Improving remote sensing contribution to soilscape maps
Résumé
Within the French national program Inventaire Gestion et Conservation des Sols (IGCS), a central purpose is to provide a DSM at regional scale (Référentiel Régional Pédologique). For the Brittany region, soils maps are rare, as the existing ones, not yet digitized, covers approximately 20% of the area to the 1/100 000 and less than 15% to the 1/25°000. Facing this lack of spatial data on soils, the idea is to map soil-landscape units (Unités de Pédo Paysages : UPP) to the 1/250 000, and inform each UPP with a definition of local soils typology (Unités Typologiques de Sols : UTS). This concept of soil-landscape units results from the general idea, established in various disciplines, that there are functional links and reciprocal constraints between environmental factors ( hydrosphere, lithosphere, atmosphere, biosphere). Specifics adaptations and the relative equilibrium of the system assumes a relation between natural vegetation and soils, either at regional scale and local scale. In Brittany, where about 80% of the surface is devoted to agricultural land since ages, the variety of landscapes are strongly related to the variety of soils. As traditional agriculture (until 1970) was much more dependant on the land (so called “terroir”) , we notice that the main agricultural landscape patterns, (size and geometry of parcels, crops rotations or grass land locations, density of hedgerow network, waste land...), still remaining, are revealing soils attributes. Thus, according to the S.C.O.R.P.A.N. model, in region like Brittany O is an important factor in order to achieve DSM, but it's also important to note where the actual human activities forces soils evolution (urbanisation, protected woodland...). Remote sensing is obviously the main source of data to map landscape units at regional scale, but one must look carefully how to analyse landscape units, including soils properties like surface hydromorphy, without simply focus on land use classification. We would like to présent two methodological aspects of our work on DSM : 1) The object-oriented image analysis present very interesting opportunities applied to landscapes mapping. Indeed, as landscape is a complex spatial organisation of various elements, pixel-based classifications are not sufficient. Taking account of the neighbourhood, with cover frequency vectors for exemple, gives information on diversity and heterogeneity, but is still limited concerning spatial informations. Using an object oriented approach for the segmentation and classification of Landsat TM images with eCognition software underline the role of patterns in landscape discrimination (linear woodland in a valley, large parcells of cereals in plains, small parcells of grass land on hills, mixed or homogenous spatial organisation ...). Beyong the accuracy of landscape classification, the object oriented image analysis has a great interest concerning DSM for at least two reasons : first because landscape patterns are more stable than the land use their contain, second because these patterns can have an influence on soils (erosion...). Object oriented image analysis can also help to deal with scale issues in DSM using a hierarchical approach. 2) Choosing specific variables according to the scale of DSM can be a way of working with the SCORPAN model. Mappping soil-landscape units (UPP) at regional scale, and soils types at local scale, we have to analyse how the SCORPAN factors interfere according to the scale level, looking for the best variables at each level. Obviously scale issues are not just a question of spatial resolution of the digital data or techniques for down or upscaling. We would focus on SCORP variables that are discriminating for units caracteristics at regional scale but insufficient for local variability (agricultural morphology, elevations, spectral radiometrics, climatics indice based on temperature and précipitations...). Then we also examine the consistency with local variables (slopes, distance to the rivers, parent material, aspect, land use, ...) witch could be confusing at regional scale. These are combine using different techniques (fuzzy logic) and both largely use remote sensing as the main -but not unique- source of data (Landsat TM, MODIS, AVHRR...).