Uso de machine learning para elaboração de mapas de fertilidade do solo

Abstract

The advance of Brazilian agriculture in recent years has been driven by advances in soil fertility research and applied technologies, which have enabled the efficient use of correctives and fertilizers, improving soil quality and agricultural productivity. In this scenario, Digital and Precision Agriculture have revolutionized agricultural practices, with the use of sensors, fertility maps, geolocation and homogeneous management zones,proving to be effective in increasing efficiency in the use of resources, optimizing the recommendation of inputs and minimizing environmental impacts. In this context, the aim of this work was to draw up soil fertility maps to determine and analyze management zones on a property located in the municipality of Frei Paulo, Sergipe. The characterization of the soil’s chemical attributes was carried out in an experimental area of 180 ha. An area with georeferenced points and sample grids were generated using the free software QGIS (3.28.2). There were 7 sample points, with 3 samples per hectare and two depths, 0 to 20 cm and 20 to 40 cm, and 4 sub-samples were collected 1 meter apart for each main sample. The data was submitted to descriptive statistics and the maps were interpolated using the Smart-Map plugin (version 1.4). According to the maps generated, distinct patterns were revealed between the two layers analyzed: pH, calcium, phosphorus and potassium showed higher values in the surface layer (0-20 cm), reflecting the influence of agricultural management and leaching. The saturation by essential cations indicated good surface fertility, but the low effective CEC at depth requires attention to fertilization.