Recent advances in sensing, machine learning and process modeling have expanded on what is possible within Pedometrics, a branch of research at the interface of soil science and data science. However, many methodological challenges remain. This session focuses on methodological advances that improve the reliability, reproducibility and pedological soundness of digital soil mapping, spectroscopy and spectral transfer functions, pedo-transfer functions or other predictive statistical applications in soil science. We specifically invite contributions that address the underlying statistical, computational and conceptual tools needed to
1) produce pedologically consistent multi-property, multi-depth, multi-timepoint soil maps,
2) create suitable sampling designs for current state-of-the-art mapping approaches,
3) provide physically or chemically meaningful pedo-transfer functions,
4) quantify and propagate uncertainty in space–time and for multivariate outcomes,
5) provide methods for up- and downscaling (change-of-support) and realistic aggregation of uncertainties,
6) contribute to explainable machine learning and robust soil science-informed model diagnostics, and
7) integrate mechanistic soil process knowledge with data-driven algorithms into so-called hybrid approaches or soil science informed machine learning.
The session welcomes contributions that develop, compare, or benchmark novel approaches or present (open) software implementations that strengthen the foundations of pedometrics and enable robust and reproducible frameworks. We welcome developments towards all kinds of applications such as global to local soil health monitoring, soil function assessments or for farm level and forest management. The session aims to bring together statisticians, process modelers, computer scientists and pedologists to discuss methodological solutions that are general, reproducible and transferable across regions and applications.