Soil variability across scales and its drivers: A multi-proxy approach

Ing. Jessica Reyes Rojas

Doctoral thesis

Abstract:

Variability of the soil properties fundamentally influences many aspects of land use, soil management, soil threats and protection. It can be observed and assessed at different vertical and horizontal scales, driven by various factors and their interactions: i) microscale of a soil profile determined by the interplay of biological, chemical and physical factors, and specific mechanisms such as leaching, accumulation, redox processes, bioturbation or sedimentation, ii) local scale with significant interaction of soil and relief, iii) regional and global scale governed by lithology, geomorphology and climate. Human activities affect the soil variability at all these levels. The thesis aims to i) describe soil variability at different scales (micro, local and large scale), ii) to determine the changes in the intensity of individual factors of soils variability at the spatial scales studied and iii) to test different methods of digital soil mapping and image spectroscopy, including progressive methods of machine learning, for the soil variability assessment. The results showed that soil variability at different scales was explained by proxy indicators that are highly scale-dependent. At the local scale, lithology, hydrological conditions, and land-use history are the primary drivers of soil heterogeneity. At the large scale, wider patterns associated with substrate, climate and topography dominated. Testing modern digital soil mapping techniques coupled with hyperspectral imaging and machine learning showed that the random forest nonlinear model provided the best performance for both SOC (R² = 0.75) and CaCO₃ (R² = 0.76) content. This study demonstrated that a comprehensive assessment of soil variability incorporates classical methods that provide detailed insights into soil formation factors and processes, sedimentation history, and stratigraphy. Moreover, this investigation showed that the variability of soil properties at different scales was driven by an interplay of factors whose relative importance shifts in intensity in function of scale. However, land use history and land management practices consistently influence soil variability across all scales. The application of machine learning models that further enable mapping of soil properties demonstrates the potential of these techniques for predicting and quantifying soil variations, while the use of imaging spectroscopy offers new opportunities for characterising soil properties and processes at the microscale level with a potential for assessment at local and large scales.

Keywords

Micromorphology, Soil spectroscopy, Soil legacy data, Machine learning, Indicators, Mineralogy

Language used: English

Date on which the thesis was submitted / produced: 25. 8. 2025

Thesis defence

• Supervisor: doc. RNDr. Tereza Zádorová, Ph.D.
• Reader: Bořivoj Šarapatka, Bc., externi, Lukáš Brodský, externi, Lubica Pospíšilová, externi

Czech University of Life Sciences Prague

Faculty of Agrobiology, Food and Natural Resources

Doctoral programme:
Exploitation and Conservation of Natural Resources