Intelligent division of agricultural farms into homogeneous management zones for precision irrigation using remote sensing and artificial intelligence

Precision irrigation requires accurate spatial delineation of agricultural fields into homogeneous management zones to optimize water use and improve crop productivity and support water stress mapping. This study proposes an integrated and scalable framework for dividing farms into homogeneous zones by combining multisource remote sensing data, terrain-derived indicators, geospatial analysis, and machine learning techniques. The methodology follows a three-level environmental segmentation strategy: (i) separation of agricultural and non-agricultural areas using supervised artificial intelligence models, (ii) topographic-based subdivision to account for terrain-driven hydrological variability, and (iii) soil-based zoning using in situ soil studies when available, or global SoilGrids products otherwise. Sentinel-2 optical imagery, terrain derivatives, and soil attributes were processed within Google Earth Engine and Python-based machine learning workflows. Several machine learning classifiers, including Random Forest, XGBoost, and LightGBM, were tested for agricultural land classification. Among them, LightGBM performed best, achieving an overall accuracy of 94.1 % and an F1-score of 0.92 for the agricultural class. The results demonstrate the feasibility and internal consistency of the proposed workflow at the farm scale. Because the framework relies on standardized and widely available datasets and clearly defined processing steps, it is methodologically reproducible and designed to facilitate application in comparable agricultural contexts.