Detecting changes in soil water content response under climate extremes using long-term lysimeter data

Soil water content dynamics describe the response of soil functions to atmospheric forcing and provide insight into soil hydraulic properties and soil health. Abrupt changes in climatic conditions may lead to persistent shifts in this response, reflecting structural alteration rather than short-term variability. Detecting and reproducing such changes remains challenging, as most modelling approaches assume stationary soil properties and are not designed for long-term monitoring.

In this study, to detect persistent changes, we analyse multi-year lysimeter observations from the SOILCan network that is part of the TERrestrial ENvironmental Observatories (TERENO). For that purpose, we use a data-driven approach that combines a neural network with seasonal–trend decomposition. The model is trained on a lysimeter exhibiting stable soil water dynamics and subsequently applied to lysimeters of different soil origins exposed to contrasting climatic conditions. Differences between observed and modelled soil water content are tracked over time to test whether the soil moisture-climate relationship remains stable under changing conditions.

Persistent changes in soil water response are identified when model residuals exhibit a sustained bias over time, indicating a shift in the underlying soil–climate interaction. Based on this behaviour, soil dynamics are classified as stable, resilient, or changed. Application to the extreme drought of summer 2018 in Germany shows that while soil water dynamics are often preserved under typical conditions, extreme drought and exposure to new climatic regimes can induce lasting changes, even when soil texture remains unchanged. The proposed approach thus provides an early-warning capability for detecting emerging changes in soil hydraulic functioning from long-term monitoring data.