|
MISR Level 2 Top-of-Atmosphere/Cloud Products |
This statement applies to MISR Level 2 TC Stereo and Albedo for September 27, 2001, and beyond until such a time as further improvements to MISR software are made.
An extensive review of product quality has not yet been performed. Please read the summary words of caution if you have not done so already.
Many of the algorithms used in the product retrievals have been developed specifically for the MISR instrument, and as such, are relatively untested. We expect to improve on these algorithms as we gain experience with the data. Trade-offs with the stereo-matching algorithms have been made at times to sacrifice accuracy or coverage for speed and vice-versa.
In spite of all the warnings, the MISR Level 2 TC Stereo and Albedo software which generated these products is believed to be functioning quite well except where noted below. This statement highlights major known problems with the products, as well as functionalities which are currently not implemented.
The Restrictive and Expansive albedos are still not ready for public release and thus have been set to NoRetrieval throughout the swath. Clear-Sky Modeling for the Local Albedo has not yet been implemented, nor have the Grey-Level Difference Vectors.
The Cloud Classifiers product, which contain the Angular Signature Cloud Mask (for detection of high cloud) and the cloud classifiers, is not yet available.
REGISTRATION
Cloud motion calculations are quite sensitive to the quality of registration of
the D camera L1B2 ellipsoid-projected radiance products. Since Level 1 does not
yet utilize Reference Orbit Imagery (ROI) when performing registration
correction, the registration relies on a fairly static camera model. The camera
model changes periodically, and although the registration is typically much
better, the camera models in use only guarantee accuracy of 2 pixels or less in
the D cameras. Cloud height accuracy is nominally 562 m, corresponding to 1
pixel of accuracy in the A cameras. Under the best of conditions, the heights
often appear quantized. Further, they are occasionally made worse due to
errors in cloud motion caused by misregistration of the D cameras. A 2 pixel D
camera error translates to a 10-15 m/s error in the cloud motion vectors, which
propagates to an error of 1100 m in height. We expect the registration
reliability to improve significantly when the ROI is used, and we anticipate a
reduction in height and wind uncertainties of approximately a factor of 2. For
more details, including a link to a list of orbits with known registration
problems, see the
Registration Page.
DOMAIN ARTIFACTS
Cloud motion retrievals are made on 70.4 km domains. This may at times
result in discontinuities at domain boundaries for cloud heights.
STRIPES
Horizontal stripes may occasionally appear in the product for some parameters.
This is due to one or more missing lines of data in Level 1, and often shows up
in Level 2 parameters as "No Retrieval" flag values. For more
details, read sections on
Gaps
and
Instrument Out-of-Sync in the Level 1 Statement.
BLUNDERS
Blunder detection has not been implemented. As a result, spikes may
occasionally appear in the cloud heights.
ALGORITHM UPDATES
The cloud motion and height retrievals have changed somewhat from the Level 2
Cloud Detection and Classification ATB (JPL D-11399, Rev. D). These changes
will be reflected in the next release of the document, Rev E. Highlights
include:
The histogramming of the cloud motion for each domain now involves the identification of clusters of points which may cross the histogram bin boundaries, resulting in a matrix which identifies the clusters. The clusters chosen for the two cloud layers must be local maxima in this matrix. The histogram now includes a center bin which is centered on 0.0 in each direction. We no longer concatenate bins with the same population together but rather choose the one with the smallest height range.
External meteorological inputs such as MODIS and NSIDC are not yet used. Instead, a static monthly climatology (the TASC dataset) is used.
The Level 1 Radiometric Camera-by-camera Cloud Mask (RCCM) still has not been successfully validated over land, but is working well over ocean. Therefore, the Stereoscopically Derived Cloud Mask relies solely on stereoscopically matched data to determine the presence over land, but does include the Level 1 Radiometric data as input over ocean.
If there is no successful stereo retrievals for a given 1.1km subregion, the stereoscopic height for that region is set to NoRetrieval except under the following circumstances: if the pixel is located over ocean and the RCCM indicates ClearHighConfidence, the surface height at that point is substituted in for the final stereo height.
When calculating the Reflecting Level Reference Altitude (RLRA) at 2.2km resolution, it is set to NoRetrieval if all of the corresponding StereoHeights are also NoRetrieval.
RLRA DISCONTINUITIES
The process of registering the BRF's from the surface ellipsoid to the
Reflecting Level Reference Altitude (RLRA) is dependent on the quality of the
incoming RLRA. If the RLRA is discontinuous due to blunders in the
stereoscopic height retrieval, this will feed through to the Top-of-Column
BRF's and the Local Albedo and show up as discontinuous values in those
products. If there was no valid stereoscopic height retrieval anywhere within
a 2.2 km region (the resolution of the RLRA), the RLRA (and consequently the
reprojected BRF's, the number of un-obscured pixels and the Local Albedos) will
all be set to NoRetrieval.
MODEL DIFFERENCES
The Local Albedo calculation is first attempted by Deterministic Modeling (if
the scene is homogenous), then Stochastic Modeling and finally solid-angle
weighting. No modeling is attempted for clear-sky pixels or where the solar
zenith angle is < 25.8 degrees. The (Deterministic - Stochastic) model
difference is peaked at 0.0, with the bulk of the differences being less than
0.02 with no appreciable bias. The peak of the (Deterministic-SolidAngle)
difference distribution is also located at 0.0, but the SolidAngle albedos are
biased consistently higher than the Deterministic ones, with differences up to
0.08. There is some slight band-to-band variation present in the
model-difference distributions, but the shapes and peak locations of the
distributions are similar.
BAND DIFFERENCES
The Local Albedo of the Red, Green and NIR bands are all very similar, with
differences between them on the order of 1% or less. For low clouds, the Blue
band can be up to 5% higher than the Red due to the increased Rayleigh
scattering above the cloud-tops. This difference decreases as the cloud-top
height increases.
