five multi-temporal scenes with an overall accuracy of 85.7%. A three-scene combination resulted in an overall accuracy of 82.2%. The analysis showed that seasonal aspects are very important when selecting adequate observation dates. Grassland use intensity was also assessed on peatlands using a peat soil map, since land use intensity significantly affects greenhouse gas emissions from peatlands. The results demonstrate the potential of targeted multi-spectral, high spatial resolution remote sensing for the large-scale monitoring of dynamic habitats, which is of vital importance to support various environmental conservation schemes through improved monitoring and reporting capabilities.
This study was the first to use high-resolution IKONOS imagery to classify vegetation communities on sub-Antarctic Heard Island. We focused on the use of texture measures, in addition to standard multispectral information, to improve the classification of sub-Antarctic vegetation communities. Heard Island’s pristine and rapidly changing environment makes it a relevant and exciting location to study the regional effects of climate change. This study uses IKONOS imagery to provide automated, up-to-date, and non-invasive means to map vegetation as an important indicator for environmental change. Three classification techniques were compared: multispectral classification, texture based classification, and a combination of both. Texture features were calculated using the Grey Level Co-occurrence Matrix (GLCM). We investigated the effect of the texture window size on classification accuracy. The combined approach produced a higher accuracy than using multispectral bands alone. It was also found that the selection of GLCM texture features is critical. The highest accuracy (85%) was produced using all original spectral bands and three uncorrelated texture features. Incorporating texture improved classification accuracy by 6%.
A review is presented of the image processing literature on the various approaches and models investigators have used for textures. These include statistical approaches of autocorrelation function, optical transforms, digital transforms, textural edgeness, structural element, gray tone co-occurrence, run lengths, and auto-regressive models. A discussion and generalization is presented of some structural approaches to texture based on more complex primitives than gray tone. Some structural-statistical generalizations which apply the statistical techniques to the structural primitives are given.
Ecologists commonly collect data on vegetation structure, which is an important attribute for characterizing habitat. However, measuring vegetation structure across large areas is logistically difficult. Our goal was to evaluate the degree to which sample-point pixel values and image texture of remotely sensed data are associated with vegetation structure in a North American grassland-savanna-woodland mosaic. In the summers of 2008-2009 we collected vegetation structure measurements at 193 sample points from which we calculated foliage-height diversity and horizontal vegetation structure at Fort McCoy Military Installation, Wisconsin, USA. We also calculated sample-point pixel values and first- and second-order image texture measures, from two remotely sensed data sources: an infrared air photo (1-m resolution) and a Landsat TM satellite image (30-m resolution). We regressed foliage-height diversity against, and correlated horizontal vegetation structure with, sample-point pixel values and texture measures within and among habitats. Within grasslands, savanna, and woodland habitats, sample-point pixel values and image texture measures explained 26-60% of foliage-height diversity. Similarly, within habitats, sample-point pixel values and image texture measures were correlated with 40-70% of the variation of horizontal vegetation structure. Among habitats, the mean of the texture measure ‘second-order contrast’ from the air photo explained 79% of the variation in foliage-height diversity while ‘first-order variance’ from the air photo was correlated with 73% of horizontal vegetation structure. Our results suggest that sample-point pixel values and image texture measures calculated from remotely sensed data capture components of foliage-height diversity and horizontal vegetation structure within and among grassland, savanna, and woodland habitats. Vegetation structure, which is a key component of animal habitat, can thus be mapped using remotely sensed data.