iSOIL - Interactions between soil related sciences -
Linking geophysics, soil science and digital soil mapping

WP 4: Pedometrics

Lead Partner: UT

Pedometrics is an integrative part of the iSOIL project dealing with data mining and data fusion towards the generation of digital soil maps at different scales. Within the iSOIL project, the analysis of sensor combinations to derive soil properties was most crucial for providing valid, high resolution and transferable results. A further key component was to follow a hierarchical approach for data collection, validation and implementation. Following objectives are dealt with:

  • Design of sampling schemes
  • Improvement and development of digital soil mapping approaches
  • Development of new models and transfer functions
  • Cross-Scale Analysis
  • Methodology for preparation of soil property and soil functional maps
  • Methods for implementation of soil monitoring

Sampling
Several approaches are documented to select samples based on available covariates such as a weighted version of the conditioned latin hypercube sampling (cLHS), fuzzy k-means sampling (FKMS) and response surface sampling The results show that in average the combination of a weighted version of the conditioned latin hypercube sampling approach and the random forests method should be recommended.

Landscape Segmentation
The first step towards mapping large areas on the basis of linearly operated sensors should be a detailed examination of the landscape settings. If the pattern of spatial variation of a soil property is expected to differ between geomorphological or pedological regions, which is often the case, they should be modelled separately). Therefore, landscape segmentation into smaller, more homogeneous plots is an important first step.

Landscape Patches
Following landscape segmentation derivation, a core patch within each landscape segment is required to reduce the area on which the geophysical sensors will be applied for intensive measurements and analysis, where quasi full coverage sensing is possible. In accordance with the landscape segmentation and generation of these patches, we used the soil map and land use map as input data.

Validation Samples
While calibration samples need to be drawn from within the patches, validation samples should be independent and collected outside of the patches. Therefore, validation was based on a decided number of fully independent soil samples based on a wLHS scheme to cover the geographical space.

3D Mapping
The main objective of 3D mapping activities is to develop and compare different methods of 3D DSM. We selected the depth function, basing our choice upon the trade-off between the simplicity of the function and its level of ‘best fit’. This is due to the fact that even though a very complex mathematical function should be able to describe a soil profile very well, its complexity can certainly become a problem during the mapping. The activities of the Partner CU were following:

  • Creation of 3D soil maps for several properties: soil compaction, clay content, silt content, sand content, SOC, S, N
  • Test of several method for 3D soil mapping
  • Test of different types of covariates (non geophysical and geophysical) for the use in 3D digital soil mapping

Cross Scale Mapping
The major aim of the Fuhrberger Feld study is to develop and test concepts that incorporate field scale geophysical data for landscape scale digital soil mapping. The final result should reflect an approach that can integrate field-/patch-scale geophysical sensing data (EMI and gamma-ray spectrometry) – in terms of cross-scale modelling – best suited for landscape scale predictions of soil properties. We focus on sand, silt, clay, and SOC at the depth intervals of 0-10 cm and 10-30 cm. For modeling purposes, we used the ConMap hyper-scale digital terrain features (Behrens et al., 2010) as the only predictors for landscape-scale digital soil mapping using Random Forests in this study. We validated the modeling results based on independent validation data.

Data Mining and Sensor Importance
An important part of multiple sensor data analysis in integrated measuring-mapping approaches (as in the iSOIL project) and which has not been widely discussed thus far is the analysis of feature or sensor importance. Random Forests provides built-in options to analyse feature importance which we used to analyse the influence of the sensor signal for modelling soil properties.
Clay is manly related to the ECa signals, whereas silt and clay are related to gamma-ray spectrometry signals. It is also interesting to analyze the (mostly continuous) changes in importance of one sensor and one soil property compared to depth. Analysing such effects and the interactions of sensor signals for mapping and interpreting soil properties is important for validation of the models, as well as for extracting new knowledge from integrated soil sensing and mapping studies.

iSOIL banner

Important Information

Project ended in November 2011

CWA 16373 "Best Practice Approach for electromagnetic induction measurements of the near surface" is published and available at CEN or DIN Berlin

iSOIL data are available at http://eusoils.jrc.ec.europa.eu/projects/isoil/data.html

iSOIL is a member of the SOIL TECHNOLOGY CLUSTER of Research Projects funded by the EC

European Flag
FP7