Land cover mapping is an important tool to acquire useful information for making decision by different managers, such as committees of watershed, urban administration, state and federal environmental institution, among others. From land cover maps it is possible to identify areas of preservation whose management has not been correctly done. This information supplies aids to develop management plans of watersheds and monitoring of land cover. The basin of drainage is widely used as geographical unit because it is a natural system well defined spatially by a topographically drained area by a main river and its tributaries. Due to its large territorial extension, land cover mapping is produced by using remote sensing images, whose synoptic view of the environment allows identifying different types of land cover in a large area. Remote imagery on board spacecraft has been done since 1970's. Since then, new modern sensors have been designed in order to register images of the surface with better quality. Images from Landsat satellite are used in several applications in earth observation and, besides, are available freely. In this context, the present project proposes mapping the land cover in a drainage basin from multispectral images of the Landsat satellites. The study area is the drainage basin of the Três Irmãos hydroelectric reservoir, located on low course of the Tietê River, São Paulo State. We aim to assess the evaluation of the land cover before flooding caused by dam building until this day. This evaluation allows identifying the influence of the dam building over the economic development and the evolution of the deforestation in the watershed. In addition, land cover is associated with the water quality in the reservoir. The deforestation facilities runoff, whose sediments that are transported into water bodies decrease the water column transparency affecting limnological and ecological processes. Moreover, the sediments can led to silting the river and alterations in their courses. Land cover mapping from multispectral images requires classification algorithms. Two classifiers will be tested: maximum likelihood and k-means. The assessment of the classifications will be done from a dataset extracted of the images themselves. The kappa coefficient and global accuracy will be used in this evaluation. As result, it is expected to obtain maps of the land cover for different periods in order to be possible identifying landscape changes.
News published in Agência FAPESP Newsletter about the scholarship: