Reducing Striping and Non-uniformities in VIIRS Day/Night Band (DNB) Imagery

The Suomi National Polar-orbiting Partnership (SNPP) VIIRS Day-Night Band (DNB) offers quantitative measurements of visible and near-infrared light over a dynamic range from full daylight to the dimmest nighttime scenes. This range presents a challenge to radiometric calibration, but the instrument has exceeded all of its absolute radiometric requirements. Nevertheless, striping and banding are still discernible, day or night, but especially in low-light scenes. The causes may be a combination of detector cross talk, stray light or hysteresis in the data used for calibration. These issues combine to reduce the scientific utility of these unique observations for gaining new insight on the nocturnal environment. However, with techniques described here, striping and other nonuniformities can be reduced to the extent that shot noise is the only limiting factor in nighttime DNB imagery. The success of these techniques depend on a detailed understanding of the inner workings of the DNB instrument and its calibration process. This paper will demonstrate destriping algorithms that effectively remove striping from DNB imagery without reducing the overall radiometric accuracy. The focus is on improved nighttime and twilight imagery, but removal of striping in daytime imagery is also demonstrated. We evaluate removal of fixed-pattern non-uniformity in dark scenes on a per-orbit basis using both the calibration sector signals as well as filtered twilight and nighttime Earth-view scenes. DNB calibration is complicated by its utilization of 32 aggregation modes from nadir to edge of scan. For gain uniformity we discuss some problems with the current calibration methods, and demonstrate a way to minimize the correlated error between detectors and aggregation zones using the moment matching technique for moonlit scenes. A similar technique can be used for daytime and twilight scenes. The benefits and problems with an alternative cross-calibration technique between gain stages using indirect illumination of solar diffuser view is also discussed. Histogram equalization is shown to be effective for minimizing striping and banding. In all cases, data with stray light is filtered out to prevent contamination of the destriping process. These techniques do, however, improve uniformity even in areas where stray light contamination exists. Since the launch of SNPP, the low-light detection capabilities of VIIRS DNB have opened up several new and exciting areas of research including the detection of nightglow emission and features illuminated by nightglow. These new areas of investigation have elevated the usefulness of DNB data beyond the originally envisioned design limits. Because of this, striping and other nonuniformities have emerged as limiting factors to these new areas of research. Reducing these artifacts would extend the usefulness and maximize the potential of the highly successful DNB. Examples will be presented demonstrating the benefits of these destriping and uniformity techniques for the following DNB applications: observation of city lights through thin clouds and haze; the study of nightglow distribution and gravity wave measurement using nightglow; observation of clouds illuminated only by nightglow; study of the auroras at high latitudes; reflective imagery in the twilight with the Near Constant Contrast product; observation of fires and gas flares at night. All of these applications would benefit greatly by reducing non-uniformities in the DNB sensor data records.