Detecting Vegetation Patterns in Satellite Images: SILVIA, a Segmentation-Based Approach

One of the key indicators of ecosystem dynamics is the presence of vegetation patterns, which provide valuable insights into the interactions between biotic and abiotic factors. Recognising these patterns is critical for ecosystem resilience studies; automated analysis of satellite imagery is a cornerstone in environmental monitoring, requiring the capability to detect objects on the Earth’s surface. One of the first steps in image object detection is segmentation. This study presents SILVIA (Segmentation and Identification for satelLite Vegetation pattern ImAges), a novel pipeline that leverages the Segment Anything Model (SAM) for automated segmentation of vegetation structures in satellite images. SILVIA generates segmentation masks for spectral images and vegetation index-derived representations; masks can be filtered using confidence thresholds and spatial criteria tailored to the morphology of Fairy Rings. SILVIA, after the multistep filtering and characterisation process, also implements image fusion mechanisms, providing a more robust approach regarding false-positive and false-negative segments: this constitutes the main original contribution of this work. The results demonstrate that this approach effectively identifies fairy rings with minimal human intervention, improving the scalability and reproducibility of vegetation pattern monitoring.