# ArcGIS 10.5 - Testing the accuracy of a satellite image's georeferencing using dGPS ground control points

I want to test the accuracy of the georeferencing of a satellite image by using a set of independent ground control points collected in the field via differential GPS.

Is there a tool/method/procedure in ArcGIS 10.5 to quantify/assess the deviation of a location in the satellite image from that of a dGPS control point.

Basically, what you need is to compute the RMS of the positionning error. This type of analysis requires a manual input of points that you should place on your satellite image where you SEE that it should be according to the location on the ground where you took the measurement. So the procedure in ArcGIS would be :

1) create a point feature class

2) In an edit session for your feature class, draw one point at each location on the satellite image where you have the homologous DGPS measure (and give it an ID that matches your DGPS survey point)

4a) (if necessary) project your your points in the coordinate system of your DGPS point

4b) create a X and a Y field, and populate them with the X and Y coordinates (calculate geometry after right clicking each field)

5) join the DGPS values to the control point values (this can be done based on the ID (join by attribute) or based on the position DGPS point if you added them to the map (join by location) )

6) compute the square error for each point, that is the square of their distance (field calculator : ((!X_satellite! - !X_DGPS!)**2 + (!Y_satellite! - !Y_DGPS!)**2)

7) compute the average of the square errors (right click on field, then "statistics")

8) take the square root of the result, this is the RMS (root mean square error) that is used as a standard for evaluating the quality of the geolocation of an image.

Remark: If you used the georeferencing tool from ArcGIS to update the georeferencing of your image, you also get a value of RMSE, but this is the RMSE of the model based on the points that are used to update the georeferencing. Using points that are independent of the model is more rigourous.