Effect of landscape structure on image rescaling

Scale is an important issue in ecological and environmental studies. Image rescaling from fine resolution to coarse resolution was studied in this research. Two kinds of rescaling methods were used in this study: nonarithmetic sequence (NAS) and arithmetic sequence (AS). Compared with reference images, the rescaled image was strongly impacted by landscape structure. For rescaling factors less than the geostatistical nugget distance in a landscape, both NAS and AS rescaling methods could effectively rescale images from fine to coarse scale, but the NAS rescaling method was better for preserving image statistical variation and spatial heterogeneity compared to the AS rescaling method. When the image rescale factor was between the geostatistical nugget and range, the NAS methods could keep image invariance of statistical information, but image spatial variance decreased with the resolution coarser in both NAS and AS methods. When image rescaling factor exceeded the range, both image statistical and spatial variances decrease with coarser rescaling image resolution. After the rescaling factor over image range, only maximum likelihood method can retain image variance and texture similar to those of the reference images. The NAS rescaling methods increased image variance and spatial heterogeneity, whereas the AS scaling methods (WA and SA) reduced image variance. Overall, the maximum likelihood rescaling method was most effective for preserving image variance and texture.