• Home
• Program
• Search
• Author Index
• Sponsors
• Committees
• Contact Us
• Tech Support

Session:

Environment 1

Abstract No.:

O-011

Title:

Cokriging for Large Spatial Datasets: Mapping Soil Properties at a Region Scale from Airborne Hyperspectral Imagery.

Author(s):

P. MONESTIEZ, INRA (FR)
E. WALKER, INRA (FR)
C. GOMEZ, IRD (FR)
Ph. LAGACHERIE, INRA (FR)
R CIAMPALINI, INRA (FR)

Abstract:

Recent developments in soil sensing technologies, initially oriented towards soil mapping at the field scale for precision agriculture, show high potential for digital soil mapping (DSM) of large areas. We present here a spatial statistical model that combines hyperspectral remote sensing, field measurements and, potentially soil types from existing pedological maps, to predict soil properties as clay or calcium carbonate contents at increasing resolutions from 5m to 100m over large regions. Methodological difficulties arise from dimensional aspects. From a spatial point of view, the geostatistical model have to be inferred from rare field soil samples and remote sensing data that are patchy - only informative on bare soils - and very numerous - several thousand records at fine resolution. From a multivariate point of view, soil properties have to be predicted using PLS from high dimensional ? 256 bands ? hyperspectral data. To illustrate the proposed approach, a 25-square-km area located in the vineyard plain of Languedoc and a 600-square-km area in Tunisia were surveyed with both airborne hyperspectral remote sensing data at a 5-m resolution and complementary field survey. Various maps of clay and calcium-carbonate content were produced by block cokriging and represent different compromises between prediction accuracy and spatial resolution.