Those radiometric characteristics of the Landsat 4 Thematic Mapper (TM) that can be established without absolute calibration of spectral data have been examined. Subscenes of radiometric all raw data (B-data) were examined on an individual detector basis: areas of uniform radiance were used to characterize subtle radiometric differences and noise problems. A variety of anomalies have been discovered with magnitude of a few digital levels or less: the only problem not addressable by ground processing is irregular width of the digital levels. Essentially all of this non-ideal performance is incorporated in the fully processed (P-type) images, but disguised by the geometric resampling procedure. The overall performance of the Thematic Mapper is a great improvement over previous Landsat scanners.

The effective resolution in radiance is degraded by about a factor of two by irregular width of the digital levels. Several detectors have a change of gain with a period of several scans, the largest effect is about 4%. These detectors appear to switch between two response levels during scan direction reversal; there is no apparent periodicity to these changes. This can cause small apparent difference between forward and reverse scans for portions of an image. The high-frequency noise level of each detector was characterized by the standard deviation of the first derivative in the sample direction across a flat field. Coherent sinusoidal noise patterns were determined using one-dimensional Fourier transforms. A "stitching" pattern in Band 1 has a period of 13.8 samples with a peak-to-peak amplitude ranging from 1 to 5 DN. Noise with a period of 3.24 samples is pronounced for most detectors in band 1, to a lesser extent in bands 2, 3, and 4, and below background noise levels in bands 5, 6, and 7.

The geometric fidelity of the GSFC film writer used for Thematic Mapper (TM) images was assessed by measurement with accuracy bette than three micrometers of a test grid. A set of 55 control points with known UTM coordinates was measured on a digital display of part of band 5 of the TM image of the Washington, D.C. area and fitted to the control points. The standard error of the fit of the TM image to the control is 37 meters, or 1.3 pixels, with no consistent distortion. These test indicate that the geometric fidelity of TM images is likely to be higher than the ability of film recorders to reproduce the images.