Digital soil mapping (DSM) is defined as the ‘computer-assisted production of digital maps of soil type and soil properties, by use of mathematical and statistical models that combine information from soil observations with information contained in explanatory environmental variables’. It is rapidly becoming the most popular soil mapping technique. One aspect that has received little interest so far is that the soil observations used for calibration and interpolation are themselves not error-free. The aim of this MSc research is to adapt DSM such that it can take measurement errors in soil observations into account. Application to a real-world case study, such as available from Namibia or Scotland, will be an essential part of the thesis work.
The thesis research begins with getting acquainted with DSM by studying a few key articles and application of DSM to example datasets (using R scripts). Next you will study a known method to incorporate measurement error in regression kriging and apply it to your selected case study. You will compare results with those obtained if measurement error is ignored and explain the differences, both for the resulting soil maps as well as for the associated uncertainty maps (i.e., regression kriging standard deviation maps). Next you will either continue the regression kriging road but work out and test how measurement error influences the calibration of the DSM model (including calibration of variogram parameters) or you will analyse how measurement error can be incorporated in DSM interpolation methods based on machine learning. One possible solution approach might be to use Monte Carlo simulation: sample repeatedly from the probability distribution of the uncertain measurements, run the machine learning method for each of these simulations, and integrate over all simulation results. This will work fine but it is computationally demanding and perhaps there is a faster solution. The MSc-research must also include a cross-validation of the final maps and their associated uncertainty maps. The results of this MSc research will likely be very useful to the ISRIC SoilGrids project (www.soilgrids.org).
- Learn about DSM, in particular regression kriging and machine learning for soil mapping
- Learn how measurement errors in soil observations can be included in regression kriging for DSM and analyse their effects through a real-world case study
- Develop and test new methodologies for including measurement errors in DSM
- Summarise how the ISRIC SoilGrids project can benefit from the findings of this research
- Digital Soil Mapping, e.g. Minasny and McBratney (2016)
- Regression kriging with data contaminated by measurement errors, e.g. Delhomme(1978)
- Machine learning for soil mapping (e.g. SoilGrids publications)
- Solid background in statistical modelling, such as obtained through the Spatial Modelling and Statistics course
- Experience with programming in R
- Affinity with Digital Soil Mapping
Theme(s): Modelling & visualisation