Relative Radiometric Gain Correction over the Dynamic Range of All Reflective Channels of the Landsat 5 Thematic Mapper

Detector response in sensors such as the Landsat 5 Thematic Mapper (TM) is typically modeled as a linear function of the incident radiance reaching the focal plane. Due to differences in gain and bias among the detectors, the resulting imagery can contain significant striping artifacts that must be removed. Given knowledge of the detector bias, such ‘destriping’ of the imagery can be achieved through the application of linear corrections assumed to be valid across the entire dynamic range of the instrument; the slopes in the correction models are given by the reciprocals of the estimated relative detector gains. This paper presents an analysis intended to evaluate the performance of corrections based on previously developed lifetime relative gain models for the Landsat 5 TM at the extreme ends of the dynamic range. A striping metric (SM) was used to validate the relative gain correction in subregions of image data with pixel values between 180 and 220 DN; at these signal levels, relative gain correction tended to be quite effective, with striping metrics typically greater than 80% (100% indicates absence of striping). In subregions with pixel values between 1 and 30 DN, the estimated striping metric tended to be much smaller; for very low signal levels, the metrics became negative, indicating enhancement of striping. Subsequent analysis of these cases indicated variations in the noise margin on the order of 2 DN, the same order of magnitude as the signal level. The results of these analyses indicate that reliable correction cannot be guaranteed at low signal levels dominated by noise. At higher signal levels, however, where the effects of noise are proportionally smaller, reasonably accurate corrections based on the lifetime models can be achieved, allowing preservation of the overall radiometric fidelity of the image data.