WP3 improvement of SOTER spatial and attribute data

WP leader: Cranfield University, Thomas Mayr

We will make improvements in three fields, working at a scale of 1:250 000:

  1. reform the analysis of DEM data by using natural breaks. Two approaches will be investigated: (a) a hierarchical classification of landform types (MacMillan et al. 2000) through a combination of dendogram analysis and area statistics, followed by a more detailed subdivision based on classification of landform elements using slope-break analysis; (b) a bottom-up approach of identifying natural breaks using local self-adjusting thresholds and up-scaling. The tewo approaches will finally create independent landform datasets.
  2. derive parent materials by using feature-enhanced, multispectral image processing to MODIS and Landsat-TM data - with edge and texture filtering and visual interpretation, making use of geological maps. There will be a link with the global 1:1 million geological mapping program. Especially in the non-European windows, spectral-mixture modelling will be applied to landscape-unit analysis in combination with DEM variables. The landscape will be divided according to lithology and vegetation.
  3. develop Europe-wide additional predictors of soil properties at various spatial, spectral and temporal scales. The multi-scale approach will have two baselines and a collaborative component: (a) we will focus on classification and regression-tree analysis using ancillary data to predict refined attribute classes of e-SOTER; (b) we will use evidential reasoning to either constrain or assimilate ancillary remote sensing data at European scale. The combination of constraint, regression and assimilation analysis involves a risk of over-fitting which may cause uncertainty. Therefore, in (c), an expert panel will evaluate variable and parameter dependencies that may affect spatial and attribute outputs for e-SOTER.
  4. integrate the products of the first 3 tasks of this WP for the 4 pilot windows resulting in 1:250 000 enhanced SOTER databases that will be valuated by local experts.

In the first component we will compile attribute data previously not used to constrain e-SOTER attributes (e.g. advanced land use/cover products, partly from the FP6 ECOCHANGE project), productivity and vegetative change including phenology indicators, the Soil Water Index (SWI) being a measure of the profile soil moisture content obtained by filtering the surface soil moisture time series with an exponential function, and the relation between soil groups to MODIS-derived surface albedo). These data will be used in a classification and regression-tree (CART) analysis.

The second component will use the same data refine the derivation of e-SOTER attributes spatially.

The third component will assess potential contradictions in the derived data, and specify the required additional information, complementary to WP1 and WP2 outputs. This will require field survey data from the pilot areas. An iterative feedback loop to WP4 is established in this task.