DES Seminar by Jose Padarian
14 March 2014
Provision of soil information for biophysical modelling.
Soil is an important actor in ecosystem processes and data that represents it, is not always available due to its intrinsic complexity. Several techniques have emerged to try to overcome this issue, including the use of pedotransfer functions (PTFs) and digital soil mapping (DSM).
The aim of this project is to derive a framework for addressing soil data need, using as example water holding capacity of soils: drainage upper limit (DUL) and crop lower limit (DLL).
PTFs with uncertainty assessment are not always available, hence there is a necessity to generate new ones and to identify if a PTF prediction is valid for a given soil domain. We selected Australia as example to generate a set of pedotransfer functions which predict soil water retention properties required by commonly-used biophysical models. PTFs were generated using symbolic regression and the fuzzy k-means with extragrades algorithm was used to estimate the uncertainty of prediction and identify when an observation is within the PTF data domain.
Using DSM techniques, soil avalilable water content was mapped at five depth intervals (0-5, 5-15, 15-30, 30-60, and 60-100~cm) with the help of different combinations of environmental information (topographic, climatic, soils, Landsat imagery, gamma-ray spectrometry) as covariates. The modelling techniques used were symbolic regression (GP), Cubist, and support vector machines (SVM). In addition, two averaging method were used to generate an ensemble model. The main focus was to balance model parsimony (number of covariates), accuracy (numerical performance) and realism of the visual representations (maps).
All welcome to attend
Location: Room 422, Biomedical Building (C81) ATP
Contact: Dr Uta Stockmann
Phone: 02 8627 1147