A dependency learning approach to delineate site-specific management zones

Precision Agriculture offers solutions based on scientific knowledge to help farmers make better decisions in the agricultural cycle. In the initial phase of this cycle, it is crucial to identify an adequate delineation of site-specific management zones to optimize inputs and enhance crop yields. Based on a soil property that can be chemical, physical, or biological, the delineation of a site-specific management zone problem consists of minimizing the number of zones reaching a homogeneity level according to the soil property selected. Additionally, the proposed solution should consider the use of agricultural machinery to create orthogonal shapes, thereby assisting farmers with delineation tasks. In this work, we propose a novel approach, not previously presented in the literature, for the Site-Specific Management Zone problem, based on dependency trees that account for the various interactions between variables. Additionally, to minimize the number of zones, we incorporated into the objective the minimization of the variances within each zone. The proposed new approach was evaluated using two different objective functions: one aimed at minimizing the number of zones, and the other focused on minimizing both the number of zones and the sum of the variances within those zones. The computational results show an improvement of the 100\% for instances of small and medium size, considering the best literature approach to generate orthogonal site-specific management zones.