Adaptive neuro-fuzzy inference system for predicting sandy soils bulk density

Soil compaction, often the result of intensive mechanization, is a major concern because it increases the bulk density of the soil, fundamentally altering its structure and reducing its capacity to support essential ecological and agricultural functions. Therefore, soil monitoring is essential, particularly through its main quality indicator, bulk density. However, the analysis of this attribute requires the collection of soil samples with subsequent laboratory analysis, which demands time and qualified labor. Thus, this study aimed to develop a neuro-fuzzy model to estimate the bulk density of sandy soils based on variations in the proportion of its particle size fractions. To achieve this, undisturbed samples were collected from soils in four municipalities in northern Bahia, located along the shores of Lake Sobradinho. The proportion of particle size fractions in the composition of each sample was used to develop models that integrate fuzzy logic and neural networks. They employed the Takagi-Sugeno inference method and hybrid learning algorithms for parameter adjustment and bulk density predictions. The performance of the models was evaluated using error statistics, with the model with the highest accuracy in density predictions being selected, achieving an R² of 71%. It is important to emphasize that the neuro-fuzzy model did not require data stratification and can be applied to estimate the sandy soil density of any soil type sampled.