Detecting the resilience of soil moisture dynamics to drought periods as function of soil type and climatic region

Abstract. Abrupt changes in climatic conditions and land management can cause permanent shifts in soil hydraulic response to climatic inputs, impacting soil functions and established soil–climate interactions. To quantify the resilience of soil water content dynamics after abrupt changes in environmental conditions, we present a model framework combining a neural network with seasonal trend analysis (STL). Using data from a series of lysimeters from the TERrestrial ENvironmental Observatories (TERENO) – SOILCan lysimeter network, we identified changes in soil water content responses after an extremely hot and dry summer in Germany in 2018. The model incorporates meteorological variables decomposed into seasonal and long-term components along with a categorical indicator of current moisture conditions. It is trained on data from a reference site with stable soil water content response and applied to lysimeters from multiple origins exposed to contrasting climates. By analysing annual residual patterns—particularly mean bias over time—soil water content state dynamics is classified as ‘stable’, ‘resilient’, or ‘changed’, reflecting whether the system maintains, recovers, or diverges from its original state. We found that soils preserve the response function to environmental forcing under typical conditions but exhibit structural change when relocated to new environments, even when soil texture remains constant. The proposed method offers a scalable and non-invasive tool for tracking changes in the response of soil water content to climatic change and provides early indicators of changes in essential soil functions and soil health status.