Mapping of natural and semi-natural vegetation types in French agricultural landscapes

Authors and Affiliations: 

Christina Corbane, Sylvio Lavventure, Samuel Alleaume and Michel Deshayes

Irstea, UMR TETIS, Montpellier, France


The ‘Multi-scale Service for Monitoring Natura 2000 Habitats of European Community Interest (MS.MONINA)’ project is designed to develop new Earth Observation (EO)–based services for the cost-effective collection of information at different scales across the Natura 2000 network and beyond. The project’s design is unique in this sense as it integrates mapping approaches from the local to regional to European levels and addresses the reporting, monitoring and management needs of stakeholders at three scales: the individual local site, the Member State, and the EU. In this context, the National Research Institute of Science and Technology for Environment and Agriculture (Irstea) is developing a methodological framework for mapping of natural and semi-natural vegetation types across the French agricultural landscapes at a scale of 1/25000 using EO data. The service, under development, is tailored to lowland vegetation in open areas and is easily transferable to the different biogeographical regions characterizing the continental French territory (Continental, Atlantic, Mediterranean and Alpine).

The method consists in an object-orientated rule-based classification of multi-temporal and multi-sensor data (Landsat 7, IRS and aerial photography) acquired between 2008 and 2011. The concept relies on a coupling of ecological expert knowledge (empirical rules), information content of EO data, and ancillary data for extracting features with relevance for the assessment of specific physiognomic vegetation classes (Figure 1). Key components include:

1) the use of the multi-level segmentation of color infrared aerial photos (at a spatial resolution 0.5 meters) available at the national scale, provided by the IGN/IFN (Institut national de l’information géographique et forestière): Level 0 is a broad level corresponding to the mask of forest area; Level 1 is a medium segmentation level used in order to extract agricultural areas; and Level 2 is a fine level corresponding to small natural habitats.

2) the development and implementation of ecological rules for the classification of vegetation types,

3) the hierarchical classification starting from the discrimination of non-natural vegetation (including agricultural areas) from semi-natural vegetation and ending with the identification of herbaceous surfaces, together with their productivity levels, and low woody vegetation, grouped by the densities of the vegetation cover.

In the context of agricultural landscapes, the main difficulty lies in distinguishing croplands or arable areas from grasslands. Given that croplands are usually characterized by bare soils after being harvested (at least once a year), the temporal variations of three vegetation indices, derived from remote sensing time series, were used to identify the discriminate between croplands and semi-natural areas: NDVI (for photosynthetic vegetation), NDSVI (non-photosynthetic vegetation) and Background (Zhang et al., 2011). As for the identification of woody vegetation and its density, textural features (mainly “Contrast and Homogeneity” features) derived from the Grey Level Co-occurrence matrix (GLCM) [Haralick and Shamnugam, 1973]) were used.

The service cases for the Alpine and the Continental biogeographical regions were successfully implemented on the test sites of Belledonne and Forez plain respectively. The method is tested over the Cévennes National Park located in the Mediterranean biogeographical region.



Zhang Y., Smith A. M., Hill M. J., Estimating Fractional Cover of Grassland Components from Two Satellite Remote Sensing Sensors (2011), ISPRS conference proceedings.

Haralick, R.M. and S. Shamnugam, Textural Features for Image Classification, with S. Shamnugam, IEEE Transactions on Cybernetics (1973), Vol SMC-3, No. 2.