Seletividade ambiental de castanheiras (Bertholletia excelsa Bonpl.) no estado do Amazonas

Abstract

The study aimed to analyze the relationship between physical and chemical soil attributes of native Brazil nut trees in Amazonas, in order to contribute to the expansion of knowledge about the natural environments in which Brazil nut trees occur. Soil samples were collected at Lago das Pedras Community – Barcelos/AM, Jutica Property – Tefé/AM and Jatuarana Community – Manicoré/AM. In each area of the experiment, soil samples were collected in the 0-20 cm surface layer of a permanent plot (300 x 300 m), obeying the distances of 50 m between lines and 30 m between points, totaling 60 samples. In each of the plots of the areas under study, an inventory of all Brazil nut trees with DBH ≥ 10 cm was also carried out. Soil data were evaluated by descriptive statistics, followed by Geostatistical analysis, using ordinary Kriging to analyze the data from the Lago das Pedras Community and Jutica Property and interpolation by IDW to analyze the data from the Jatuarana Community; in the univariate study of the DAP, a slicing DAP ≥ 50cm and DAP < 50cm was used by means of Indicative Kriging for the data from the Lago das Pedras Community and Jutica Property and for the DAP data from the Jatuarana Community, we followed the slicing criterion of the data using the median with DBH < 145.4 and DBH ≥ 145.4. Subsequently, the multivariate process of spatial regression was carried out using Machine Learning, through the Random Forest (RF) Algorithm for spatial predictions. The results showed that most of the attributes studied have a moderate degree of heterogeneity. The RF prediction generated maps of probability of occurrence of Brazil nut trees that contribute to characterize and understand the spatial distribution of Brazil nut trees with DBH ≥ 50cm and DBH < 50 cm in the areas. The RF model indicated the main attributes of the soil in the regions under study, generating maps of the most important variables in the areas, where, according to the RF model created, the most relevant attributes in the three regions are total sand and sodium.