INFERRING FLUCTUATIONS OF THE AQUIFER BY MONITORING THE AREA OF SMALL LAKES IN A BRAZILIAN SAVANNA REGION USING A TEMPORAL SEQUENCE OF 50 LANDSAT IMAGES

Water availability is subjected to a complex dynamic involving fluctuation of the aquifer level, itself subject to climatic and edaphic factors as well as land use and land cover. Human pressure can have a drastic effect on the aquifer level, the effect of which are often only noticeable after years of continuous usage. In this article we are using a temporal sequence of 50 Landsat images to study a complex of small lakes in Northern Minas Gerais. Our objective is to quantify the fluctuations of the aquifer for the 1984-2009 period by monitoring the area of these lakes bi-yearly and compare it with the hydric balance to understand its evolution. Having no measurements of the aquifer, we developed a methodology entirely based on remote sensing and meteorological data. Because these lakes are very small, the 30 m resolution of the Landsat data rendered very poor results. Based on the postulate that the water pixels behave like a smooth continuous surface, we increased the resolution to 5 m using minimum curvature interpolation. Two methods were tested for extracting the lake surface pixels: supervised classification and thresholding of the normalized difference water index. The results show that the interpolated Landsat data compared well with a high resolution Ikonos image of the same date and can improved the contouring of water bodies. For extracting the water pixel the classification approach performed better by about 15%. Statistical tests showed that the lakes have been systematically decreasing but that these changes cannot be attributed to climatic factors.