Weed Distribution Mapping and Site-Specific Characterisation in Lentil Fields Using AI and Geostatistics

Sustainable weed management in lentil (Lens culinaris Medik.) production requires accurate knowledge of weed spatial distribution and the site-specific factors influencing infestation patterns. Limited pre- and post-emergence herbicide options for pulse crops underscore the need for precise pre-emergence applications guided by spatially explicit weed mapping. This study integrates YOLOv11 object detection with Slicing-Aided Hyper Inference (SAHI) framework, advanced geostatistical techniques, and soil electromagnetic induction measurements to develop a comprehensive precision agriculture framework for species-specific weed management. High-resolution drone imagery (4K, 1.5 m altitude) was systematically collected across a 3.42-hectare commercial lentil field in Chile’s Central Irrigated Valley, complemented by satellite-derived vegetation indices (Sentinel-2 NDVI) and soil electrical conductivity mapping (EM38-MK2). The YOLOv11 model achieved robust detection performance with F1-scores of 0.87 for lentil crops and 0.84 for Ambrosia artemisiifolia, the dominant weed species, enabling species-specific density mapping at 5 m × 5 m resolution. Geostatistical analysis revealed significant spatial autocorrelation in weed distributions (Moran’s I = 0.667, p < 0.001) with strong bivariate associations between weed density and environmental variables, particularly soil electrical conductivity (spatial r = 0.633) and vegetation indices (spatial r = 0.818). Fuzzy clustering successfully delineated four distinct management zones, with 31.9% of the field requiring critical intervention and 51.7% suitable for maintenance-level management, enabling potential 35-50% reduction in herbicide use while maintaining effective weed control. The demonstrated multi-scale approach enables transition from satellite-guided field reconnaissance to ultra-precise drone-based treatments, supporting cost-effective implementation across extensive agricultural areas. This integrated AI-geostatistical framework addresses critical limitations in current precision agriculture technologies by combining high-accuracy species detection with spatial analysis capabilities that enable predictive modelling and evidence-based management optimisation, establishing foundations for scalable precision weed management in sustainable agricultural production systems.