Super-resolution mapping is becoming an increasing important technique in remote sensing for land cover mapping at a sub-pixel scale from coarse spatial resolution imagery. The potential of this technique could increase the value of the low cost coarse spatial resolution imagery. Among many types of land cover patches that can be represented by the super-resolution mapping, the prediction of patches smaller than an image pixel is one of the most difficult. This is because of the lack of information on the existence and spatial extend of the small land cover patches. Another difficult problem is to represent the location of small patches accurately. This thesis focuses on the potential of super-resolution mapping for accurate land cover mapping, with particular emphasis on the mapping of small patches. Popular super-resolution mapping techniques such as pixel swapping and the Hopfield neural network are used as well as a new method proposed. Using a Hopfield neural network (HNN) for super-resolution mapping, the best parameters and configuration to represent land cover patches of different sizes, shapes and mosaics are investigated. In addition, it also shown how a fusion of time series coarse spatial resolution imagery, such as daily MODIS 250 m images, can aid the determination of small land cover patch locations, thus reducing the spatial variability of the representation of such patches. Results of the improved HNN using a time series images are evaluated in a series of assessments, and demonstrated to be superior in terms of mapping accuracy than that of the standard techniques. A novel super-resolution mapping technique based on halftoning concept is presented as an alternative solution for the super-resolution mapping. This new technique is able to represent more land cover patches than the standard techniques.