Agrominerais e sensoriamento proximal: avaliação e predição de atributos químicos em solo ácido

Abstract

The use of agrominerals in the agricultural sector has emerged as a promising alternative for supplying macro- and micronutrients to soils, thereby positively influencing the nutritional quality of food. Through rock fertilization, a technique that employs agromineral rock powders, it is possible to enhance soil fertility while reducing the reliance on chemical inputs. The present research aimed to analyze the chemical and spectral behavior of soils subjected to incubation tests with carbonatic phosphate rock powder as a sustainable alternative for soil remineralization. The methodology comprised three main stages. The first stage involved field activities, including the collection of rock material and soil samples from the municipalities of Irecê and Feira de Santana, Bahia, Brazil, respectively. The second stage consisted of laboratory procedures, including sample preparation, chemical, X-ray diffraction (XRD), and spectral data acquisition, recommendations for phosphate fertilization using rock powder (RP) and monoammonium phosphate (MAP), and the establishment of experimental units. A total of 105 experimental units were assembled, including 50 units for each material and five control treatments, with applications of two materials (RP and MAP) and five progressive doses for each source. The third stage involved monitoring and monthly sampling over a 120-day incubation period, with chemical and spectral analyses conducted every 30 days, followed by database construction, data analysis, and interpretation. The results were presented in two scientific articles. The first article aimed to evaluate, over time, the chemical changes in acidic soils incubated with increasing doses of carbonatic phosphate rock powder and monoammonium phosphate fertilizer, and their effects on soil fertility. Temporal analyses and graphical representations of soil pH and macronutrient behavior were generated. The results indicated: (i) variations in soil pH as a function of material, dose, and incubation time, with RP treatments exhibiting the most favorable pH values; (ii) other chemical parameters, such as sum of bases (SB) and macronutrients, also showed improved values in RP-treated samples, although nutrient imbalances were observed, with predominance of Ca and P due to the intrinsic composition of the incubated rock powder. In contrast, MAP-treated samples intensified soil acidification and gradually reduced nutrient concentrations. Overall, the most favorable soil fertility conditions were observed in samples incubated with rock powder. The second article aimed to apply predictive models—Partial Least Squares Regression (PLSR) and Random Forest (RF)—using spectral data to estimate phosphorus and pH in soils incubated with different phosphate sources. Six spectral preprocessing techniques were applied: reflectance (R), smoothing (SMO), standard normal variate (SNV), multiplicative scatter correction (MSC), continuum removal (CR), and absorbance (AB). The results showed that: (i) for soil pH, both models exhibited good predictive performance, likely due to the relatively low variability in chemical behavior, which contributed to greater homogeneity and improved model accuracy; (ii) phosphorus prediction required spectral preprocessing, mainly due to its higher variability in chemical concentrations; (iii) PLSR residuals for pH showed lower bias and more stable error distribution, whereas RF models for both pH and phosphorus exhibited error patterns associated with the chemical behavior of the samples. Consequently, PLSR performed better for pH prediction, while RF showed superior performance for phosphorus estimation. In addition, the predictive models demonstrated an innovative alternative for evaluating agromineralbased inputs. Thus, this study successfully integrated sustainable soil management practices with analytical radiometric techniques, contributing to the development of research approaches aimed at minimizing environmental impacts.