CALIPSO Quality Statements
Lidar Level 2
Vertical Feature Mask Product
Version Releases: 3.01, 3.02
|
CALIPSO Quality Statements |
|
This document provides a high-level quality assessment of the Level 2 lidar vertical feature mask product, as described in section 2.7 of the CALIPSO Data Products Catalog (Version 3.2) (PDF). As such, it represents the minimum information needed by scientists and researchers for appropriate and successful use of these data products. We strongly suggest that all authors, researchers, and reviewers of research papers review this document for the latest status before publishing any scientific papers using these data products.
The purpose of these data quality summaries is to inform users of the accuracy of CALIOP data products as determined by the CALIPSO Science Team and Lidar Science Working Group (LSWG). This document is intended to briefly summarize key validation results; provide cautions in those areas where users might easily misinterpret the data; supply links to further information about the data products and the algorithms used to generate them; and offer information about planned algorithm revisions and data improvements.
The CALIPSO Level 2 Lidar Vertical Feature Mask (VFM) consists of a sequence of bit-mapped integers, with one 16-bit integer being recorded for each range resolution element in the Level 0 lidar data downlinked from the satellite. Decoding the bits in the individual integers yields information on feature type (e.g., cloud, aerosol, or clear air) and subtype (e.g., water cloud or ice cloud) at each location. Information about cloud thermodynamic phase, and the amount of horizontal averaging required for detection is also included, as are quality assessments for all classification decisions. No new parameters have been added for the version 3 VFM product. However, the interpretation assigned to the bits describing cloud ice-water phase has changed slightly from version 2. These changes are described in detail below.
Because validation for different parameters can require different levels of effort, and because the uncertainties inherent in some retrievals can be substantially larger than in others, the maturity levels of the parameters reported in the different data products files are not uniform. Therefore, within this document, maturity levels are provided separately for each scientific data set (SDS) included with the data files. The data product maturity levels for the CALIPSO layer products are defined in the table below.
| Beta: | Early release products for users to gain familiarity with data formats and parameters. Users are strongly cautioned against the indiscriminate use of these data products as the basis for research findings, journal publications, and/or presentations. |
| Provisional: | Limited comparisons with independent sources have been made and obvious artifacts fixed. |
| Validated Stage 1: | Uncertainties are estimated from independent measurements at selected locations and times. |
| Validated Stage 2: | Uncertainties are estimated from more widely distributed independent measurements. |
| Validated Stage 3: | Uncertainties are estimated from independent measurements representing global conditions. |
| External: | Data are not CALIPSO measurements, but instead are either obtained from external sources (e.g., the Global Modeling and Assimilation Office (GMAO)) or fixed constants in the CALIPSO retrieval algorithm (e.g., the 532 nm calibration altitude). |
The CALIPSO Level 2 lidar vertical feature mask data product describes the vertical and horizontal distribution of cloud and aerosol layers observed by the CALIPSO lidar. Each range bin in the Lidar Level 0 data is characterized by a single 16-bit integer, with the various bits in the integer representing flags that describe some aspect of the data measured within the bin. Instructions on how to decode these integer data are given in the sections below. Also provided are quality summaries for each classification reported. The data are recorded in nominal increments of 15 consecutive laser pulses, which is nominally equivalent to a distance of 5-km along the laser ground-track.
| Bits | Field Description | Bit Interpretation |
|---|---|---|
| 1-3 | Feature Type |
0 = invalid (bad or missing data) 1 = "clear air" 2 = cloud 3 = aerosol 4 = stratospheric feature 5 = surface 6 = subsurface 7 = no signal (totally attenuated) |
| 4-5 | Feature Type QA |
0 = none 1 = low 2 = medium 3 = high |
| 6-7 | Ice/Water Phase |
0 = unknown / not determined 1 = randomly oriented ice 2 = water 3 = horizontally oriented ice |
| 8-9 | Ice/Water Phase QA |
0 = none 1 = low 2 = medium 3 = high |
| 10-12 | Feature Sub-type | |
| If feature type = aerosol, bits 10-12 will specify the aerosol type |
0 = not determined 1 = clean marine 2 = dust 3 = polluted continental 4 = clean continental 5 = polluted dust 6 = smoke 7 = other |
|
| If feature type = cloud, bits 10-12 will specify the cloud type. |
0 = low overcast, transparent 1 = low overcast, opaque 2 = transition stratocumulus 3 = low, broken cumulus 4 = altocumulus (transparent) 5 = altostratus (opaque) 6 = cirrus (transparent) 7 = deep convective (opaque) |
|
| If feature type = Polar Stratospheric Cloud, bits 10-12 will specify PSC classification. |
0 = not determined 1 = non-depolarizing PSC 2 = depolarizing PSC 3 = non-depolarizing aerosol 4 = depolarizing aerosol 5 = spare 6 = spare 7 = other |
|
| 13 | Cloud / Aerosol /PSC Type QA | 0 = not confident 1 = confident |
| 14-16 | Horizontal averaging required for detection (provides a course measure of feature backscatter intensity) |
0 = not applicable 1 = 1/3 km 2 = 1 km 3 = 5 km 4 = 20 km 5 = 80 km |
User notes for the feature classification flags.
The probability distribution functions (PDFs) of χ′ vs. δv vs. <β′532> for clouds and aerosols that are used by the V3 CAD algorithm were developed based on a four month test data set. These PDFs are binned by altitude (1 km increments between 0 km and 20 km) and latitude (between 90°S and 90°N in 10° increments). Despite the use of these updated 5D PDFs, which significantly enhance overall performance, the V3 CAD algorithm (PDF) may continue to have some difficulty correctly classifying moderately dense dust and smoke layers presented in high latitudes and/or high altitudes as aerosol. Users should also be aware that clouds embedded within optically dense aerosols may be identified by the feature finder algorithm as a single layer. While this happens less in version 3 than in the earlier releases, these misidentified heterogeneous features will likely be classified as clouds.
In the polar regions where polar stratospheric clouds (PSCs) are observed, there may be times when stratospheric layers are misclassified as cloud. This typically happens when the base of a PSC drops below the GMAO-reported tropopause or when a PSC is vertically adjacent to a cloud system in the troposphere.
| high confidence | = | |CAD score| ≥ 70 |
| medium confidence | = | 50 ≤ |CAD score| < 70 |
| low confidence | = | 20 ≤ |CAD score| < 50 |
| no confidence | = | |CAD score| < 20 |
| high confidence | = | layer base ≥ (Ht + 2.5 km) |
| medium confidence | = | (Ht + 1.0) km ≤ layer base < (Ht + 2.5 km) |
| low confidence | = | layer base < (Ht + 1.0 km) |
| Value | Version 2 Interpretation | Version 2 Interpretation |
|---|---|---|
| 0 | no confidence | no/low confidence |
| 1 | low confidence | phase based on temperature only |
| 2 | medium confidence | medium confidence |
| 3 | high confidence | high confidence |
Bits 10-12, Feature Sub-type
In summary, the algorithm classifies aerosol layers that have volume depolarization ratio (δv) greater than 0.2 as desert dust and 0.075 < δv < 0.2 as polluted dust. Note that polluted dust could be a component of urban pollution, i.e., it is not confined to desert regions but is any type of aerosol composed of some dust-like particles. Of the non-depolarizing aerosols, layers lofted above 1 km are assumed to be smoke, and layers less than 1 km above the surface are either clean continental if the layer IAB is small or polluted continental if the layer IAB is large.
An assessment of the CALIOP aerosol subtyping scheme can be found in Mielonen et al., 2010.
Bit 13, Cloud / Aerosol / PSC Subtype Quality Assessment
 |
|
The Feature_Classification_Flag values are stored as a sequence of 5515
element arrays (i.e., as an N x 5515 matrix, where N is the number of
separate records in the file). Each array represents a 5 km
"chunk" of data, and each array element contains the feature
classification information for a single range resolution element in the
Level 0 lidar data downlinked from the satellite. As shown in the table
below, the vertical and horizontal resolution of the CALIPSO data varies
as a function of altitude above mean sea level (see
Hunt et al., 2009). The image above provides a
pictorial representation of the mapping of the one-dimensional array of
Feature_Classification_Flag values into a two-dimensional array of
range-resolved lidar data samples. The numbers in each block of the image
indicate the 1-D array indices associated with each spatial averaging
regime in the 2-D lidar backscatter data. Only the starting and ending
indices are shown.
Example code for transforming a 1-D array of feature classification flags into a rectangular, altitude-registered matrix is available for both IDL and MatLab.
|
| Altitude Region | Vertical Resolution (meters) |
Horizontal Resolution (meters) |
Profiles per 5 km |
Samples per Profile |
|
|---|---|---|---|---|---|
| Base (km) | Top (km) | ||||
| -0.5 | 8.2 | 30 | 333 | 15 | 290 |
| 8.2 | 20.2 | 60 | 1000 | 5 | 200 |
| 20.2 | 30.1 | 180 | 1667 | 3 | 55 |
| Total | 545 | ||||
| Lidar Level 2 Vertical Feature Mask (VFM) Information Half orbit (Day) geolocated data radiances |
|||
|---|---|---|---|
| Release Date | Version | Data Date Range | Maturity Level |
| December 2011 | 3.02 | November 1, 2011 to present | Validated Stage 1 |
| May 2010 | 3.01 | June 13, 2006 to February 16, 2009 March 17, 2009 to October 31, 2011 |
ValStage1 |
The CALIPSO Team is releasing Version 3.02 which represents a transition of the Lidar, IIR, and WFC processing and browse code to a new cluster computing system. No algorithm changes were introduced and very minor changes were observed between V 3.01 and V 3.02 as a result of the compiler and computer architecture differences. Version 3.02 is being released in a forward processing mode beginning November 1, 2011.
Version 3.01 of the Lidar Level 2 data products is a significant improvement over previous versions. Major code and algorithm improvements include
The sections below highlight important changes to the layer detection, scene classification, and extinction algorithms that have implications for the overall quality of the Lidar Level 2 data products.
| Figure 6: Histograms of CAD scores for Version 2 (red) and Version 3 (blue) |
 |
This figure in below presents the relationship between the CAD score and the IAB QA ( = 1- Column IAB cumulative probability). The IAB QA is a measure of the backscatter overlying a cloud or an aerosol layer. A value of 1 corresponds to a situation where the atmosphere above the cloud or aerosol layer under consideration is clear. Values smaller than 1 indicate the presence of overlying features (aerosol and/or cloud); the smaller of the IAB QA value, the more of the overlying features. The IAB QA is highest for high magnitude CAD scores and slopes down gradually for small CAD score magnitudes. This relationship reflects the fact that the presence of overlying features tends to add difficulty to the cloud and aerosol classification and therefore reduce the confidence of a classification made. The dip between -10 and 20 corresponds to the outlier features, indicating that these outliers are mostly likely overlaid by some relatively dense features. The cloud layers with special CAD scores (103 and 104) have the smallest IAB QA values. The relatively big value at CAD = 0 corresponds to the features having zero CAD values at high altitudes where the probability of the presence of overlying features is low. At high altitudes the separation of clouds and aerosols is not as good as at low altitudes because of the presence of cirrus clouds.
| Figure 7: Relation between CAD score and Layer IAB QA Factor |
 |
Overall, because of the better separation between clouds and aerosols in the 5D space, the 5D algorithm significantly improves the reliability of the CAD scores. The improvements include:
Dense aerosol layers (primarily very dense dust and smoke over and close to the source regions), which are sometimes labeled as cloud in the V2 release, are be correctly identified as aerosol, largely because of the addition of the integrated volume depolarization ratio. In addition, in the open oceans, dense aerosols that were previously classified as clouds are now frequently observed in the marine boundary layer. Improvements are also seen for these maritime aerosols. Note, however, dense dust/smoke layers found at single-shot (0.333 km) resolution will be classified as cloud by default. This issue will be revisited for post-V3 releases.
Because the V2 CAD algorithm used a latitude-independent set of 3D PDFs, a class of optically thin clouds encountered in the polar regions that can extend from the surface to several kilometers were sometimes misclassified as aerosols. In version 3, these features are now correctly classified as cloud.
Correct classification of heterogeneous layers is always difficult. An example of a heterogeneous layer would be an aerosol layer that is vertically adjacent to a cloud or contains an embedded cloud, but which is nonetheless detected by the feature finder as a single entity in the V2 release. By convention, heterogeneous layers should be classified as clouds. The version 3 feature finding algorithm has also been improved greatly, and can now much better separate the embedded or adjacent single-shot cloud layers from the surrounding aerosol. This improvement in layer detection contributes significantly to the improvement of the CAD performance.
Some so-called features identified by the layer detection scheme are not legitimate layers, but instead are artifacts due to the noise in the signal, multiple scattering effects, or to artificial signal enhancements caused by non-ideal detector transient response or an over estimate of the attenuation due to overlying layers. These erroneous "pseudo-features" are neither cloud nor aerosol and are distributed outside of the cloud and aerosol clusters in the PDF space. The V3 CAD algorithm can better identify these outlier features by assigning a small CAD score (the bump between -10 and 20 in the V3 CAD histogram) and classify most of them as cloud by convention. A CAD threshold of 20 can effectively filter these outliers out.
The extinction products are produced by first identifying an aerosol type and then using the appropriate values of Sa and the multiple scattering factor, η(z). Note that multiple scattering corrections have not yet been implemented for the current data release, so that η(z) = 1 for all aerosol types. The accuracy of the Sa value used in the lidar inversions depends on the correct identification of the type of aerosol. In turn, the accuracy of the subsequent optical depth estimate depends on the accuracy of Sa.
The underlying paradigm of the type classification is that a variety of emission sources and atmospheric processes will act to produce air masses with a typical, identifiable aerosol 'type'. This is an idealization, but one that allows us to classify aerosols based on observations and location in a way to gain insight into the geographic distribution of aerosol types and constrain the possible values of Sa for use in aerosol extinction retrievals.
The aerosol subtype product is generated downstream of the cloud-aerosol discrimination (CAD) scheme and, therefore, depends on the cloud-aerosol classification scheme in a very fundamental way. If a cloud feature is misclassified as aerosol, the aerosol subtype algorithm will identify this 'aerosol' as one of the aerosol subtypes. The user must exercise caution where the aerosol subtype looks suspicious or unreasonable. Such situations can occur with some frequency in the southern oceans and the polar regions.The version 2 algorithm included a rudimentary ability to identify a specific subset of high confidence instances of HOI. These clouds were classified as ice clouds, and flagged with a 'special CAD score' of 102, indicating that they had been further classified as HOI. The new version 3 algorithm implements a much more sophisticated scheme for recognizing HOI that correctly identifies many more instances of these sorts of ice clouds. The special CAD score of 102 is no longer used to identify these layers. Instead, the "ice cloud" and "mixed phase cloud" classifications have been eliminated, and replaced as shown in the table below.
| Value | Version 2 Interpretation | Version 3 Interpretation |
|---|---|---|
| 0 | unknown/not determined | unknown/not determined |
| 1 | ice | randomly oriented ice (ROI) |
| 2 | water | water |
| 3 | mixed phase | horizontally oriented ice (HOI) |
The Ice/water Phase QA flags have also been redefined slightly for Version 3, as follows:
| Value | Version 2 Interpretation | Version 3 Interpretation |
|---|---|---|
| 0 | no confidence | no/low confidence |
| 1 | low confidence | phase based on temperature only |
| 2 | medium confidence | medium confidence |
| 3 | high confidence | high confidence |
A confidence flag of QA=1 indicates the phase classification is based on temperature. Initial classification tests are based on layer depolarization, layer-integrated backscatter, and layer-average color ratio. Layers classified as water with temperature less than -40 C are forced to ROI and given a confidence flag of QA=1. Layers classified as ROI or HOI with temperature greater than 0 C are forced to water and also given a confidence flag of QA=1. Clouds for which the phase is 'unknown/not determined' are assigned a confidence value of 0 (no/low confidence).
Layers classified as HOI based on anomalously high backscatter and low depolarization are assigned QA=3. These layers are rare after the CALIOP viewing angle was changed to 3° in November 2007. The Version 3 algorithm computes the spatial correlation of depolarization and integrated backscatter and uses this as an additional test of cloud phase. Layers classified as HOI using this test are assigned QA=2. The majority of the layers classified as HOI result from this test and typically have higher backscatter than ROI but similar depolarization. These layers are common even at a viewing angle of 3°. We interpret this as clouds with significant perpendicular backscatter from ROI but containing enough HOI to produce enhanced backscatter. These layers tend to be found at much colder temperatures than the high confidence HOI (see Hu et al. 2009).
PLEASE NOTE: Users of the CALIOP optical depths should read and thoroughly understand the information provided in the Profile Products Data Quality Summary. This summary contains an expanded description of the extinction retrieval process from which the layer optical depths are derived, and provides essential guidance in the appropriate use of all CALIOP extinction-related data products. Validation and improvements to the profile products QA are ongoing efforts, and additional data quality information will be included with future releases.
