1 Parameter analysis that optimizes lossy compression of a multispectral and multitemporal Landsat image series L. Pesquer, A. Zabala, and X. Pons. Abstract. — The present work belongs to the group of methodologies that facilitate the access to large spatial datasets in environments with limited storage and the efficient management of such information. In fact, the paper inspects which of the various possible strategies are considered optimal for a lossy compression of remote sensing time series images, specifically Landsat-5 TM images. A wide range of tests has been carried out to analyze the impact on the compression fidelity depending on spectral or temporal aggregation. Considering this main segregation, other auxiliary parameters have been considered in order to improve outcomes such as: thermal versus optical bands, regions with and without nodata values and clean cloud zones with respect to a significant cloud cover. Studies show that compression in multispectral segregated groups improves compression fidelity in most compression ratios (eg. Multispectral compression obtains 3 dB PSNR additional units in a 10:1 compression in nodata analysis); but compression of multitemporal series presents the most irregular results. However, the study concluded that 2D-compression applied to image spectral and temporal series could be useful for obtaining homogenous quality, and sometimes better global efficiency, than compressing separately every image composing the series. Index Terms—Lossy compression, remote sensing images, multiespectral and multitemporal analysis. I. INTRODUCTION EMOTE sensing, has provided an enormous amount of data to scientific community and continues to grow at an important rate. Image sensors have increased their capabilities, in terms of spatial, spectral, radiometric, and temporal resolution [1] [2]. With this progression becomes necessary to develop strategies for manage and share massive data [3]. Moreover, all this information is required from any location and from any device, sometimes with low capacities as PDAs. In this scenario, compression methodologies, and in particularly lossy compression, offer a suitable solution [4], but before compressing data is necessary to analyze what is the optimal procedure to compress and which factors are important to consider for obtaining a good balance between quality and ratio compression. This work was supported by the Ministerio de Ciencia e Innovación of Spain, project G160106/1315. L. Pesquer is with the Center for Ecological Research and Forestry Applications, Autonomous University of Barcelona, Edifici C 08193 Cerdanyola del Vallès Spain; (phone: +34-93-5811312; fax: +34 93 5814151; e-mail: l.pesquer@ creaf.uab.cat). A. Zabala is with the Department of Geography, Autonomous University of Barcelona, Spain (e-mail: alaitz.zabala@uab.cat). X. Pons is with the Department of Geography, Autonomous University of Barcelona and with the Center for Ecological Research and Forestry Applications, Spain (e-mail: xavier.pons@uab.cat). The present work is a first step to a more complete study to answer these questions. In this first step, the authors have analyzed 2D-compression, and in near future they will analyze and will compare it to 3D-compression. The difference between both compressions consists in the dimensions implied in the compression procedure, it means that, in remote sensing images, 2D-compression searches correlation patterns and data redundancy in 2D spatial domain, and 3D-compression add spectral or temporal dimension to this analysis, searching patterns in a more complete domain. A priori, 3D- compression could be more suitable for a remote sensing series, but this study would not be completed without this first step. 2D-compression applied to image series (spectral and time series) could be useful for obtaining homogenous quality, and sometimes better global efficiency, than applying separately to every image that composes a series. In this context, this study analyzes which the influence is, in quality compressed images, of parameters like: temporal o spectral segregation, presence of nodata values, cloud proportion cover, etc, at different compression ratios. This work is based on lossy compression using JPEG2000 standard [5] [6] developed by the Joint Photographic Experts Group (JPEG). II. METHODOLOGY In this study, we have worked with multispectral and multitemporal series of Landsat-5 TM. These images dated from 04-05-2008 to 27-10-2008 and cover three study areas of 197-031 path-row. These three regions have a similar extension, around 25000 ha, but different landscapes (figure 1) and are located in Catalonia, a region of approximately 32000 km 2 situated in the northeast of the Iberian Peninsula, at the extreme southwest of Europe. All images have been acquired in L1G processing level and georeferenced and radiometrically corrected with MiraMon GIS software [7] R