Abstract: Reggiani SpA, via Tonale 133, 21100 Varese, Italy; - (dario.simonetti, silvia.carboni)@ext.jrc.it Abstract – At the JRC, a methodology is developed to monitor the tropical forest cover in Latin America, Southeast Asia and Africa. The results will provide quantitative measurements of changes for the year 1990 and 2000 and will be a major input for the FAO FRA 2010 remote sensing survey (Global Forest Resources Assessment). The project is based on object-based classification of a systematic sampling of Landsat imagery at each longitude and latitude intersect. The area covered at each sample site is a box of 20km x 20km for which Landsat data are used for both dates. Prior to the classification and the change detection, a robust approach applicable to a very large amount of data had to be developed to put the multi-temporal and multi-scene data on the same radiometric scale. The paper presents all the pre-processing steps applied to a total of circa 4,000 image pairs. Starting with the conversion to TOA (Top of Atmosphere) reflectance, the image normalization is described as well as the haze correction process. A two-level segmentation is applied on the multi-date imagery. The results of these processing steps prior to the supervised classification are presented. Keywords: Land cover change, pre-processing, Landsat, tropical forest 1. INTRODUCTION Tropical regions are currently undergoing rapid changes in land cover (Achard et al ., 2002). These changes, in particular forest clearing can have different impacts such as greenhouse gas contribution and biodiversity loss. A global and systematic estimation of the status and monitoring of forest cover changes is important to accurately access these potential impacts and to inform better decision-makers at national and international scale (Mayaux et al. , 2005; Lepers et al. 2005). Remote sensing imagery offers repetitive data acquisition, a synoptic view of inaccessible areas and consistent image quality. Deforestation and vegetation changes have been thus widely studied with remotely sensing technologies. The JRC is developing a methodology to monitor the tropical forest cover in Latin America, Southeast Asia and Africa, based on the FAO global Forest Resources Assessment 2010 (FRA2010) systematic sample, in support to FRA2010 remote sensing survey. The project is based on an automatic object-based classification of Landsat imagery validated by visual interpretation of regional experts (Achard et al. , 2009). To produce accurate change detection results, radiometric and atmospheric correction of multi-date images is of crucial importance. This paper presents all the pre-processing steps applied to a total of 4,000 image pairs covering Latin America, Southeast Asia and Africa. Starting with the conversion to Top of Atmosphere (TOA) reflectance, the image normalization is described in details as well as the haze correction process. To facilitate the interpretation and improve the classification, a two-level segmentation is applied on the multi-date imagery providing objects of similar pixels both spectrally and temporally. 2. SAMPLING STRATEGY Figure 1 depicts the study areas covered by the JRC. The grid system selected for the global systematic sampling is a rectilinear grid based on degrees of geographical latitude and longitude enabling a straightforward implementation, easy location and understanding (FAO, 2007). To monitor changes at a scale relevant to land management, boxes of 10 km by 10 km plus 5 km buffer to facilitate interpretations have been extracted at each intersection point. On our study areas, the one by one degree sampling frame results in 2055 intersection points in Sub-Saharan Africa, 1230 in Central and South America and the Caribbean and 741 in South and Southeast Asia. This sampling scheme results in a circa 3 % sampling rate (Achard et al. , 2009).