Mapeamento digital de solos e predição de atributos utilizando Machine Learning e lógica fuzzy na bacia do Ribeirão Inhaúma, Iconha/ES

Abstract

Obtaining information about the soil and its attributes is essential for proper planning of the use and occupation of it. Conventional mapping and routine laboratory analysis are the most used methods in Brazil for soil spatialization and obtaining information about its attributes. In the case of mapping of soil classes this method is based on mental models developed by the pedologist, which makes it highly subjective, dependent on the professional experience and difficult to replicate, while the laboratory analyzes are expensive and present the possibility of generating impacts through the use of chemical reagents. Therefore, new computational techniques have been presented as a tool for mapping classes and attributes of soils, providing greater speed and decrease of cost in relation to conventional methods. Thus in this work the use of these techniques was studied in Ribeirão Inhaúma, in Iconha / ES. In this study, we used 10 different models for the digital mapping of soils. In this case, 10 models were tested, which were efficient, with emphasis on fuzzy logic, Randon Forest, Ranger, Extreme Gradient Boosting and Weighted Subspace Random Forest , chapter 1 also compared this model to conventional mapping, finding 24% agreement between the maps. Chapters 3 and 4 tested the use of several models for mapping attributes, highlighting in chapter 3 the Randon Forest method and in 4 the use of spectral techniques for mapping organic carbon, phosphorus and clay with good accuracy. Chapter 5 analyzed the degree of correlation between information collected by remote sensing through aerogeophysical data and data obtained by fluorescence analysis using an X-ray analyzer, comparing the values found with the land use and occupation map. It was possible to identify the correlation between the use and occupation of soils with gamma spectrometry and temperature data. High correlation with elevation and low correlation with thermofluorescence data was observed in the aerogeophysical data of gamma spectrometry, indicating that data from the thermofluorescence analysis can not replace those from the gamma spectrometry and vice versa, for the purpose of analyzing characteristics of the land. Chapter 6 proposes an initial study on the possibility of using fluorescence as a tool in the determination of Fe, Al, Si and Ti oxides in order to reduce the impact on the environment caused by the chemical reagents used in the sulfuric attack and to increase the speed analysis and preparation of the sample. X-ray fluorescence analyzes were promising for the determination of the oxides of Cambisols and Oxisols, and a larger number of samples were required using the two methods, in order to prove the correlation between the contents found