International Satellite Cloud Climatology Project (ISCCP) Cloud Analysis (C1 and C2) Langley Atmospheric Science Data Center Data Set Document

ISCCP was established as the first project of the World Climate Research Program (WMO, 1984), to collect and analyze satellite radiance measurements to infer the global distribution of cloud radiative properties and their diurnal and seasonal variations. These data and analysis products will be used to improve the understanding and modelling of the effects of clouds on climate.

This document includes information for the following data sets.

ISCCP_C1_NAT

International Satellite Cloud Climatology Project (ISCCP) Stage C1 3-Hourly Cloud Products in Native (NAT) Format (ISCCP_C1_NAT)

ISCCP_C2

International Satellite Cloud Climatology Project (ISCCP) Stage C2 Monthly Cloud Products in Hierarchical Data Format (ISCCP_C2)

ISCCP_C2_NAT

International Satellite Cloud Climatology Project (ISCCP) Stage C2 Monthly Cloud Products in Native (NAT) Format (ISCCP_C2_NAT)

1. Data Set Overview
2. Investigator(s)
3. Theory of Measurements
4. Equipment
5. Data Acquisition Methods
6. Observations
7. Data Description
8. Data Organization
9. Data Manipulations
10. Errors
11. Notes
12. Application of the Data Set
13. Future Modifications and Plans
14. Software
15. Data Access
16. Output Products and Availability
17. References
18. Glossary of Terms
19. List of Acronyms
20. Document Information

The cloud analysis products of the International Satellite Cloud Climatology Project (ISCCP), called Stage C1 and C2 data, are constructed from the original B3 radiances, the results of the three parts of a cloud algorithm, and the correlative data used in the analysis. Stage C1 data represent the global, merged results reported every 3 hours with a spatial resolution of 250 km (nominal); Stage C2 data are the monthly averages and other summary statistics of the Stage C1 quantities.

The data set names are as follows:

ISCCP_C1_NAT:	International Satellite Cloud Climatology Project (ISCCP) Stage C1 3-Hourly Cloud Products in Ntive (NAT) Format (ISCCP_C1_NAT)
ISCCP_C2:	International Satellite Cloud Climatology Project (ISCCP) Stage C2 Monthly Cloud Products in Hierarchical Data Format (ISCCP_C2)
ISCCP_C2_NAT:	International Satellite Cloud Climatology Project (ISCCP) Stage C2 Monthly Cloud Products in Native (NAT) Format (ISCCP_C2_NAT)

Dr. William B. Rossow
NASA, Goddard Institute for Space Studies
Room 322A
2880 Broadway
New York, NY 10025
USA
Telephone: 212 278-5567
FAX: 212 678-5662
E-mail: clwbr@giss.nasa.gov

Chris Brest
NASA, Goddard Institute for Space Studies
Room 322A
2880 Broadway
New York, NY 10025
USA
Telephone: 212 678-5565
FAX: 212 678-5552
E-mail: cbrest@giss.nasa.gov

GOES-5
GOES-6
GOES-7
GMS-1
GMS-2
GMS-3
GMS-4
Meteosat-2
Meteosat-3
Meteosat-4
Meteosat-5
NOAA-7
NOAA-8
NOAA-9
NOAA-10
NOAA-11

ISCCP focuses on the study of the distribution and variation of cloud radiative properties. Scientific objectives are as follows:

1. To produce a global, calibrated radiance data set for studying cloud variations;
2. To coordinate basic research on cloud retrieval techniques for satellite data, and to derive and validate a global cloud climatology;
3. To improve the understanding of the role of clouds in the Earth's radiation budget and hydrological cycle.

The NOAA spacecraft are a series of satellites in 870 km (nominal) circular, near-polar, sun-synchronous orbits with an inclination angle of approximately 99 degrees (retrograde) to the equator. They cross the equator during local morning and afternoon (and corresponding night times), with an orbital period of approximately 102 minutes. Each sequential orbit covers adjacent longitudes near the equator and overlapping longitudes near the poles. The Advanced Very High Resolution Radiometer (AVHRR) on-board this series is composed of up to five spectral channels with a nadir resolution of 1.1 km. In addition, temperature sounding and ozone observations are made by the TIROS Operational Vertical Sounder (TOVS) and are used in the ISCCP analysis of B3 data.

The Geostationary Operational Environmental Satellite (GOES) series consist of spin-stabilized spacecraft in geostationary circular orbit located over 75 degrees west longitude for GOES-East (5 & 7), and 135 degrees west longitude for GOES-West (6). The GOES-6 satellite was routinely moved to provide better coverage of seasonal weather events until its failure on 1/21/89. The GOES-7 satellite is now utilized in this manner. Data are collected in the visible and infrared bands by the Visible Infrared Spin-Scan Radiometer (VISSR). The visible channel detector consists of eight identical photo-multiplier tubes that scan the Earth in parallel, producing a visible channel resolution of 0.9 km. The IR detector produces a resolution of 8 km.

The Meteorological Satellite (METEOSAT) series; operated by the European Space Agency, is in geostationary circular orbit over the equator centered at the Greenwich meridian (0 degrees E longitude), with the exception of METEOSAT-3. METEOSAT-3 is now centered at 50 degrees west and is functioning as a replacement for GOES-7 which was moved westward in response to the failure of GOES-6. The Multispectral Imaging Radiometer (MIR) on METEOSAT-2, -3, -4, and -5 collects data over the Earth in three spectral regions, one in the visible and two in the infrared. Each satellite scans the Earth from east to west and, if the water vapor channel (6.7 um) is turned off, is capable of producing a resolution of 2.5 km.

The Geostationary Meteorological Satellites (GMS) are a series of satellites are operated by the Japan Meteorological Agency and are positioned in geostationary circular orbit over the equator centered at 140 degrees east longitude. The VISSR onboard the GMS satellite collects data with four identical detectors operating in parallel, producing a resolution of 1.25 km.

The Multispectral Imaging Radiometer (MIR) sensor on METEOSAT is a scanning radiometer which provides images in the visible and thermal IR regions of the spectrum. The instrument produces images of the full Earth disc viewed from a geostationary orbit. A reduced image format, corresponding to a limited band across the Earth's disc, may be selected by telecommand. The optical reflector system of the radiometer includes a movable Ritchey-Chretien telescope with primary and secondary mirrors. This includes a mirror located in the center of the primary mirror inclined at 45 degrees to the optical axis, four folding mirrors, and a separation mirror for diverting light to the visible sensor.

The optically-collected visible and IR signals are converted into analog electric signals by five detectors. These are divided into two subsets, two visible and three IR. The detectors are distributed across the focal plane of the radiometer and as a result of the relative displacement of the detectors in this plane, their respective fields-of-view (FOV) do not coincide but are displaced relative to each other.

The two visible detectors are positioned in the focal plane of the primary telescope. Their instantaneous FOV at the Earth's surface (2.5 square km) is determined by their physical size (250 X 250 micrometers sensitive area) and the telescope's focal length (3650 millimeters). While the visible detectors function properly at ambient temperatures, the three IR detectors must be cooled to less than 95 degrees K.

Each IR detector is 70 square micrometers and generates an instantaneous 5 km square FOV at the subsatellite point. One visible channel is time sharing with the water vapor channel so that the resolution of the visible image changes depending on the choice of channels.

The following table lists the measuring geometry characteristics for the satellites employed by the ISCCP program:

SATELLITE	SCAN SYSTEM	SCAN DIRECTION	IMAGE VIEWING ANGLE
NOAA	Cross-track scan mirror	Moving south to north, scanning west to east	55.4 degrees
GOES	Spacecraft spin motion plus scan mirror	Stepping north to south, scan west to east	20 x 20 degrees
METEOSAT	Spacecraft spin motion plus scan mirror	Stepping south to north, scan east west	18 x 18 degrees
GMS	Spacecraft spin motion plus scan mirror	Stepping north to south, scan west to east	18 x 18 degrees

Global coverage is provided by the set of satellites employed in ISCCP. The coverage of each sensor is listed below in degrees of longitude and latitude:

SATELLITE	SENSOR	LONGITUDE RANGE	LATITUDE RANGE
NOAA-7,8,9,19,11,12	AVHRR	Global	Global (1)
GOES-5(E)	VISSR	15 W to 135 W	60 N to 60 S
GOES-6(W)	VISSR	75 W to 165 E	60 N to 60 S (2)
GOES-7(E)	VISSR	15 W to 135 W	60 N to 60 S (3)
METEOSAT-2,3,4,5	MIR	60 W to 60 E	60 N to 60 S (4)
GMS-1	VISSR	160 W to 80 E	60 N to 30 S
GMS-2,3,4	VISSR	160 W to 80 E	60 N to 60 S
Global coverage over a period of approximately 24 hours. For better coverage of seasonal weather events, the position of GOES-6 alternated between 98W and 108W with a position of 98W during the tropical convective season, beginning August 31, 1994 and continuing until April 30, 1987 when it arrived at 135W, where it remained until its failure on January 21, 1989. For better coverage of seasonal weather events, the position of GOES-7 alternated between 98W and 108W, with a position of 98W during the tropical convective season, beginning March 1, 1989 and continuing until April 20, 1992 when it arrived at 112W, where it will remain indefinitely. For better coverage of atlantic weather events, METEOSAT-3 was moved to 50W on August 1, 1991.

There are 132 quantities reported for the ISCCP-C1 product for each 250 X 250 km map cell. These quantities are arranged into two groups: the first 74 values are integers representing various counts, while the last 58 values are real numbers representing physical quantities. These quantities are listed in Item 13 at the end of this catalog. There are 72 quantities reported for the ISCCP-C2 product for each 250 X 250 km map cell, consisting of averages of the hour-monthly mean values obtained separately for each of the 00, 03, 06, 09, 12, 15, 18, and 21 Greenwich Mean Time (GMT) time periods. In addition to mean cloud properties, the frequency of occurrence and average properties of ten cloud types are reported. These quantities are listed in Item 13 at the end of this catalog.

Count values on the tape for ISCCP-C1 and ISCCP-C2 range from 0 to 255. Values are coded with counts from 1 to 253, with count 0 reserved for underflow and count 254 reserved for overflow. Overflow occurs at lower values for pressures and cloud thicknesses. The count 255 is reserved to signify NO DATA. If a count value is less than 1 OR greater than 253 is converted to a physical value, using the current version of the conversion tables described, the table returns a value of -100.0 for a count of 0, -200.0 for a count of 254, and -1000.0 for a count of 255. The physical quantities for these data are IR radiance, temperature, pressure, VIS radiance, reflectance, optical thickness, precipitable water amount, and ozone column abundance.

The units defined for the physical quantities in the C1 and C2 data are:

PARAMETER	UNITS
Infrared (IR) Radiance	Degrees Kelvin
Temperature	Degrees Kelvin
Precipitable Water	Centimeters
Pressure	Millibars
Visible (VIS) Radiance	Unitless
Ozone Column Abundance	Dobson Units
Reflectance	Unitless
Cloud Opitical Thickness	Unitless
Snow Cover	Percent
Ice Cover	Percent

NOTE: VIS radiances and actual reflectances are both coded using the same table. However, the radiances are not reflectances. Their value represents the measured intensity as a fraction of the effective solar constant of the radiometer. This quantity divided by the cosine of the solar zenith angle is equal to a reflectance. Both brightness temperatures (representing IR radiances) and physical temperatures are also coded using the same table.

The count value to physical relationship is not always linear. Since the radiometers measure radiances (generally with a linear response), the sensitivity and accuracy of the physical quantities derived from these radiances may not be the same over their whole range. For example, warm temperatures are measured more precisely by most radiometers than cold temperatures. It is misleading to present the data with the same "apparent" precision over the whole range. Hence, the temperature and optical thickness conversion tables are non-linear in a way that approximates the linear response of the AVHRR instrument: a difference of a single count represents a larger temperature difference for cold temperatures than for warm temperatures or a larger optical thickness difference for larger values than for smaller values. This means that a linear average of the count values representing temperature or optical thickness produces an energy-weighted result. If a different weighting is desired, then the conversion tables should be applied before averaging. All other conversion tables are linear.

The cloud detection step analyzes the radiance data twice: first to determine an estimate for the radiance values that represent clear conditions and second, to determine which radiance measurements deviate from these clear sky values. Cloudy conditions are defined to be those that exhibit radiance values that are sufficiently different from the clear values.

To avoid spurious diurnal variations of cloudiness caused by changes in methodology associated with the presence or absence of VIS data, the clear sky composite procedure relies primarily on IR radiance tests to obtain both the VIS and IR clear radiances. However, since the daytime results can be improved by use of the VIS channel measurements, these results are incorporated so that the IR-only results can be reconstructed.

The cloud detection part of the analysis must separate the radiance data set into two parts: the radiance values representing clear scenes and those representing cloudy scenes (in this case a "scene" is one image pixel with a field-of-view size ranging from 4 to 8 km, depending on satellite). Many studies have suggested that the radiances associated with clear scenes are generally, though not always, less variable in space and time than those associated with cloudy scenes (Rossow et al., 1985, and references therein, see Item 11.2). Thus, the cloud detection process starts by first testing all the radiances for low spatial and temporal variability to determine a best estimate of the clear scene radiance value at each location and time. The distribution of these clear radiance values is referred to as the clear sky composite.

The clear sky composite values are obtained as the result of two tests and the accumulation of three kinds of statistics over two time periods. The two tests for low spatial and temporal variability are meant to isolate many, but not all, of the clear image pixels from the images. The statistics are used to check whether these radiance values have the proper characteristics thought to represent clear conditions. This latter aspect of the algorithm is necessary because the magnitude of the radiance variability associated with different surface types and different cloud types is highly changeable. In some climate regimes the surface properties are much more variable; in some cases, the cloud properties are not highly variable. Thus, no single aspect of the cloud detection algorithm is successful everywhere. The final step of the process is an intercomparison of the results of all the preliminary steps to determine the best value of the clear radiances.

The first step tests the spatial variability of the IR radiances within small regions (about 100 km on land and 300 km over ocean). All pixels determined to be much colder (by 3.5 K over ocean and 6.5 K over land) than the warmest pixel are labeled "cloudy"; all others (including the warmest) are labeled "undecided". The second step tests the time variability of the IR radiances over three days at the same GMT. All pixels determined to be much colder (by 3.5 K over ocean and 8.0 K over land) than the values at the same location either yesterday or tomorrow are labeled "cloudy"; all pixels found to be the same temperature (to within 1.1 K over ocean and 2.5 K over land) as they were either yesterday or tomorrow are labeled "clear". (Performing comparisons at the same local time each day avoids the large diurnal variation of land surface temperature.) The remaining pixels with intermediate variability are labeled "undecided".

Three statistics are collected: the differences of the extremum radiances (maximum IR and minimum VIS) for consecutive 5 day periods, the number and average values of the pixels labeled "clear" for 5 day periods, and the number and average values of the pixels labeled "clear" by the time test and "undecided" by the space test. Large differences in the extremum radiances generally indicate remaining cloud contamination; low populations of "clear" pixels indicate persistent cloudiness. If these statistics suggest little cloud contamination, then the clear sky composite is formed from the 5-day-average radiances of "clear" pixels. If the statistics indicate contamination, then the clear sky composite is formed from the 30-day-average. If the 30-day statistic (number of "clear" pixels) is too small, then the 30-day extremum radiance is used.

Each pixel is labeled by the combined results of the two tests and the statistics. The two tests produce four categories, depending on whether the tests agree or disagree. The combinations (space-time of cloudy- cloudy, cloudy-undecided, and undecided-cloudy are labeled "cloudy"; the combination of undecided-clear is labeled "clear". The combination undecided-undecided is labeled "undecided", while the combination cloudy-clear is labeled "mixed". The statistics indicate whether the short- term or long-term values were employed in the clear sky composite, (i.e., whether the clear radiance value is more or less accurate). All of these labels are used to evaluate the success of the analysis by checking their consistency.

The version of the algorithm used to produce this C1 data does not use any correlative data to construct the clear sky composite, except a classification data set that indicates whether a particular location is land, water or coast. The revised algorithm will use an augmented classification that includes topography and snow/ice cover.

The final decision on the status of an image pixel is made by the bispectral threshold test (IR-only threshold at night). Using the clear sky radiances derived for each location and time, all image pixels with radiance values that are sufficiently different from clear sky conditions are declared to be cloudy. To improve the detection of cirrus and low-level clouds, single channel detections are allowed. Thus, if the VIS radiance or the IR radiance is different from the clear sky values, the pixel is called a cloud. The magnitude of the difference required is set by the estimate of the uncertainty in the clear radiance values. In these results, the VIS thresholds are 3.5 to 6.0 (land) and the IR thresholds are 3.0 deg K (ocean) and 8.0 deg K (land). (VIS radiances are represented as a percent of the instrument response obtained when measuring the full solar flux; IR radiances are represented as brightness temperatures in deg K.) Thus, if a pixel has a VIS (IR) radiance greater than (less than) the clear value by more than the threshold amount, it is considered to be cloudy.

This labeling of pixels as cloudy or clear is performed without regard to the previous labels derived from the clear sky composite analysis; however, the success of the cloud detection is indicated by whether these labels generally agree or disagree. The clear sky composite analysis is meant to be more conservative than the threshold test because the low variability of the surface properties means that not all measurements need to be included to get a good estimate of the clear radiance values. Thus, the image pixels are generally equally divided among the "cloudy", "clear", and "undecided" categories obtained from the clear sky composite analysis; the "mixed" category is usually very small except when the algorithm is having difficulty distinguishing the clouds. The threshold decision is made for all pixels; most pixels labeled as "cloudy" or "clear" in the clear sky composite analysis are similarly labeled by the threshold decision, confirming the success of the method. The "undecided" category contains radiances that are too ambiguous to make a reliable decision in the composite analysis. These pixels are usually divided into cloudy and clear in rough proportion to the number of pixels already labeled as cloudy or clear. This two stage decision process provides an "error" check from the internal consistency of the result.

The actual relationship between the pixel radiances and the clear radiances is recorded for each pixel by a two-part code: each part records a value from 0 to 5 representing the VIS and IR results separately. This preserves the ability to reconstruct the IR-only results in the final C1 data.

• VIS code - 0 no decision (no data)
• VIS code - 1 less than clear VIS by more than threshold
• VIS code - 2 less than clear VIS by less than threshold
• VIS code - 3 greater than clear VIS by less than threshold
• VIS code - 4 greater than clear VIS by more than threshold
• VIS code - 5 greater than clear VIS by more than 2 X threshold

• IR code - 0 no decision (no data)
• IR code - 1 greater than clear IR by more than threshold
• IR code - 2 greater than clear IR by less than threshold
• IR code - 3 less than clear IR by less than threshold
• IR code - 4 less than clear IR by more than threshold
• IR code - 5 less than clear IR by more than 2 X threshold
  
  

In the C1 data, various additional combinations of the two codes (VIS-IR) are used as follows to diagnose the performance of the algorithm. Classes 2, 3, 4, and 5 are also counted in class 1, while classes 7 and 8 are also counted in class 6.

• 0 = bad (no data)
• 1 = cloudy
• 2 = VIS-only-cloudy
• 3 = IR-only-cloudy
• 4 = uncertain-cloudy
• 5 = bad-cloudy
• 6 = clear
• 7 = uncertain-clear
• 8 = bad-clear
  
  

Uncertain-cloudy (class 4) pixels are pixels with radiance values that are between the threshold and two times the threshold amount. Because these radiances are often caused by broken clouds or occur when the scene is ambiguous, a count of the pixels lying near the threshold indicates how much the result would change if a different threshold value were used. This count provides a dynamic sensitivity monitor or uncertainty estimate. VIS-only-cloudy (class 2) and IR-only-cloudy (class 3) pixels are those that were determined to be cloudy by a single channel decision; that is, VIS-only-cloudy pixels have a VIS radiance that is greater than the clear sky VIS value by more than the threshold amount but have an IR radiance that is similar to the clear IR value. (At night there are no VIS-only- cloudy pixels and all cloudy pixels are IR-only-cloudy pixels.) Uncertain- clear (class 7) pixels are those which have a radiance value similar to the clear value in one channel but deviate from clear conditions too much in the other channel. For example, the VIS (IR) radiance may be similar to the clear value, but the IR (VIS) radiance is much warmer (darker) than the clear value. Bad-cloudy (class 5) pixels are similar to uncertain-clear pixels except that one channel has a radiance value similar to cloudy radiances. Bad-clear (class 8) pixels have VIS and IR radiances that are much too dark and warm compared to the clear radiance values. Bad (class 0) pixels are daytime pixels with no VIS decision or any pixels with no IR decision (usually due to lost data).

A summary of the pixel-by-pixel threshold information is provided for each region (nominal size is 250 X 250 km) in the C1 data set. Cloud amount is defined in the ISCCP data as the number of cloudy pixels within the region. Although this approach is thought to overestimate cloud coverage when using "low" resolution satellite data, no technique is yet available that determines fractional cover of individual pixels for all cases. Hence, the meaning of the cloud amount reported in C1 data is defined by this procedure: the cloud amount obtained from this analysis is an "effective value," which indicates the variation of actual cloud amount on a scale >5-10 km. In C1 the number of cloudy pixels is reported together with the total number of pixels; cloud fraction is the ratio of these numbers.

The uncertainty estimate is provided by the number of pixels classified as uncertain-cloudy by the threshold and by the number of pixels labeled as undecided in the composite step; however, several other counts are summarized which indicate in different ways that the algorithm is or is not performing as expected. In particular, when the scene is ambiguous or the clear sky composite value is contaminated, then the number of mixed, uncertain-clear, bad-clear, and bad-cloudy pixels grows. Generally, the number of bad-clear pixels should be small. Further investigation of the behavior of these "error" counts will improve their interpretation.

1. Retrieval of surface temperatures from the clear pixel IR radiances (if present) and the clear sky composite IR radiances. The surface is assumed to be a blackbody. The effects of atmospheric water vapor absorption are calculated.
2. Retrieval of surface reflectances from the clear pixel VIS radiances (if present) and the clear sky composite VIS radiances. The surface is assumed to be an isotropic reflector. The effects of Rayleigh scattering and ozone absorption are calculated. No retrieval is performed at night.
3. Retrieval of cloud top temperature and pressure from the cloudy pixel IR radiance (if present) utilizing the surface temperature from the clear sky composite and assuming clouds to be an opaque single layer.
4. Retrieval of cloud optical thickness from the cloudy pixel VIS radiances (if present) utilizing the surface reflectance and assuming the cloud is a single, conservative Mie scattering layer. Over oceans a model of surface reflectance is employed in place of the retrieved surface reflectance. No retrieval is performed at night.
5. If the optical thickness of the cloud is small, the cloud top temperature is recalculated to account for transmission of radiation from the surface using the retrieved optical thickness. The revised cloud top pressure is then used to recalculate the optical thickness. No adjustment is performed at night.

A number of consistency checks are made to determine if the radiative analysis is performing as expected. These checks generally detect problems with the data or errors in the cloud decision. Among the most interesting results so far is the check on the altitude adjustment. This procedure encounters difficulties with the optically thinner clouds because the VIS radiance measurement accuracy and the uncertainties in the calculation of the clear radiances from the retrieved surface reflectance prevent a meaningful measurement of the optical thickness. In other words, even though the cloud may be "obvious" in the IR image, its VIS radiance effect may be negligible. The presence of this condition is tested by solving the radiative transfer equation of the minimum cloud optical thickness consistent with the coldest possible cloud top temperature (cloud top at the tropopause), the surface temperature and the observed brightness temperature. If the retrieved optical thickness is less than this value (often the retrieved value is zero because there is no measurable difference between the observed radiance and the clear radiance), then the cloud top temperature is set to the coldest possible value and the optical thickness is set to its minimum value. The success of this technique appears good in some preliminary tests, but a refinement of this approach is possible.

Interesting results also include the check on the surface temperature retrieval when low cloud contamination of the clear sky radiances is present. This refers to the effect of undetected clouds on the surface temperature retrieval because the retrieval overestimates the atmospheric emission over the higher, but mislabeled, cloud tops. This causes the retrieved temperature to be colder than the observed brightness temperature, the opposite of the expected relationship. Further tests are underway to develop useful error flags.

The average and variance of each cloud and surface parameter are provided for the 250 km (nominal) region. The average value is reported directly, along with the total number of pixels used to calculate it and the root mean square value. The variance is calculated from the number of pixels, the average value, and the root mean square value. Both the average and the root mean square values are decoded in the same way.

The reported cloud parameters represent averages over all cloudy pixels in each region at that time. These average values do not indicate the structure of the clouds present, however. Original ISCCP plans called for reporting the properties of five cloud types: low, middle, high, cirrus, and deep convective clouds. The latter two types were qualitatively defined to be optically thin and thick high clouds, respectively. Consideration of how best to define these types precisely, as well as studies of the adequacy of this classification scheme to represent the actual cloud structures, was part of a pilot study on the uses of ISCCP data. Recommendations from that study were to make the definitions more flexible and to preserve greater resolution in cloud top location and cloud optical thickness. Thus, cloud type information is presented in the C1 data by counting the number of cloudy pixels with optical thicknesses and top pressures in each of 25 categories. (Only the five cloud top pressure categories exist at night. These results are reported in the first optical thickness class within each pressure category.)

Several features of this method of classification are:

1. Definitions of low clouds, for example, can be formulated by combining or interpolating among the five cloud top pressure categories. In particular the height categories can be adjusted to the levels in a climate model.
2. Different definitions that depend on the optical thickness, such as cirrus (i.e., "thin", "high" clouds) can be adopted. Users can construct their own cloud classes.
3. The mean values of optical thickness and top pressure are assumed to be equal to the central value within each class.
4. The cloud top temperature can be obtained by interpolation of the reported atmospheric temperature profile at the cloud top pressure.
5. To allow for diurnal studies, the number of vis-only cloudy pixels contributing to each cloud top pressure class is reported separately. For diurnal comparisons, these numbers should be subtracted from the totals reported in each pressure class.

A single C1 data file represents the merging of analysis results from all available satellites within the three hour time period; however, in one map cell the values from only one satellite are reported. Each location has an established hierarchy of satellite observations based on the variations of viewing geometry and time coverage characteristic of each satellite. For low latitudes, observations from the nearest geostationary satellite are preferred, while the polar regions (poleward of 55 degrees latitude) are covered only by the polar orbiter. If data from the primary geostationary satellites are not available, then a secondary geostationary satellite may be used if the viewing geometry is not too extreme. If no geostationary data are available at low latitudes, then polar orbiter data are used, if available. Since the time period of each data set is 3 hours long, anywhere from zero to two polar orbiter observations may be reported within this time period. The satellite that contributes the specific results is identified with a code number which is defined in the Volume ID file.

The basic objective of the ISCCP-C2 analysis is to summarize the cloud analysis results (Stage C1 data) on a monthly time scale. To preserve information about diurnal variability, the results are first averaged over the calendar month, separately for 00, 03, 06, 09, 12, 15, 18, and 21 GMT.

These eight time periods are referred to as the hour-monthly means. The number of days of observations contributing to the average values is recorded as the sixth parameter in each map grid cell. Then, the hour- monthly mean values are averaged to obtain the monthly mean values. Hour-monthly mean values which consist of less than three daily observations are excluded from the monthly mean. Before averaging over the eight hour-monthly mean time periods a number of adjustments are made.

Averaging the quantities from Stage C1 data to produce the Stage C2 data can be done in two ways, depending on the purpose. Some quantities, such as cloud optical thickness or cloud top temperature, are related to the effect of clouds on radiation in a non-linear way. hus, an average value meant to be indicative of the average radiative effect of clouds must give equal weight to these values proportional to their effect. Since these quantities were retrieved from radiation measurements, this weighting is also related to the variation of relative measurement precision over the range of the parameters. All quantities in Stage C2 data are averaged in this way, except for parameter 20, called PATH. For most parameters, this weighting procedure produces an average value that is not much different than that given by a simple linear average. This is not the case for cloud optical thickness, where a simple linear average produces a global monthly mean value that is about 60arger than that produced by an energy-weighted average. Parameter 17, TAU, gives the value which represents the average radiative effect of the clouds. Since cloud optical thickness is proportional to cloud water content, parameter 20, PATH, records the result of a simple linear average of optical thickness values. For a constant cloud particle size distribution (as assumed in the retrieval of optical thicknesses), cloud water path, WP, is given by

WP = [40/3]*[r~ * PATH]/Q    kg/m**2

where r~ is the average particle radius in cm, and Q is the normalized Mie extinction efficiency at 0.6 micrometers wavelength. For the cloud particle size distribution used, with r~ approximately equal to -.001 cm,

WP = 6.292 PATH    g/m**2

1. Cloud Detection

   1. Clear Sky Composites

      Since diurnal-uniform quality is desired, algorithm design has emphasized improvements in the accuracy of the IR tests. A number of very good techniques for improving the accuracy of the clear VIS radiance composite (cf., Rossow et al., 1985b) already exist. In addition, accurate models of the clear VIS radiances over ocean are available (Minnis and Harrison, 1984). Relatively good models of the land surface reflectance are also being constructed (see Matthews and Rossow, 1987, and references therein). Thus, to preserve the uniformity of the IR dependent results, while improving the daylight analysis results, the final clear-sky composite will be modified using the ocean model and the stable statistics of the land surface reflectances (Matthews and Rossow, 1987). All results dependent on VIS tests will be held separately to facilitate diurnal studies. A comparison of the IR-only results with the full daytime results will also permit estimates of the errors derived from the nighttime analysis. Early tests show that removing cloud contamination only changes the clear-sky IR radiance by 1-3 degrees K.

      Spatial variations of radiances can also be caused by changes in surface properties. Thus, the clear-sky composite will also be modified to avoid comparisons where snow or ice cover has changed.

      An improved version of the statistics for the clear-sky condition will also be implemented. The primary purpose of the revision is to avoid tests that depend on single data values. For example, the image from GOES-East for 19 July at 12 GMT was affected by a spurious gain change in the IR channel. This event caused 50 degree K colder than normal brightness temperatures, and caused the algorithm to classify the image as cloudy. Although the B3 data are undergoing quality inspection that should remove this "large" error, small errors of this type will be detected using statistical intercomparisons of the clear radiances over small regions. Thus, the alternative method will allow for identification and elimination of spurious data values.

   2. Threshold Decision

      The only planned refinements in the threshold step are the possible addition of two extra values for snow and ice surfaces and the implementation of thresholds that are linear in radiance. Many algorithms apply thresholds in terms of physical quantities. However, not all the radiometers made measurements that are linear in these quantities. To properly account for the radiometer performance and the changing sensitivity to clouds under different circumstances, the thresholds will be made linear in measured radiances. This actually introduces no change for the VIS channel; however, the IR radiance data are currently handled as brightness temperatures rather than linear counts. This change is most important in the polar regions where the radiometers are less sensitive to clouds because of the very low temperatures. This approach is more consistent with the interpretation of the threshold magnitude as an estimate of the clear sky radiance uncertainty. In addition, some refinement of the uncertainty and error counts discussed above may occur after study of the distribution and behavior of these quantities.

2. Radiative Transfer Model Analysis

   The planned refinements for the radiative transfer model analysis are intended to improve the reporting of the error checks thereby providing better documentation of algorithm performance. In particular, those "error" conditions that indicate cloud contamination of the clear radiances or improper labeling of the pixels will be added to the summary statistics to provide internal error estimates. There have been some minor changes to the radiative analysis to prevent unnecessary data losses. The most noticeable problem in these results is that small uncertainties in the retrieval process can cause very low clear radiances to produce surface reflectances that are slightly less than zero. The current code discards these data causing a loss of data near the terminator in the images. This result suggests that the daytime analysis should be ended at a somewhat larger solar zenith angle (the cutoff is currently 81 degrees). However, these small negative values are also valid, but "inaccurate" measurements of a small value; hence, processing can proceed by setting small negative values to zero. Another problem is that the B3 data contain no geostationary VIS images for the three time periods near local midnight, even when small portions of the image actually have solar zenith angles larger than the cut-off value. The current analysis code interprets this situation as a loss of data; hence, portions of these images were unnecessarily discarded. Changes in the software logic will avoid these problems.

3. Statistical Analysis

   Based on further accuracy tests and studies of the information content of the statistics, there may be some refinements or changes in definitions. Suggested changes include adding more pressure categories in the cloud classification, adding an alternative VIS cloud parameter equivalent to the surface reflectance, and improving the height adjustment information to indicate how many pixels actually changed height categories.

The method used to reconstruct the variances utilizes the mean and root mean square (rms) values of each quantity; however, this approach does not retain enough precision to report small variances accurately. In this version of C1 data, the variances are only accurate to the nearest 100f the mean value. Since most variances are smaller than this amount, most variances are reported to be zero.

The mean cloud properties reported in the C1 product are the final values from the radiative analysis. This means that the daytime values of cloud top temperature (TC) and cloud top pressure (PC) have been altered by the effects of the VIS channel measurements. Since the same adjustment is not performed at night, direct comparison of the day and night values of cloud top temperature and pressure must be interpreted with caution. However, the vertical distribution of clouds can be reconstructed from cloud classes and the mean IR radiance values. Visible only (VIS-ONLY) numbers can be subtracted from the total number of pixels at each pressure level, while the IR radiances can be used to estimate the cloud top temperature and pressure without TAU corrections.

Producing the ISCCP-C2 product involved performing a number of adjustments on the ISCCP-C1 data before determining the monthly averages. The adjustments necessary included VIS adjustments during daytime, VIS adjustments during nighttime, calibration adjustments, standard adjustments, special METEOSAT adjustments, and diurnal adjustments

To produce Stage C1 data, results from several satellites are merged into a single global data set. In regions where more than one satellite provides results, the merger process selects the preferred satellite according to a specified hierarchy that favors data continuity and observations made closer to nadir view. Frequency histograms of the differences in the overlapping measurements between all pairs of satellites are collected and the modal value estimated from the average of the mode value and the three nearest values above and below the mode value. These estimated differences for each satellite, when compared to the reference polar orbiter, are applied to adjust for small residual radiance calibration differences. The corrected quantities in the hour-monthly mean are: cloud top and surface temperature, cloud optical thickness and water path, and surface reflectance. Magnitudes of these corrections are illustrated in the following table. Actual calibration adjustments for each month are reported in the record prefixes for each parameter for each satellite.

The magnitude of the calibration adjustments applied to Stage C2 data to remove small residual calibration differences are shown here as the standard deviation and range of all corrections applied to each satellite over the period July 1983 - February 1987.

PARAMETER	STD DEV	RANGE
Cloud Top Temperature	0.74 K	+/- 2.5 K
Surface Temperature	1.10 K	+/1 3.0 K
Cloud Optical Thickness and Water Path	0.02	+/- 0.08
Surface Visible Reflectance	2%	+/- 8%

The spectral response of the METEOSAT "visible" channel is wider than that of the other radiometers used in the ISCCP analysis; normalization of METEOSAT radiances is done using spectrally uniform targets (clouds and clear ocean areas). The spectral response difference means that surface reflectances calculated for vegetated land areas from METEOSAT are larger than for the other satellites. This difference in surface reflectance is removed in the hour-monthly mean data set by using regression relations that are obtained by comparing METEOSAT and NOAA measurements as a function of vegetation type and season. A single relationship that varies with season was found to represent differences as a function of vegetation. Adjustment factors are applied for each season and are given in the following table. Unadjusted values can be recovered from Stage C2 data by multiplying by the slopes (given in the following table) and adding the intercept values.

Adjustment factors applied to METEOSAT land surface reflectances to reduce them to values measured at an approximate wavelength of 0.6 +- 0.1 micrometers are shown in the following table. Seasons are the standard three-month periods in the northern hemisphere.

Adjustment: Adjusted Value = (Original Value - Intercept)/Slope

Season	Slope	Intercept
Winter	0.893	0.1154
Spring	0.786	0.1135
Summer	0.752	0.1290
Fall	0.820	0.1362

Science, User and Data Services Office
Langley Atmospheric Science Data Center
NASA Langley Research Center
Mail Stop 157D
Hampton, Virginia 23681-2199
USA
Telephone: (757) 864-8656
FAX: (757) 864-8807
E-mail: larc@eos.nasa.gov

Document Revision Date:

July 17, 1996; November 24, 1997; August 1998; February 2000

Document Review Date:

...

Document ID:

...(currently leave this blank)

Citation:

...(currently leave this blank)

Document Curator:

Science, User and Data Services Office
NASA Langley Atmospheric Science Data Center
Telephone: (757) 864-8656
FAX: (757) 864-8807
E-mail: larc@eos.nasa.gov

Document URL:

http://eosweb.larc.nasa.gov/GUIDE/dataset_documents/base_isccp_c1_c2_dataset.html