ESTUDO EM DUAS UNIDADES DE PAISAGEM DA BACIA HIDROGRÁFICA DO RIO PITANGUI/PR MEDIANTE ESTATÍSTICA MULTIVARIADA E ANÁLISE ORIENTADA A OBJETOS

Abstract

Landscapes have structural and functional uniqueness, especially when are inserted in spaces whose dynamics are substantially different uses. To analyze a landscape, it is necessary understanding that it is the result of multiple relationship between society and nature, in which both are considered a set of elements that interact with each other. Currently, there are many methods for analysis and characterization of landscapes, however, at a massive rate, the techniques of GIS (Geographic Information Systems) and RS (Remote Sensing) have impacts on the production and availability of cartographic products, minimizing costs and optimizing validation field. One technique associated with SR and GIS data is the Multivariate Statistical Analysis of Principal Components, in this project used for dimensionality reduction and Clustering for the analysis of homogeneity / heterogeneity of the landscape. In this context, the research was conducted using qualitative classification, directed the characterization and recognition of physical standards as well as the occupation of two landscape units belonging to Pitangui River Basin, which includes the cities of Castro , Carambeí and Ponta Grossa, located in east-central state of Paraná. The study involved images of Landsat 5 TM satellite, which provide overall vision, particular units and values of reflectance targets contained therein. Data regarding the spectral bands and relief were integrated and spatially related to GIS, extracting variables of topography and hydrography in order to conduct a visual reconnaissance and topographic prior units. Prior to Multivariate Analysis (Principal Component and Cluster Analysis) images, spectral and representatives of relief were targeted by Object-Oriented Analysis which generated regions described by their spectral properties, space and texture through new image as containing minimum element regions and even a relational database containing all descriptors spectral, spatial and textural. After Object-Oriented Analysis, it was performed the Principal Component Analysis which listed the descriptors coming from the relational database of Object-Oriented Analysis. The res outcomes were highest correlations decreasing dimensionality, 39 to 17 descriptors or variables, and concurrently applying the technique of Cluster Analysis, in order to find characteristics of homogeneity / heterogeneity present in the units. In cluster analysis of the first and second landscape units were always generated three distinct subgroups of initial variables, both with the Landsat image (5R4G3B) as with the Landsat image (5R4G3B) associated with the relief (DTM). It was observed that the combination of spectral data with altimetry (DTM) data possible groupings emphasizing the physical characteristics and use and occupancy of the units, note where the distribution of the variables consistently, aggregating the regions of higher reflectance spectral band red and middle infrared. The near infrared reflected more occupations anthropogenic aim in urban area in southwest in the second unit. However the middle-infrared reflected the land use units from agriculture and exposed soil the both units. The near infrared band reflected the vegetation present in the units, especially in this case the first unit of study. The insertion of relief variable (DMT) increased groups that caused greaters merging in the regions, highlighting portions in both units, such as flat areas at high altitudes and valleys where the rivers run Pitangui and Jotuba.