IRAS Explanatory Supplement
V. Data Reduction
G. Extended Source Products
Table of Contents | Index | Previous Section | Next Section
- Processing Overview
- Quality Checking, Selection, and Weights
- Phasing, Sorting, and Gaps
- Conversion to Surface Brightness
- Compression and the Time-Ordered Files
- Projection into Sky Maps
- Consistency Checking and Removal of Bad Data
- Final Map Image Generation
The basic approach to the compilation of the two-dimensional images and related time-ordered files was to select high-quality data meeting criteria based on observing conditions and the performance of individual detectors. Since extensive automatic confirmation tests were not applied as in the case of the discrete sources, internal consistency between different detectors within each survey observation was imposed by adjusting individual baselines and responsivities to produce the same mean brightness as the band average for that observation (the "destriping" operation). Weighted averages of the data were used to mosaic data from multiple survey scans of the same region into single digital images. Consistency between the scans forming an hours-confirming coverage was checked by human scanning of difference images; significantly discrepant data were deleted from the images. Finally, since the zodiacal emission toward any given direction on the celestial sphere depended upon the Earth's orbital position and Sun-referenced observation angles at the time of the observation, data obtained in the three hours-confirming coverages of the sky were processed and presented separately. A file needed to reconstruct the observing geometry for each observation of a given region of the sky, the Zodiacal Observation History File, was produced.
Calibrated time-ordered detector data were used for the extended emission files if they had been obtained under acceptable observing conditions with respect to nuclear radiation event rate, adequate off-axis distance from strong potential sources of stray light, and satellite scanning rate. Data were deleted near the beginning and end of scans when the scan rate deviated from the nominal survey value, or if the detector signal saturated the analog-to-digital converer.
The radiation blanking time introduced by triggering of the spike "deglitcher" in two detectors of each band was averaged over 2.5 sec intervals. Data were deleted from any band during any interval in which the dead time introduced by the deglitcher exceeded 10% in that band.
The processing was designed to allow definition of a cone of avoidance for closest approach to the Earth, Moon, Mars, Jupiter, and Saturn. Stray light performance of the instrument was examined early in the mission, and it was concluded that the nominal scan constraints on closest approach to the Earth and Jupiter produced data of acceptable quality for extended source processing, and that no constraint on data near Saturn or Mars would be imposed. No data within 30° of the Moon were included in the extended source products. This resulted in a series of lune shaped gaps spread across the ecliptic plane in each of the three sky coverages (Fig. III.C.6).
A weight factor was assigned to each sample from each detector as a quality indicator for that sample. The weights followed the samples through all subsequent averaging, image creation, and image mosaic processes, yielding a weight map corresponding to each intensity map produced. A general weight was assigned to data from each detector based on the nominal sensitivity and noise of that detector observed early in the mission. Inoperative detectors were assigned zero weight. The weight was proportional to the variance of the data sample, so that it could be used as a statistical weight in an averaging process.
The extended source images were not intended to give high-quality photometric information for point or small extended sources. However, to maintain consistency for small sources, data from the two edge detectors in each band which were substantially smaller than nominal detectors in that band (detectors 26, 47, 39, 42, 1 1, 31, 4, and 55) were deleted.
In preparation for smoothing and binning the data for the various extended source products, the time-ordered calibrated data were adjusted on a detector-by-detector basis to align samples obtained at the same sky position of the telescope boresight. This phasing took account of both the scan rate of the telescope and of electronic phase shifts in the sampling of individual detectors. Detector data were also sorted into cross-scan order in each band according to their centroid positions. Because of the phasing operation, some data were lost at the beginning of each scan and whenever a gap in the telemetry stream occurred. The star-up time for the phasing was 11 seconds, or about ¾ ° on the sky.
Detector data were converted from calibrated in-band fluxes (W m-2) to surface brightness, (W m-2sr-1), using an effective instantaneous field of view for each detector obtained from slow scans of point sources across the array (Table IV.A.1).
The detector data were compressed in the in-scan direction prior to projection into the image domain. Compression was accomplished by filtering and then decimating in order to prevent aliasing. A symmetric filter was desired to provide zero phase shifts, and a Lanczos single smoothed filter ((sin x)/x)2 weighting in the window) was selected. The filter width was chosen for each band to provide two output samples per second corresponding to a nominal 2' sample spacing. No compression or interpolation was applied in the cross-scan direction because the construction of the telescope's focal plane array produced samples with 2' spacing. These compressed data formed the time-ordered input for the sky plates and Galactic plane maps.
A lower resolution time-ordered file of intensity data was also created. The Zodiacal Observation History File (ZOHF) contains the data samples obtained by averaging data from all detectors in each band (according to their weights) over an interval of 8 seconds of time. This provides data samples at approximately 30' intervals in both in-scan and cross-scan directions. Thee file also retains, in addition to celestial coordinates, the Sun-referenced observation angles and time, and information necessary to evaluate the zodiacal emission contribution to the intensity in each sample (see discussion below). Data from this file were used to create the low-resolution all-sky maps.
In the course of reducing the scan data into images, it was found that, in spite of the efforts to calibrate the gain and offset for each detector individually, small inconsistencies remained between detectors in a given band within a single scan. Such systematic inconsistencies showed up clearly as "stripes", i.e., persistent streaks in the in-scan direction in the images on the cross-scan scale of a detector length. A "destriping" procedure was developed to remove these high spatial-frequency inconsistencies while preserving the best available photometric information for the band as a whole. The procedure consisted of determining a responsivity and baseline correction for the data from each detector for each survey scan, and then applying that correction to the time-ordered data file prior to projection into images. The corrections were determined by requiring that the intensity from each detector follow the weighted band-average intensity during that observation. This procedure depended on the scan passing over areas of both low and high average background so that responsivity effects could be separated from baseline effects. Prominent small sources (high curvature regions) in the data were excluded in computing the averages. If insufficient data were available in a given observation, or if the correlation of the output of a detector with the band average was poor, the corrections determined in the previous observation were used.
This process reduced the striping on a scale of a few arc minutes introduced by variations among the detectors in the focal plane. It did nothing to remove the ½ ° wide stripes caused by variations in calibration or sky brightness between different scans of the same piece of sky. The treatment of these wider stripes is discussed below under the topic of consistency checking in Section G.8.
After phasing, smoothing, and resampling to produce the compressed data the time-ordered detector data formed the equlavent of a two-dimensional array 15 or 16 samples wide (depending on the band) and several thousand samples long. Exact positional information was carried for every other sample of the two detectors at the edges of the array in each band the "tie points". The map projection was accomplished by projecting the exact positions of the tie points into the tie and sample space of the map images with the mapping transformation, and doing a bi-linear interpolation in line and sample space to get the projections of the other samples. All projections were done so that the sky coordinate associated with a pixeI refers to the position at the center of the pixel. The map projections used are discussed in Section X.D. Once assigned a position in image space, a particular data sample was binned by dividing the weight of the sample among the nearest four bins with an inverse bi-linear interpolation, multiplying the sample value by the divided weights and accumulating the weighted samples into the four bins of the intensity image and the weights into the four bins of the weight image. Maintaining these separate weighted intensity and weight images facilitated combining data from multiple scans into a single image. The final average intensity image was produced by dividing the weighted intensity image, pixel-by-pixel, by the weigh image. These procedures applied to the processing of the low-resolution all-sky map data as well as the sky plate data.
After the data had been assembled into maps, the images were examined to check for anomalies. One coverage of the sky consisted of ½ ° wide survey scans spaced every ¼ °. By selecting alternate scans, a coverage could be split into two sets of non-overlapping scans, each of which essentially covered the whole sky. The check consisted of examining the difference image between these two halves of a sky coverage. Anomalies discovered in this way were then classified and if necessary, examined further in the two direct intensity images. If the anomaly was severe, the offending data were removed from the map. Three classes of anomalies were removed:
- Obvious flooding of the focal plane by near-field objects.
- Scans or sections of scans which were substantially noisier than their neighbors. The random noise level needed to be about two times larger than normal before this effect was noticeable, so relatively few data were removed for this reason. More frequently, the destriping strategy would fail and noise in the form of objectionable 2'-wide stripes would appear. Scans causing such stripes were rejected.
- The brightness of the scan differed substantially from the generally smooth background level of its neighbors. The criterion used for removal was a difference of more than 10% of the baseline value from neighboring scans on adjacent orbits. A substantial amount of data was removed for this reason. Particularly offensive in this respect were 25 µm scans made near edges of the South Atlantic Anomaly (SAA) where bias-boost (Section III.C.4) procedures were not used.
Data removal was accomplished by adding the offending scan or section of a scan back into the map with the negative of its original weight.
Several known residual anomalies remain from this cleaning process. No attempt was made to remove or even find small scale discrepancies between different scans of the same piece of sky, and many small, generally point-hike, sources remain in the maps which DO NOT appear in the same place in two different scans. This point cannot be overemphasized; when using the sky maps to look at small sources, even bright ones, at least two separate coverages MUST BE COMPARED or THE USER WILL BE FOOLED. Many, maybe most, of the non-confirming sources are due to asteroids; however, they do occur in all parts of the sky, and caution must be exercised in all cases.
Noticeable ½ °-wide stripes run across many of the sky plates. Regularly spaced patterns of these stripes were due to the effects of non-boosted SAA crossings, as mentioned above. Isolated stripes were due to some undetermined calibration anomaly. Almost none of the stripes exceed the 10% difference criterion; however, an occasional stripe which differs by more that this has been left in to prevent creation of a hole in the map.
It is not known at this time whether the scans deleted due to non-boosted SAA crossings, or their neighboring scans, were correct. The maps have been made flatter by the deletions of data, but the possibillty of small, systematic DC calibration errors in the middle of scans, deemed unlikely, cannot be excluded.
No attempt has been made to exclude the effects of photon-induced responsivity enhancement in the data (see Section IV.A.8). For instance, passage across bright regions in the Galactic plane increased the responsivities of the 60 and 100 µm detectors for some time after passage. Similar effects near bright discrete sources are evident as well. Again, comparison among the three sky coverages should be made to avoid unrecognized problems due to these effects. All three coverages should be compared for this purpose, since the first two coverages generally passed over a given piece of sky in nearly the same direction, while the third coverage, where present, generally passed in a different direction.
The final steps in production of sky plate maps were the reduction to average intensity, conversion from in-band average intensity to brightness density (Jy sr-1) and application of final calibration data. Reduction to average in-band intensity was accomplished by division of the weighted intensity image by the corresponding weight image. The conversion to brightness density was based on the knowledge of the shapes of the filter passbands of the instrument and the assumption of a radiation source with an energy distribution flat in flux per octave. Small corrections modified the baseline and rescanned the resulting brightness on a pixel-by-pixel basis to reflect the final calibration.
Maps of the sky within 10° of the Galactic plane were made by remapping the pixels from the sky plates into Galactic coordinates using the cylindrical projection described in Section X.D. The remaping was done by the tie point technique described above, where the full map transformation was applied only to a subset of the pixels and linear interpolation is applied to the rest. Binning of a data sample into the nearest four pixels was nlso done as described in Section V.G.7.
The low-resolution all-sky maps and the Zodiacal Observation History File were subject to the same modifications as the sky plates for final generation processing.
Table of Contents | Index | Previous Section | Next Section