Automatic detection of centre-pivot irrigation from high-resolution satellite imagery in northern Buenos Aires (2016-2023)

Abstract

Center-pivot irrigation enhances agricultural productivity and adapts to climate variability amid growing food demand and water scarcity. This study analyzes its expansion in northern Buenos Aires Province, Argentina, between 2016 and 2023 using Sentinel-2 satellite imagery from the Copernicus program and automated detection algorithms. Spectral indices such as NDVI, EVI, and WRI were calculated and served as inputs for the Grounding DINO model. This model combines computer vision and natural language-based artificial intelligence (AI) and was adapted to identify the circular patterns typical of center-pivot irrigation in a region characterized by continuous vegetation, which complicates automated detection. The developed methodology detected a 159% increase in irrigation systems during the study period, linked to agricultural drought events that drove their adoption. The model achieved an F1-Score of 73.2%, demonstrating its effectiveness in the Pampa Húmeda region. The results show that integrating remote sensing and AI enables efficient monitoring of this agricultural technology's expansion. Furthermore, the study highlights the need to continue developing AI models to improve the detection and characterization of geographic features, supporting sustainable water resource management.