Reversing forest regression: Effects of shrub removal and grazing on thermophilous oak forests herb layer (Central Poland)

Description

Abstract Thermophilous oak forests rank among the most species-rich forest ecosystems in Central Europe. Preserving their centuries-old, human-shaped biodiversity now requires active management. This study evaluates the effects of three restoration treatments – shrub removal, controlled grazing, and their combination – on the herb-layer composition of an oak forest. Research was conducted on permanent plots before and after three consecutive years of treatment, assessing changes in species frequency in relation to their functional traits. All treatments contributed to partial restoration of the characteristic herb-layer composition, with the strongest response in the combined variant and the weakest in the control plots. Initial thinning, applied to enhance light availability, increased species frequency across all variants, though effects were much stronger under management. Species responding most positively included hemicryptophytes and geophytes, mainly CSR strategists, shaded forest and oak forest species, light demanding species and species tolerant to disturbances. All this changes indicate a shift in the composition of the herb layer towards thermophilous oak-forest. Overall, the combination of shrub removal and grazing proved most effective in restoring the herb-layer composition, while thinning alone produced similar but weaker effects. The historical proximity approach may support short-term forecasting of changes in species frequency.

Steps to reproduce

To ensure taxonomic consistency and enable functional trait-based analyses, raw field data were standardized according to the World Flora Online (WFO) backbone checklist. Species occurrence data from 1,200 subplots were consolidated into a unified dataset, where each taxon was assigned specific functional traits and ecological indicator values. These included Raunkiaer life forms, Grime’s life strategies, phytosociological preferences, and forest density indicators, alongside updated Ellenberg indicator values (EIVs) and disturbance tolerance metrics. This integrated framework combined experimental observations from 2016 and 2019, forecasted successional states for 2028, and reference benchmarks for Potentillo albae-Quercetum and Tilio-Carpinetum. This unified hub served as the foundational input for all subsequent ordination and forecasting procedures. The initial phase of the statistical analysis quantified the biological response to restoration treatments between 2016 and 2019 by evaluating species turnover, classifying taxa as disappeared, persisting, or new. To visualize the ecological gradients, we employed Non-metric Multidimensional Scaling (NMDS) through three distinct models. A micro-scale 6D ordination was used to stabilize high beta-diversity and project functional responses, while an independent 2D ordination focused on treatment-specific successional paths. Finally, a consolidated reference ordination, utilizing Chord transformation and Bray-Curtis dissimilarities, provided a benchmark for experimental and forecasted trajectories. Biological drivers were identified by fitting Community Weighted Means (CWM) for environmental variables and disturbance indicators onto the ordination space, with significance assessed via 999 permutations. Long-term vegetation dynamics were simulated up to 2028 using a multi-step Ensemble Forecast engine. This predictive approach integrated Random Forest, Gradient Boosting Machines, Neural Networks, and Cubist algorithms within a 5-fold cross-validation procedure. The models were trained on observed turnover using functional traits and a dynamic Local Context Frequency (LCF) variable as predictors. The simulation was executed through three 3-year cycles (2022, 2025, 2028), with final projections generated using a consensus mean approach to minimize algorithmic bias. To quantify restoration success, the effectiveness of experimental variants was benchmarked against contemporary reference releves. Success was measured as the increase in the inverse of the mean Euclidean distance between experimental plots and reference centroids in NMDS space, where positive values indicated convergence toward the historical template.

Institutions

• Jan Kochanowski University

  Świętokrzyskie, Kielce
• Warsaw University of Life Sciences

  Mazovia, Warsaw
• Cardinal Stefan Wyszynski University in Warsaw

  Mazovia, Warsaw
• University of Warsaw

  Mazovia, Warsaw

Categories

Funders

• European Comission

  Grant ID: LIFE13 NAT/PL/000038 | Acronym: ZSiNPK_Kielce_LIFE_PL

Licence