Aplicação de propriedades eletromagnéticas no diagnóstico da compactação do solo

Abstract

Soil compaction is a process in which pores are reduced, due to the relocation and approximation of particles, making the soil denser, being the main factor in the physical degradation of agricultural soils. The most accurate methods for diagnosing compaction require equipment and trained personnel, in addition to being time-consuming and laborious. Alternatively, geophysical methods are fast and practical, non-destructive and can represent an effective alternative for this purpose. The electromagnetic properties of the soil are related to several attributes of the soil, and point out patterns that can be interpreted regarding changes in the soil structure. Therefore, the objective of this work is to diagnose compaction based on electromagnetic properties of the soil. The experiment was carried out at Fazenda Muquém – UFLA in a typical dystrophic Red Yellow Oxisol with a clayey texture, with the experimental design in 3 randomized blocks in a 3x2 factorial scheme (3 levels of compaction and 2 soil management). Compaction levels were obtained through additional compaction using a tractor, and the managements were Subsoiling (S) and Control (C) – Absence of subsoiling. Undisturbed samples were collected to evaluate soil density (Ds) and porosity, and deformed samples were collected for chemical analyzes and Magnetic Susceptibility (SM). In the field, soil resistance to penetration (RP) and electrical resistivity (ρ) were measured. RP and ρ were evaluated under different soil moisture conditions and up to 60 cm depth. Correlations were carried out between the different depths evaluated, in which the input data were data on electromagnetic properties (ρ and SM), and as auxiliary input data the results of chemical analyzes of the soil were used, and as output data, we will have the physical attributes that serve as a basis for diagnosing soil compaction, namely Ds, RP and macroporosity. The modeling was carried out based on 4 prediction models (Linear, Generalized Linear, Random Forest and Categorical Random Forest), the latter two using Machine Learning. Machine Learning models were presented as more promising, both for the prediction of physical properties and the categorical prediction of whether or not the soil was compacted, the soil moisture condition and the evaluation depth influenced the results, with greater emphasis on the humidity in the range of 28% and depth of 20-30cm.