Low-Cost Machine Learning Approaches for Evaluating Irrigation and Vermicompost Impacts on Ground Cover Plants

Water scarcity is a major constraint to expanding green spaces in arid and semi-arid regions. In urban landscapes, irrigation for ground cover plants represents one of the largest portions of outdoor water use; therefore, improving its efficiency is essential for balancing water demand and reducing water stress at both city and basin scales. However, lawns and green areas provide important aesthetic, social, and psychological benefits to residents, creating a challenge between conserving water resources and maintaining urban greenery. To address this challenge, a field experiment was conducted to evaluate alternative ground cover species, including Phyla , Frankenia , Oxalis , and Bermudagrass, and to assess the role of vermicompost in improving plant performance under different irrigation regimes. Treatments included three irrigation levels (100%, 75%, and 50% field capacity) and two soil conditions (with and without vermicompost). To ensure objective plant evolution while reducing labor-intensive field work and enabling whole-plant scale monitoring, Fractional Vegetation Cover (FVC) was estimated using classical image processing and advanced deep learning models. Deep learning achieved high segmentation accuracy (~ 91%) and showed strong robustness to environmental variability, outperforming traditional methods. Results showed that vermicompost significantly enhanced growth performance and relative water content under both full and deficit irrigation. Oxalis and Frankenia were more sensitive to water stress, while Bermudagrass and Phyla demonstrated greater drought tolerance. Overall, the findings suggest that drought-resilient ground covers, supported by AI-based monitoring tools, can reduce outdoor water consumption, improve resilience to water stress, and provide a sustainable alternative to high-water-demand plants in urban landscapes.