Integrating information on soil forming processes in digital soil mapping

PhD project Marcos Angelini together with ISRIC on digital soil mapping.

Resulting PhD thesis: Structural equation modelling for digital soil mapping (2018)

Problem definition and objective

Since the 1950’s, soil spatial information has been delivered through soil maps, which have been produced by soil surveyors with conventional techniques. However, the poor reproducibility of these methodology, the lack of uncertainty information, as well as a new demand of finer soil map resolution have been the main reasons for development of new methodologies for soil mapping. These methodologies are referred to as Digital Soil Mapping (DSM), which is defined as ‘the creation and population of geographically referenced soil databases generated at a given resolution by using field and laboratory observation methods coupled with environmental data through quantitative relationships.’ Current DSM methods are mostly empirical that confers some limitations to the technique. One of these is the difficulty to predict a large number of soil properties simultaneously, while preserving the relationships among them, and to include pedological knowledge in prediction models. This project investigates if structural equation modelling (SEM), a technique used in various fields within the earth and environmental sciences, can fill these gaps in DSM. Then, we will extend DSM with soil forming process information through the development, calibration, application and validation of a structural equation model (SEM).

Figure 1: Above: Research area in Argentina and below: fieldwork in progress. 


The first objective is to build and calibrate a SEM for the Argentinian Pampas region. The purpose is to develop a simple SEM to show how it can be integrated within a DSM approach. Model development will be described in order to give a guidance for applying SEM for soil properties prediction. Validation of the model predictions will be done using independent data from 100 locations collected with probability sampling. The second objective is to investigate the model complexity, and data availability (a data-rich versus a data-poor situation), and calibration method (Bayesian versus maximum likelihood) on model accuracy. The third objective is to compare the performance of a SEM with legacy maps and empirical DSM models through independent validation. Thereby, this objective will test if SEM can be an alternative to empirical DSM methods. The fourth objective is to investigate if the model can be applied to other regions with similar soil-landscape features. Both, the model developed under the first objective and a model adapted to the conditions in the new study area will be tested to identify which interrelationships can extrapolated and which cannot. Finally, the fifth objective is to explore how to present and deliver soil information generated with a SEM. Because it is expected that SEM outcomes provide sufficient information, a prototype soil information system will be implemented to reconstruct the pedon at any location as well as to predict soil classes as a function of space.

Start project: August , 2013

End project: September, 2017


SGL Soil Geography and Landscape, ISRIC, INTA