Integration of topographic correction in a pixel-based compositing algorithm in the Romanian Carpathians Steven Vanonckelen, Stefaan Lhermitte, Anton van Rompaey Departement of Earth and Environmental Sciences Katholieke Universiteit Leuven Heverlee, Belgium steven.vanonckelen@ees.kuleuven.be Patrick Griffiths Geomatics Lab, Department of Geography Humboldt University of Berlin Berlin, Germany Abstract—Within geosciences, there is a growing interest in regional scale studies that encompass large areas. Remote sensing is a privileged tool for data acquisition to support such studies. Data-integration of information from various images remains a challenge. A major breakthrough is the development of image compositing techniques that aim at composing homogeneous images by sampling the most suitable pixel from various scenes. At present, it is not clear to which extent topographic correction is able to improve pixel-based composites. In this study, a pixel- based Minnaert correction was implemented on a pixel composite in the Romanian Carpathians to correct for topographic distortions. Visual results show that the shadowing effects are reduced in the corrected composites. Furthermore, the topographic correction resulted in more homogenous reflectance values across the study area. Keywords— atmospheric correction; topographic correction; pixel-based compositing; Landsat; Carpathians; remote sensing, large scale mapping I. INTRODUCTION At the end of 2008, all archived Landsat data became freely available. The opening provided opportunities to reconstruct land cover and land use change (LULCC) [1, 2]. Despite the free availability of Landsat data, monitoring of land use changes over large areas still poses methodological challenges [3]. Especially in mountain areas, land cover mapping based on remote sensing instruments is affected by topographic and atmospheric effects [4]. However, comparisons of the impact of multiple correction methods on classification accuracy are still lacking [5]. Furthermore, Landsat footprints have a small spatial extent of 30x30 meter and phenologically consistent datasets are required when analyzing LULCC [6]. When the 16-day Landsat repeat cycle is considered, it is possible that only a few unclouded scenes are provided per growing season [7,8]. Data availability is further limited by discontinuities due to sensor or data related errors (f.e. the failure of scan line correction in Landsat 7) [9]. Pixel-based compositing of Landsat imagery offers advantages such as cloud free and homogeneous data [10]. Therefore, image compositing techniques improve large area LULCC mapping. Recent developments have encouraged image compositing for Landsat data: free data access [2], optimized preprocessing methods [11, 12] and improvements in computational resources [13, 14]. Existing approaches do not implement surface reflectances or are based on the evaluation of a single, often spectral criterium [15, 16]. Griffiths et al. (2013) implemented a state-of-the-art pixel-based compositing algorithm that implements the advantages of Landsat preprocessing [8]. However, the influence of the implementation of a topographic correction method in the pixel-based compositing has not been researched. Therefore, in this paper, a pixel-based topographic correction is integrated in large scale mapping and applied on a mountainous region. II. METHODOLOGY A. Study Area The study area is located in the Romanian Carpathian Mountains with an elevation range up to 2 544 meter and a temperate-continental climate. Figure 1 indicates that the study area is covered by 9 Landsat footprints. The majority of the land cover in the Romanian Carpathians comprises deciduous and coniferous forests. The growing season occurs between April and October, and varies in response to annual rainfall and elevation [17].