|
MISR Level 1 Alpha Products |
See also Statement dated June 15, 2000, concerning data from June 1 to July 31, 2000.
This statement applies to MISR Level 1 Products for August 1, 2000, and beyond until such a time as further improvements to MISR software are made.
An intensive, systematic data quality assessment of these Level 1 products has not been completed; however, limited assessments have been performed and the radiometric and geometric calibrations used to support data production have been improved significantly since the previous release. It should be noted that new problems in the production software and surprises in the raw data stream are still being discovered. Due to these circumstances quantitative conclusions from scientific investigations making use of these Alpha products should be treated with caution.
In spite of the above warnings, the MISR Level 1 software which produced the product files is believed to be functioning nominally except where noted below. The list below highlights major known problems with the products as well as some product attributes which might be confusing.
No problems have been found with the current release of PGE7 software. Preliminary analysis indicates that the software is meeting all of the requirements. Further analysis is planned.
The Geometric Parameters exhibit one algorithmic quirk which has surprised some users. Solar zenith and azimuth angles near the swath edge occasionally appear to jump around. This inconsistency is the result of an intentional choice of algorithm whereby solar angles are computed at the mean time at which MISR cameras viewed the ground point in question. Adjacent points are not always visible to the same set of cameras. This can cause a bias in solar angle towards cameras which acquired that point.
This portion of the list is rather long, so the sub-headings are listed for quick reference.
The Ancillary Geographic Product (AGP) contains eleven fields of geographical data. For the purpose of this quality statement, the AGP data fields are categorized into following groups:
Group A: Geographical locations corresponding to a 1.1 km resolution grid defined in the Space Oblique Mercator (SOM) map projection 1) geographic latitude, and 2) geographic longitude.
Group B.1: Elevation data corresponding to a 1.1 km resolution grid defined in the Space Oblique Mercator (SOM) map projection: 1) point elevation, 2) average scene elevation, and 3) standard deviation of scene elevation.
Group B.2: Regional elevations data corresponding to a 17.6 km resolution grid defined in the Space Oblique Mercator (SOM) map projection: 1) regional average scene elevation, and 2) regional standard deviation of scene elevation.
Group C: Surface orientation data corresponding to a 1.1 km resolution grid defined in the Space Oblique Mercator (SOM) map projection: 1) average surface azimuth angle, 2) average surface zenith angle, and 3) standard deviation to the mean surface slope.
Group D: Land/Water identification data corresponding to a 1.1 km resolution grid defined in the Space Oblique Mercator (SOM) map projection: 1) land/water identifier 2) dark water algorithm suitability mask.
The major portion of the AGP data fields are created using the following datasets as the input: 1) Digital Terrain Elevation Dataset (DTED), 2) Digital Chart of the World (DCW) Hypsography, 3) ETOPO5, and 4) World Vector Shoreline. Mostly, the AGP generated data directly reflect the quality of the inputs without degradation. However, there is a slight reduction in quality due to the processing algorithm used for some of the data fields. The details of the algorithm underlying the creation of the AGP can be found in "Level 1 Ancillary Geographic Product ATB". The following are quality statements associated with previously defined groups of AGP data fields.
Group A quality statement: A point-by-point map projection computation was used in order to produce latitude and longitude data for the centers of the 1.1 km resolution grid and achieved a one centimeter accuracy.
Group B1 quality statement: The point elevation data are directly obtained from the input datasets and there is no degradation in quality. However, the specific definition of the average scene elevation and standard deviation of scene elevation required a processing step which included resampling between the input and output grid of the same resolution. This processing step caused some reduction of quality, especially in the regions with high relief. Quality information resulting from a global validation of these fields will be added in the next release of AGP dataset.
Group B2 quality statement: Due to the low resolution of the recording grid, there is no quality degradation for the data fields in this group.
Group C quality statement: For average surface azimuth angle and average surface zenith angle, the processing step based on the resampling caused some reduction in quality, especially in regions with high relief. Quality information resulting from a global validation of these fields will be added in the next release of AGP dataset. The standard deviation to the mean surface slope should not be considered reliable. The global validation results indicate that produced value does not match its definition as specified in the Algorithm Theoretical Basis document.
Group D quality statement: Overall, the land/water identifiers are as accurate as the input dataset (i.e. World Vector Shoreline) used to derive these data fields. It should be noted that discrepancies exist between some of the Arctic and Antarctic continental and island shorelines as delineated in the elevation fields versus those delineated in the land/water identifier fields. This is the result of the input products chosen for data preparation, and not a computational error. The WVS is used to delineate all shorelines globally because it represents the most up-to-date delineation at the best scale map available.
Radiances in the MISR Level 1 products are based on release 4 of the In-flight Ancillary Radiometric Product (ARP) database. We believe this calibration is the most accurate delivered to date. It provides the radiometric response of the MISR cameras calculated using the MISR On-Board Calibrator. In particular, the Blue HQE photodiode has been shown to be quite stable since launch, and its measured radiances has now been validated using independent means. We therefore believe the absolute calibration of MISR to be better than 4 percent at the 1.0 equivalent reflectance level (one sigma confidence level). At the lower end of the cameras dynamic range respose (0.001 in equivalent reflectance), the uncertainty increases to 10% due to photon noise limitations.
A transfer from the HQE Blue detector standard to the Green, Red, and NIR HQE detectors has in addition enabled a calibration of the MISR cameras which has a high accuracy in the band-to-band calibration. This is but one reason why we believe the camera-to-camera and band-to-band radiometric errors of MISR are lower than its absolute errors, as reported in the ARP data product.
One error is known to exist in version 4 of the ARP. It was intended to reverse the order of all uncertainties in going from Version 3 to Version 4. We desire that the new values be ordered from low to high in equivalent reflectance. This was not done, as intended, for the total-uncertainty values. It was done, however, for all other uncertainty and SNR parameters. This error will be corrected in the next ARP release, but is not considered significant. No higher level processing uses these parameters.
A final consideration for users is that the radiance reported for an individual pixel in one of the L1B2 products is obtained by resampling to the SOM map projection. Therefore an individual L1B2 pixel does not necessarily correspond directly to an observation made by a single camera-CCD pixel.
MISR Level 1 products, generated with the Camera Geometric Model (CGM) version 5, represent a significant improvement in terms of georectification and coregistration accuracy if compared with the data produced with the previous CGM version 4. CGM version 5 is based on a full in-flight geometric calibration utilizing 120 Ground Control Points (GCP's) and MISR image data obtained during a time period of 45 days. As a result, the expected mean geolocation error across all cameras is below 80 meters with the standard deviations ranging between 200 meters for the nadir view angle, up to 600 meters for the most oblique angles (i.e., the D cameras). A verification of these estimates was performed using data acquired up to 90 days from the time the calibration was completed. Only the D-aft camera has shown a slight deviation from the estimates in terms of a bias of about 450 meters. CGM version 6 accounts for that bias and provides a consistent geolocation accuracy across all nine cameras. It should be noted that occasional and temporary degradations in attitude accuracy have been observed. These attitude degradations ultimately impact product gelocation and registration. Nevertheless, we expect a very small percentage of data to be affected.
In addition, the MISR Level 1 software will now correct line/time discontinuities in the raw data. These occasional input errors, which were caused by instrument out-of-sync events, resulted in gross geolocation errors in the line direction in some L1B2 products. L1B2 line misregistrations as large as 10 kilometers were observed. Such large geolocation errors should now be a thing of the past.
The raw MISR data contains occasional gaps. These gaps usually consist of a few lost lines. Straight lines of raw data are resampled to gentle curves in the SOM map projection. Radiances in the gap regions are filled in with pre-defined fill values. Gaps then usually look like narrow, curved, bright, horizontal stripes in the L1B2 image. There is at least one small gap in almost every swath. In rare cases, data gaps of many lines have been observed.
L1B2 software processes a half-block at a time. If the software encounters fatal processing errors in a half-block region, then that region is skipped, and processing is attempted in the next region. The current version of the software does not adequately fill in the regions which are skipped. Therefore, whatever happens to be in computer memory gets written to the product file for that region. This behavior is often evident at the beginning of the swath where the geolocation process attempts to bootstrap itself, often taking a few regions to get running smoothly. In these skipped regions, radiance values of zero with occasional high spikes and regions of echoed image from previous regions may be visible.
The MISR instrument tends to go out-of-sync momentarily if the data rate from the hardware exceeds the real-time flight computer's capacity to write data out. This condition can occur whenever the instrument is put into Local Mode as well as when the Aa camera exposure correction is sent. The later occurs every orbit when the Aa camera begins global mode acquisition. The resultant error is seen most often in the forward and nadir cameras near the beginning of the swath.
Image lines acquired while the MISR instrument was out-of-sync may contain sporadic fill and/or repeats of previous lines. The resulting image contains a brief vertical smear across the swath. Normally, this phenomenon only lasts for a handful of lines. In order to avoid geolocation errors, fill values are inserted in the line time fields in these regions by L1B1.
The line-of-sight between an off-nadir camera and a ground point is sometimes blocked by a topographic feature, such as a mountain. In such cases, fill values are reported instead of radiances in the terrain product. Large patches of obscuration fill can be seen in the D cameras over mountainous regions.
Blocks which encompass no land at all get entirely filled with ocean-fill values in the terrain product. Terrain algorithms are wasteful over ocean since height variation is negligible there. The Ellipsoid product already contains radiances for these blocks. If ocean blocks are required, blocks from the Ellipsoid product may be substituted.
