Influence of the nature and number of ground control points to the quality of remote sensing geometric corrections

Georeferencing satellite images is an essential procedure to carry out most remote sensing applications. The quality of this process will affect all the ulterior procedures and products. Independent test ground control points (GCPs) are required to assess the quality of the correction. However, a representative number is hardly obtained when they are manually located. This work studies the effect of the number of GCPs in the geometric correction quality when they are manually located. The methodology has been applied to Landsat TM images in a region with complex relief (heights ranging from 0 to 3000+ m). The work presents a spatial representation of the error and discusses its role in the visualisation of the quality. Moreover, we critically discuss the usage of indicators as the RMS error without considering the number of GCPs or the method used in their placement in the realistic assessment of the geometric quality of the imagery. Indeed, it is shown that, for the studied scenes, a minimum of 25 GCPs is needed to achieve a test RMS smaller than a pixel and that not using independent GCPs leads to unrealistic quality indicators. Moreover, manual placement of GCPs gives clearly worst results than automatic procedures.