Simulation Study On The Impacts Of Sensors’ PSF On Mixed Pixel Decomposition

Mixed pixels decrease the precise of recording spectral information for a spatial resolution and improvement of resolution could not decrease the number of mixed pixels. For improving identification of features and extraction of information from a remotely sensed image precisely, different kinds of methods are proposed under the assumption that signals in a mixed pixel are the sum of different kinds of objects within the projection of the pixel weigthed by response, so areal proportions within the pixel could be acquired if end-members were known. Compared to the assumptions made in unmixing, adjacency effect and nonuniformity effect caused by sensor’s spatial response introduce significant uncertainty into the results of unmixing.The impacts of adjacency effect and nonuniformity effect on unmixing are related to spatial configuration of objects on the surface of the Earth. Diversity of spatial configurations is not considered in the researches over the two effect mentioned above though uncertainty in unmixing caused by them is one of the hotpoints in remote sensing, and impacts of spatial configurations on unmixing are not understood by people. Characteristic of spatial response of sensors are different to each other and the value of digital number(DN) for the same scene changes with the variation of characteristic, and as a result, estimated areal proportions are also different. To solve the above problems, Mente-Carlo Simulation is used in the paper to find the relationships between both biases and random errors and characteristics of spatial responses and scenes, the purpose is achieved by acquiring distributions of frequencies of DNs for different spatial responses and scenes when the areal proportions within the scenes are known. Conclusions acquired in this research are displayed as following:1. Impacts on DNsThe value of DN is a random variable as the impacts of nonuniform response of PSF, the width of range of DNs increases with the proportion being close to 50% and the frequency distribution is also more normal. Adjacency effect also introduces uncertainty into the DN, and the max uncertainty is determined by the contribution of adjacency effect and the frequency distribution is similar to that introduced by nonuniformity effect. The uncertainty of DNs introduced by PSF’s nonuniformity increases with the asymmetry of PSF increasing, but parameter of PSF decreases it. The asymmetry of PSF has an effect on the distribution of DNs, meantime, the mean and distribution of DNs inctroduced by adjacency effect change with the increase of parameters. With increasing the sizes of the smallest patches, the range of DNs introduced by PSF’s nonuniformtiy also increases and probability of each DN introduced by adjacency decreases.2. Impacts on systematic errorsThe systematic error introduced by nonuniformity effect increases with the symmetry and parameters decreasing, and the parameters also increase systematic errrors introduced by adjacency effect. Besides, the parameters change the way of the general systematic error changing. The value and trend of systematic errors introduced by nonuniformity effect change with the sizes of patches increasing and he sizes of patches have on effect on systematic errors introduced by adjacency effect, and linearity of systematic errors decreases with the sizes of patches increasing.3. Impacts on random errorsRandom errors introduced by nonuniformity effect increase with the symmetry and parameters increasing, and the parameters also increase the general random errors and the characteristic of piecewise is more significant. The sizes of the smallest patches increase the random errors introduced by nonuniformity effect and the symmetry of the way of changing decreases.