Relações solo-paisagem e modelo de predição de solos da bacia do córrego Marinheiro, Sete Lagoas (MG)

Abstract

Located in the peri-urban area of the city of Sete Lagoas, Minas Gerais, Marinheiro stream watershed is characterized by high relevance in the geographical context in which it inserted, been the target of impacts of urban expansion and speculation on its water potential. The aim of this work is the study of soil-landscape relationships of Marinheiro stream watershed, Sete Lagoas (MG), and the proposition of a soil prediction model for the watershed. It conducted a semi-detailed conventional soilmapping and an analysis of interactions between elevation, slope, lithology, phytophysionomy and soil from crossing the thematic maps. The digital soil model was built from the weights of evidence method and executed on the modeling platform open source Dinamica EGO. It was used a sample of 42 soil profiles and 157 checkpoints analyzed and described in the fieldwork as the result variable. The conventional soil survey comprised 21 mapping units and 19 soil suborders, featuringa high variability of classes for an area of 14.8 km². Latossolos (Brazilian soil classification systems classes) were the mapped classes of higher spatial occurrence in the watershed, while Cambissolos Háplicos (Brazilian soil classification systems classes) presented the highest participation among the mapping units. From an initial analysis of the relations between lithology, phytophysiognomies and topographic variables derived from a digital elevation model of high spatial resolution, It was possible to identify a high variability physiographic in the watershed. Irregular relief in the upstream portion and at points located in the watershed showed a higher association with the granite-gneiss of Belo Horizonte Complex and calcareous of the Sete Lagoas Formation, while the more gentle slopes areas predominated over siltstones materials of Serra de Santa Helena Formation. The different lithologies in general showed greater influence on physical andmorphological properties of soils than in the classes. The digital soil model resulted in a probability map for each 16 soil classes identified, which were combined into a single map, resulting in 12 mapping units (MU). Latossolos, Argissolos and Cambissolos (Brazilian soil classification systems classes) were the most frequent soil on the MU. Latossolos was the class with higher participation ( 50%). Whencompared visually, the predicted and conventional soil maps showed similarities between the outlines and the distribution of mapping units. The predicted map validation has resulted in a global accuracy of 71,4%, demonstrating the efficiency of the model for predicting the spatial occurrence of soil classes. The digital mapping proved to be a promising technique for enabling its reproducibility for regions without information and with similar physiographic, optimizing resources and time.