Mapeamento digital de solos no estado do Mato Grosso do Sul a partir de dados legados e modelos preditivos baseados em árvores

Abstract

Soils are a natural resource of great relevance, mainly due to their importance in the production of food, the support of biomes and in the storage of water, assuring the replacement of springs and water sources, as well as other environmental services. Thus, the knowledge about the soil properties and their distribution in the landscape is important for their management and for the territorial planning. The principal way to obtain information about the soils and its distribution is through soil surveys. The main hypothesis of the study is that from the techniques of digital mapping by using reference areas it is possible to predict the spatial distribution of soil units. Also, through the selection of predictor variables and the evaluation of quantitative predictive methods, the soil survey can be improved, reducing the subjective character of the interpretation, besides giving a quantitative character to the final product. The objective of this study was to evaluate the efficiency of tree-based predictive methods, Random Forest and Decision Tree, for the extrapolation of soil mapping unities in the municipalities of Nova Alvorada do Sul and Rio Brilhante, from 46 profiles described in Sidrolândia and Campo Grande, both located in Mato Grosso do Sul state. The approach through tree-based models enabled a quantitative evaluation of the factors involved in pedogenesis, which contributed to a better understanding of each factor and its direct contribution to soil formation. The use of the reference area proved to be adequate for the process of learning the morphometric patterns, as well as for the extrapolation of the mapping units to the entire area. Both predictive models tested proved to be quite efficient in the extrapolation of the mapping units, and the Random Forest model presented the best predictive performance, in all statistical indices evaluated, in relation to the Decision Tree model. The models based on trees can contribute to the knowledge about the soil formation factors and to qualify their contribution in the pedogenesis, as well as in the understanding of the pedoambientes and distribution of the classes of soil in the landscape; also, to support the mapping of similar areas not yet surveyed.