Paisagem do Parque Nacional da Serra da Canastra e de sua zona de amortecimento - MG: análise de padrões espaciais a partir de árvore de decisão e métricas de paisagem

The National Park of Serra da Canastra (PNSC), located in the southwestern portion of the state of Minas Gerais, is an area of great environmental relevance, considered as a "priority area of flora and fauna preservation. It is the second largest protected area of Minas Gerais, and in accordance with the management plan , the landscape is composed of preserved and anthropogenic environments and its surroundings are being increasingly degraded by different types of its soil uses, fragmenting the landscape, increasing the anthropogenic pressure on the PNSC and causing environmental problems such as erosion, soil compaction and water contamination. It is in this subject, that this study takes place: the proposal is to construct a methodology for landscape analysis based on conceptual approaches of systems analysis in geography and landscape ecology, associated to computational methods of spatial analysis to identify patterns, which result, may support the planning of this landscape, to ensure the maintenance of their ecological viability. Based on the landscape ecology theory, metrics that describe the composition and configuration of the landscape, were obtained through its spatial pattern, allowing the identification of the main forms of anthropic pressure and the fragmentation mapping of their degrees, and based on the systemic analysis of the landscape , the adoption of the proposed area subdivision in Landscape Units . To identify those units, were considered the following: geology (CPRM), hypsometry (SRTM), slope (SRTM), hydrographic density (SRTM), soils (UFLA) and vegetation cover and soil use (mapping trough Landsat 8 images rating). The integration of these elements, were processed through the intersection of their respective layers in ArcGIS 10.1 software, generating 151.646 combinations of information. These were subjected to a process of data mining, with classification based on decision tree, at Weka Explorer software, using J48 algorithm. The classification showed an overall accuracy of 81.92% and a very good performance , according to the Kappa Index value. The Associations created, indicated subdivisions in the area, whose boundaries were refined considering homogeneity and contiguity, generating the Landscape Unit (UPs) of the Canastra's Sierra National Park and its Buffer Zone. The methodology proved to be effective, because the units were identified, demarcated, and characterized according to their physical attributes support, biotic, soil use and its structure was quantified from landscape metrics that were calculated. It was then possible, to identify the main weaknesses of this landscape, pointing out aspects that require attention to public agencies and the public, focusing on sustainable management of the protected area and its buffer zone .