Collecting the right number of ground control points

The following is the minimum number of ground control points (GCPs) to collect, but it is recommended that you collect more than the minimum to ensure accuracy. However, collecting over 20 GCPs per image does not significantly improve the accuracy for most math models. To improve the accuracy, collect GCPs evenly throughout the image at a variety of elevations and in areas where images overlap. Also, the quality of the GCPs impacts the number needed to ensure accuracy.

Table 1. Minimum number of GCPs
Math model		Minimum GCPs	Recommended
Aerial photography		Three or four per project	Three per image for highest accuracy
Optical/Radar satellite:
	MERIS		Not recommended due to image bow-tie effect
	SPOT 1 TO 4	Four per image	Six to 10 per image
	SPOT 5, IRS, ASTER, EOC, LANDSAT, QUICKBIRD (Basic), PRISM, AVNIR-2, KOMPSAT-2, KOMPSAT-3, CBERS, DMC, CARTOSAT, ORBVIEW, WorldView-1 (1B product), WorldView-2 (1B product), GeoEye-1, SSOT	Six per image	10 to 15 per image
	RADARSAT, ERS, JERS, ASAR, IKONOS, PALSAR, FORMOSAT-2, EROS, QUICKBIRD (Ortho Ready Standard), TerraSAR-X/TanDEM-X, WorldView-1 (Ortho Ready Standard), WorldView-2 (Ortho Ready Standard)	Eight per image	10 to 15 per image
	RADARSAT-2 with Toutin's model	None	Optional. When using Toutin's Model with RADARSAT-2 data, ground control points (GCPs) are not required to achieve accuracy within spacing of one pixel. You can still collect GCPs if you are confident in the accuracy of the GCPs.
Radar-specific modeling:
Radar-specific model		GCPs optional	Improve accuracy with a minimum of eight GCPs for RADARSAT-1, RADARSAT-2, TerraSAR-X/TanDEM-X, and old JAXA PALSAR data.
Rational functions:
	If computed from GCPs	Five per image*	19 per image*
	If extracted from image file	None	Optional. Using zero-order RPC adjustment requires a minimum of one GCP/TP, using first-order RPC adjustment requires a minimum of three GCPs/TPs, and using second-order RPC adjustment requires a minimum of six GCPs/TPs.
Thin Plate Spline		Three per image	More than the minimum will average out errors introduced by inaccurate GCPs or terrain variations
Polynomial:
	First-order	Four per image	More than the minimum will average out errors introduced by inaccurate GCPs
	Second-order	Seven per image
	Third-order	11 per image
	Fourth-order	16 per image
	Fifth-order	22 per image

*Depends on the number of coefficients that you want to use, for more information, see Understanding the rational functions math model.