​​Rush dominance in wet pastures - spatial analysis, trajectories and drivers​

Abstract

​​Lowland wet pastures are biodiversity-rich ecosystems but are increasingly threatened by the expansion of rush (Juncus spp.), which can dominate and displace species of conservation value. This study quantified the spatial extent, temporal dynamics, and key drivers of rush distribution in Kingcombe Meadows Reserve, Dorset, using high-resolution aerial imagery (2005, 2020) and deep learning methods. A pretrained U-Net convolutional neural network (CNN) was trained on randomly selected annotated image samples to produce reserve-wide rush maps, achieving overall accuracies above 91%. Bias-corrected area estimations showed that rush cover increased substantially over the study period, rising from 7.6% in 2005 to 20% in 2020, indicating both localised spread and establishment in previously unoccupied areas. Topographic analysis revealed that rush occurred predominantly at lower elevations (114-158 m) and on gentle slopes (less than 8 degrees), where hydrological conditions favour persistence. Current grazing pressures showed no significant relationship with rush distribution, suggesting that grazing alone could not explain recent expansion patterns in the reserve. This study demonstrates that CNN-based segmentation provides a powerful framework for fine-scale vegetation mapping and ecological assessment. The results highlight the need for integrated management strategies that combine grazing, hydrological, and topographic considerations to control rush encroachment effectively. More broadly, the approach offers a transferable tool for conservation practitioners and policymakers to monitor invasive or encroaching species in wet grasslands and other sensitive habitats, supporting evidence-based management and enhancing biodiversity conservation at a landscape scale.​