A major factor contributing to the utility of LANDSAT data is the repetitive coverage of each spot on the earth's surface within the sensor's field of view at least once every 18 days. This repeat coverage makes possible: studies of crop growth: monitoring urban development and land use: and predicting water runoff from melting ice and snow, just to name a few examples. Of central importance to investigators performing these studies is the requirement that imagery and/or image data on computer compatible tapes (CCT's) register to a fraction of a pixel. This paper describes the application of all digital processing methods to the problem of precision registration of LANDSAT MSS scenes.

Registration accuracy has been evaluated by means of change detection imagery and a precision correlation technique. Change detection imagery (generated by differencing pixel by pixel the registered scene data) for full scenes and sub-scenes show the need for high order interpolation (Cubic Convolution Process), in contrast to lower order interpolation. Statistical analyses using error histograms derived from the change detection imagery provide a more quantitative comparison.

Full scene registration accuracy was evaluated by designing a uniformly distributed set of features in one scene of a registered pair, and performing the correlation between the features and those centered at identically the same location in the other scene. Various statistics have been compiled from the results. The method also makes possible an analysis of the spatial distribution of errors, results of which are also reported.

Date of this Version