The impact of misregistration on change detection - Geoscience and Remote Sensing, IEEE Transactions on

AbstructThe impact of misregistration of images on the detection of changes in land cover has been evaluated using spatially degraded Landsat MSS images. Attention is focused on simulated images of the normalized difference vegetation index (NDVI) at two of the spatial resolutions of the planned Moderate Resolution Imaging Spectrometer (MODIS), namely 250 and 500 m. In the first of two sets of experiments singledate images from seven diverse areas were first misregistered against themselves and the statistical properties of the differences were analyzed using semivariograms. In general, we would expect that the finer the spatial frequencies present within an image the greater the consequences of misregistration. The results indicate that in the absence of any actual changes to the land surface, the consequences of misregistration were very marked even for subpixel misregistrations. Pairs of images from different time periods were then misregistered. The results showed, that for four of the seven areas, an error equivalent to greater than 50% of the actual differences in the NDVI as measured by the semivariance, was induced by a misregistration of only one pixel. To achieve an error of only lo%, registration accuracies of 0.2 pixels or less are required. All these four areas were densely covered by vegetation, whereas for the other three more sparsely vegetated areas with semiarid climates, a registration accuracy of between 0.5 and 1.0 pixel were sufficient to achieve an error of 10% or less. These results indicate that high levels of registration must be achieved by operational monitoring systems if there is to be reliable monitoring of global change.