Proceedings Article | 12 August 2014
KEYWORDS: Landsat, Agriculture, Soil science, Vegetation, Clouds, Roads, Remote sensing, Image classification, Buildings, Earth observing sensors
Landsat archives (made accessible by USGS at no charge since 2011) have made available to the scientific community a large amount of satellite multispectral images, providing new opportunities for environmental information, such as the analysis of land use/cover changes, which represent important tools for planning and sustainable land management. Processing a time series images, the creation of land cover maps has been improved in order to analyze phenomena such as the soil sealing. The main topic of this work is in fact the detection of roads and buildings construction or everything that involve soil removing. This subject is highly relevant, given the impact of the phenomenon on land use planning, environmental sustainability, agricultural policies and urban runoff. The analysis, still in progress, has been applied to Bologna Province (Emilia-Romagna Region, Italy) that covers 3703 Km2. This area is strongly urbanized: 8,9% of the total surface is sealed against a national value of 6,7%, with the soil sealing rate which has been defined from recent studies as the fourth Italian value in the 2001/2011 period. Other information available for this territory derive from CORINE Land Cover and Copernicus Projects. In the first one, the minimum mapping unit is 25 ha and the one for change is 5 ha; these values are too large for an accurate detection of the soil sealing dynamics. On the other hand, the Copernicus Project provides an imperviousness layer with a better resolution (20x20 m2), but its maps start from 2006. Therefore, the potential of multispectral remote sensing analysis over large areas and the multitemporal Landsat availability have been combined for a better knowledge about land cover changes. For this work, Landsat 5 and Landsat 8 images have been acquired between 1987 and 2013, according to basic requirements as low cloud cover and a common acquisition season (summer). A supervised pixel-based classification has been performed, with maximum likelihood algorithm. Due to landscape heterogeneity, classification has been improved with auxiliary data, such as NDVI. Therefore, the obtained maps have been compared with a post-classification change detection procedure in order to quantify land use changes, with particular attention to the soil sealing increase.
