One can assess instrument performance by comparing the results of two instruments measuring the same thing.  A key tool is the correlation scatter plot in which results from one instrument are plotted as the x-coordinate and from the other as the y-coordinate of a point on the x-y plane.  Many such pairs are plotted and the relative instrument performance is evident from the scatter of points off the ideal 45° (x=y) line.  This figure shows a scatter plot of the “temperature of the lower stratosphere” (TLS) produced from AMSU ch-9 on the NOAA-18 satellite for Sep 2006, against near-coincident GPSRO data from COSMIC.  We see a relative bias of >1K, an RMS scatter of about 1K about a fitted line, and a change in the bias of more than 2 K over the observed temperature range -- far from adequate for long-term climate change monitoring (from S. Ho, NCAR, 2007).  The error is predominantly in the AMSU data.  How can we tell?