DEVELOPMENT OF A PHOTOVOLTAIC SYSTEM EXTRACTION INDEX FOR THE DETECTION OF LARGE PV PLANTS USING SENTINEL-2 IMAGES

The importance of solar power, known for its wide availability and low emissions, is underscored by the growing adoption of renewable energy, driven by environmental concerns and technological progress. Large photovoltaic (PV) plants require constant monitoring for efficiency and reliability. Remote sensing offers a cost-effective solution for accurately capturing plant size, shape, and location data. Satellite imagery, especially from open-source satellites such as Sentinel-2 (S2) and Landsat 9, has revolutionized remote sensing, enabling the development of machine-learning algorithms for PV system classification. While various spectral indices, such as the Normalized Difference Water Index (NDWI) and the Normalized Difference Vegetation Index (NDVI), enhance accuracy in water and vegetation areas, no specific index exclusively designed for PV extraction exists because of the diverse deployment settings of PV arrays. This study introduces a tailored Photovoltaic system extraction index (PVSEI) for S2 images in two regions known for large PV installations: Viterbo (Italy) and Seville (Spain). PVSEI combines different spectral bands to maximize the contrast between the solar panels and the surroundings. Employing Object-Based Image Analysis (OBIA) for accurate identification, multiresolution segmentation was used to create segments based on scale, shape, and compactness. The Decision Tree (DT) classifier consistently ranked PVSEI as the most effective. Accuracy assessment using the Overall Accuracy (OA), Kappa Index of Agreement (KIA), Producer Accuracy (PA), User Accuracy (UA), and F1 consistently yielded excellent results, with an OA exceeding 98%. KIA ranged from 0.74 to 0.82 for segmentated objects. Overall, PVSEI excelled in both study areas, with occasional challenges in distinguishing bare soil objects resembling PV systems.