Evaluation of Linear Mixture Modelling of Urban Areas in Landsat TM Satellite Imagery

In this paper, the author presents the experiences made by applying the linear mixture modelling technique to an urban area located in the north of San Diego, Ca. Landsat TM satellite imagery was chosen to perform the analysis on. The paper presents the results of linear unmixing applied to raw and first-order corrected imagery. To evaluate the accuracy of the unmixing method, the results of the sub-pixel classification were compared to a reference, which is represented by high-resolution imagery, called ADAR. images were acquired at the same day time. In both scenes, noise removal and geometric registration with terrain correction were performed. Topographic effects were also corrected. Conspicuous are the big reflectance differences of vegetation and soil in the images. This is caused by different climatic backgrounds. 1996 was a year with less precipitation and higher temperatures than usually and 1998, a year dominated by El Niño, was relatively cool and wet. The resulted reflectance variations of especially vegetation might effect the classification results. Reference values for the sub-pixel classification were obtained by an interpretation of digital airborne photographs taken by an ADAR 5500 system. The ADAR data were acquired in June 1998. The images have a ground solution distance of 1m. 3 LINEAR MIXTURE MODELLING Linear Mixture Modelling has been discussed often in articles. Therefore here will be only a very brief explanation of the technique. LMM computes the spectral constituents of a pixel by analysing its spectral signature. It assumes a single reflectance of a photon at the object within the given instantaneous field of view. In many cases, it approximates reality. Whenever there is radiation transmission through one of the materials or if there are multiple reflections within a material or between two objects a nonlinear mixing occurs. In the LMM, the DN of a pixel in each spectral band is expressed as a linear combination of the characteristic relative radiance of its component endmembers weighted by their respective areal proportions within the pixel. [Ichoku 1996] Thus, the DNi of a pixel in the i-th band is given by: