978-1-61284-848-8/11/$26.00 ©2011 IEEE Fusion of ALOS and QuickBird Imagery for Mangrove Analysis A Case Study in Beilun Estuary, Vietnam Jia Hu, Minhe Ji* Key Lab of Geographic Information Science, Ministry of Education and East China Normal University Shanghai, P.R.China mhji@geo.ecnu.edu.cn Abstract—This study compared four image fusion techniques for mangrove studies in the Beilun estuary, Vietnam. The fused image was generated from an ALOS multi-spectral image scene and a Quickbird panchromatic image scene. Four quantitative indices (i.e. mean, standard deviation, entropy, and correlation coefficient) were used to compare and evaluate the results of the fusion methods. Variation of fusion effects was observed among different methods. In addition, fused images were used to identify mangrove biomass levels via spectral mixture analysis. The fusion results are not quite very well from a series of evaluations, two possible explanations of the failure results present in the conclusion, which provide the reference for the relevant research. Keyword-image fusion; mangrove biomass; spectral mixture analysis I. INTRODUCTION Wetlands are considered the most biologically diverse of all ecosystems [1]. Mangroves as a primary plant life are both an important coastal wetland resource and one of the most highly productive seashore ecosystems around the world [2]. Due to its desirable functions for abundant primary production, carbon storage, and ecological buffering between human settlements and the marine ecosystem, mangrove’s inventory and monitoring are of utmost importance [3]. Recent research even deemed that mangroves as one of the best geo-indicators in global coastal change research and excellent procedures to detect and quantify coastal modifications [4]. Because of their broad scientific and economic, and even aesthetic and ethical values [5], mangrove wetlands deserve a great deal of attention in their monitoring and management. Remote sensing technology has been broadly used for forestry studies [6], but the current practice is still short of methods to extract detailed information about mangrove species and biomass [7]. Many studies have revealed that due to the high spectral similarity among different mangrove species, merely increasing the spatial resolution of imagery might have led to a wrong direction. A widely recognized observation is that spectral resolution is the true determinant for discrimination at species level. While it is still not viable to collect hyperspectral data for any specific study area nowadays, the proliferation of different sensor types and platforms provides a great opportunity to increase spectral dimensions through image fusion. It is the aim of image fusion to integrate image data recorded at different resolutions and/or by different sensors to obtain more ground information than otherwise derived from a single image alone. This study evaluated four pixel-level image fusion methods for the purpose of improving the identification of mangrove species in Beilun Estuary, Vietnam. The fusions involved the multispectral data from AVNIR-2 of the Advanced Land Observing Satellite (ALOS) and the panchromatic data from QuickBird. Four conventional quantitative indices were used to assess and compare the fusion results, including mean, standard deviation, entropy, and correlation coefficient. Since it was difficult to judge the quality of each fusion outcome simply based on the indices, the fused images were further evaluated through the assessment of mangrove biomass levels results from spectral mixture analyses (SMA). The summary statistics provided some useful insight into the feasibility of image fusion for mangrove analysis. II. METHODOLOGY A. HSV Transformation HSV is a cylindrical-coordinate color system comprised of three color elements: hue, saturation, and value. It is usually constructed by transformation from the RGB color space [9, 10]. Since this transformation is nonlinear and reversible, it is widely used for fusing images of different spatial resolutions. The specific fusion procedure for this study is follows. An HSV transform was first performed to convert multispectral data from the RGB color space to the HSV color space. Then the transformed data in the hue and saturation bands was resampled to the resolution of the panchromatic data. In the final step, a reverse transform was performed to convert data from HSV space back to RGB, with the value-band data being replaced by the panchromatic data [11]. Limitations have been reported for use of HSV transform in image fusion. This approach is known to work well for a moderate resolution ratio (1:4) and less reliable for ratios greater than 1:20. It is also noted that fusion by HSV transform only works for multispectral imagery consisting of three bands, as it is limited by the RGB channels for transformation. Supported by the National Natural Science Foundation of China (No. 40671074).