|
||||||
|
|
||||||
|
||||||
|
|
||||||
Data Set Documentation
The Pathfinder Land Products data sets are created using the SSM/I Pathfinder daily HDF antenna temperature (TA) files. Each daily land product file begins with the first A scan after 00:00:00 UTC and contains all data up to 23:59:59 UTC. Each file contains land classification (CLS), land surface temperature (LST), latitudes (LAT), longitudes (LON), A-scan start times (AST), orbital elements (ORB) and a file description. See section 2.0 for a description of the file structure.
The algorithms developed by Neale et al. (1990) and McFarland et al. (1990) are used to compute the land products. (See Appendix 1 for references.) The algorithm requires brightness temperatures (TBs) as input. To accomplish this the Pathfinder TAs are converted to TBs using the method of Wentz (1991).
The files were created on a Silicon Graphics VGX class computer with version 3.3, release 3 of the HDF library. If you are receiving these files on tape, they were transferred to tape with the UNIX tar facility. All files have been compressed with the IRIX UNIX compress command. The file size for an uncompressed daily land product file is 16 megabytes.
NOTE: All discussions in this text are for any row major applications written in C. If you are using the HDF FORTRAN interface to read the HDF objects, the arrays will be transposed. For example, array A(5,3) in an HDF C interface would become A(3,5) in an HDF FORTRAN interface. This will apply to all HDF data array discussions in this text.
Table 1. Contents of the Pathfinder HDF Daily Land Product File
| ITEM | HDF OBJECT TYPE | HDF REF No. |
| Version Descriptor | N/A | 1 |
| Land Classification(CLS) | Scientific Data Set | 2 |
| Land Surface Temp (LST) | Scientific Data Set | 3 |
| Latitude (LAT) | Scientific Data Set | 4 |
| Longitude (LON) | Scientific Data Set | 5 |
| A-Scan Start Times (AST) | Scientific Data Set | 6 |
| Orbit Parameters | Scientific Data Set | 7 |
| File Description | Annotation | 8 |
| Land Classification Image | Raster Image Group | 9 |
| Surface Temperature Image | Raster Image Group | 10 |
The DMSP satellite completes just over 14 orbits in a day. An orbit is defined as starting when the satellite crosses the equator going from south to north. Because the Pathfinder land product files are organized by time, it is not uncommon to have a fractional part of an orbit prior to the first full orbit, or a fractional part of an orbit following the last full orbit in the daily land product file. To accommodate this, arrays were set up to store the orbits "side-by-side" from left to right, as shown in Table 2.
The first orbit may be a fractional part of an orbit. Its first scan contains the first data for the 24 hour period. That scan may originate at a point other than the beginning of a true DMSP orbit. Following from left to right are up to 15 more orbits. The last orbit to the right may be a fractional portion of a DMSP orbit. If so, it begins before time 23:59:59 but does not complete the orbit. The remainder of the orbit will appear as orbit 1 in the daily land product file for the following day. The last DMSP orbit, or fraction of an orbit, may appear as orbit 15 or 16. Each orbit or portion of an orbit is separated from the next by a delimiter column for all objects except the A-scan start times and the orbit parameters. See Table 4. for the value used as a delimiter in each object.
Table 2. Orbit Arrangement in Arrays
ORBIT --> ------------------------------------------------------------------------ | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16| ------------------------------------------------------------------------ | | | | | | | SCAN 1 for each orbit | | | | | | | | | | | | SCAN 2 for each orbit | | | | | S| | | | | | | SCAN 3 for each orbit | | | | | C| | | | | | | SCAN 4 for each orbit | | | | | A| | | | | | | SCAN 5 for each orbit | | | | | N| | | | | | | " | | | | | | | | | | | | " | | | | | | | | | | | | " | | | | | | | | | | | | SCAN N for each orbit | | | | | ----------------------------------------------------------------------For each orbit, including the partial orbits, the data has been located in the proper scan position. Any missing scans are identified by missing data flags (see descriptions below for these flag values). This means that if the second orbit of the day is missing there will be a missing data flag in orbit position 2. For the partial orbit at the beginning of the day, any data that falls on the previous day (before 00:00:00 UTC) will not be present in this file and the scan position will be filled with a missing data flag. Likewise, for any last partial orbit, the data after 23:59:59 UTC will be flagged as missing for this day but will appear on the next day's file.
The beginning DMSP orbit (or fraction of an orbit) number and ending DMSP orbit (fraction) number can be found by reading the file description stored with each daily land product file. Use the "getfiledesc" utility described in section 4.2 to get this information from the file.
The number of scans is 1612 for all Pathfinder SSM/I HDF objects except the orbit parameters (see section 2.6). The "width" or number of columns in each product array is 1040. This is calculated from the following:
Each scan has 64 samples. There are 14 orbits plus 2 fractions for a total of 16. The orbits are separated by a delimiter column. Thus (64 samples + 1 delimiter) * 16 orbits = 1040 columns. Table 3 shows the valid data and delimiter columns.
| ORBIT | DATA COLUMNS | DELIMITER COLUMNS |
| 1 | 001-064 | 065 |
| 2 | 066-129 | 130 |
| 3 | 131-194 | 195 |
| 4 | 196-259 | 260 |
| 5 | 261-324 | 325 |
| 6 | 326-389 | 390 |
| 7 | 391-454 | 455 |
| 8 | 456-519 | 520 |
| 9 | 521-584 | 585 |
| 10 | 586-649 | 650 |
| 11 | 651-714 | 715 |
| 12 | 716-779 | 780 |
| 13 | 781-844 | 845 |
| 14 | 846-909 | 910 |
| 15 | 911-974 | 975 |
| 16 | 976-1039 | 1040 |
Table 4. shows the flags used in the geophysical products for missing data, out of bounds data, mislocated data, inappropriate surface types and column delimiters.
Table 4. Flags For Geophysical Products
| FLAG 1 | FLAG 2 | FLAG 3 | FLAG 4 | DELIMITER | |
| CLS | -10 | 30 | 25 | -20 | |
| LST | -10 | -30 | 00 | -40 | -50 |
FLAG 1 = Missing data.
FLAG 2 = Erroneous TA data; either out of bounds or mislocated.
TA values not between 50 and 315 degrees K are considered as
out of bounds.
FLAG 3 = An inappropriate surface type (coast, water, ice, possible
sea ice).
FLAG 4 = Inappropriate land classification for computation of LST.
The following sections provide further details on each of the HDF Land
Product objects.
Table 5. Land Classification Codes
0 indeterminate 1 dense vegetation 2 composite vegetation & water 3 dense agriculture/range vegetation 4 precipitation over vegetation 6 composite soil & water/wet soil 7 flooded conditions 8 precipitation over soil 9 dry arable soil/medium vegetation 10 desert 13 refrozen snow 14 dry snow 15 semi-arid surface 19 wet snowThe HDF data type used to store the land classification values is DFNT_INT16. It will be necessary to use this type in any software written to access the data. The HDF reference number is 2.
The HDF data type used to store the land surface temperature is DFNT_INT16. The HDF reference number is 3.
Table 6. Values of Stored Latitude, Longitude and Start Times
| GOOD | FLAG 5 | FLAG 6 | FLAG 7 | DELIMITER | |
| LAT | LAT*100.0 | LAT*100.0 | -29999 | LAT-200)*100.0 | -10 |
| LON | LON*100.0 | LON*100.0 | -18999 | LON*100.0 | -10 |
| AST | AST*1 | 99999.9 | -189.99 | AST*1.0 | N/A |
where:
LAT = Latitude in degrees
LON = Longitude in degrees
AST = A-scan (low resolution) start time in seconds of day
FLAG 5 = Out of bounds
FLAG 6 = Missing
FLAG 7 = Erroneous/mislocated
Note that the Pathfinder latitudes and longitudes were updated to reflect the geolocation corrections for yaw, along-track and pixel 128 as identified by Wentz (1991). The HDF data type used with this object is DFNT_INT16. The HDF reference number for latitude and longitude are 4 and 5 respectively.
Since there are 16 orbits per day and as many as 1612 scans per orbit, the array for AST is 1612 X 16. There are no delimiter columns in this object. The A-scan start times are derived by subtracting 1.9 seconds from the B-scan (high resolution) start times because the antenna temperatures are originally tagged with only the B-scan start times. Note that the B-scan start times were rounded to the nearest second for the period beginning 1 August 1987 and ending on orbit 5 of 16 September 1987. There is the possibility that the first A-scan encountered in the file may actually have occurred on the previous day. If this is the case, the scan start time for the first scan will be between 86398.1 and 86399.9. It is possible that a scan will exist near the end of the file with a start time equal to or greater than the A-scan start time found at the beginning of the file. The HDF data type used with this object is DFNT_FLOAT32 and the HDF reference number is 6.
Table 7. Orbital Elements File Format
POSITION COLUMN DESCRIPTION UNITS 1 Satellite ID (08, 10, or 11) 2 Julian Day of Data (YYDDD) 3 Epoch time of elements (DD.DDDD) 4 Inclination Angle Degrees 5 Right Ascension Degrees 6 Eccentricity N/A 7 Argument of Perigee Degrees 8 Mean Anomaly Degrees 9 Mean Motion Orbits/day 10 Semi major axis Kilometers 11 Period Seconds/orbit
The HDF data type used with this data object is DFNT_FLOAT32. The HDF reference number is 7.
This object is an ASCII description of the daily land product file. It has the file name, satellite name, Julian date, beginning orbit number, ending orbit number, software version number, file structure version number, HDF version number, Marshal Space Flight Center (MSFC)* tool set version number and the e-mail address and phone number for MSFC User Services. The description can be read with the program "getfiledesc" after you compile it with the HDF library. Section 4.0 contains instructions on how to get the HDF library. The HDF object reference number is 8.
extractlp 1.0 getorblp 1.0 getfiledesc 2.0
*The Pathfinder Land Products data set was transitioned to the National Snow and Ice Data Center (NSIDC) from MSFC January 1997.
All land surface data are stored in Hierarchical Data Format (HDF). A thorough discussion of HDF command line utilities is beyond the scope of this manual. However, HDF and HDF utilities are a public domain software developed by The HDF Group. Up-to-date HDF source code, documentation, HDF newsletters, and user support can all be found at The HDF Group Web site.
The contents of HDF are:
README describing files and subdirectories in ftp/HDF/. FAQ frequently asked questions about HDF. Documentation/ HDF documentation HDF_Current/ Points to most current release of HDF HDF3.3r4/ HDF 3.3 release 4 (latest release) HDF4.0.alpha/ HDF 4.0 alpha release HDF3.3r3/ HDF 3.3 release 3 prev_releases/ releases previous to HDF 3.3r4: HDF 3.3r3 and HDF 3.2r4 contrib/ contributions from HDF users outside and inside NCSA examples/ examples of HDF programs_good for testing, too newsletters/ HDF newsletters tarexamples/ compressed tar files of examples HDFVset/ README _ where to get old/new version of HDFVset HDF-UCD/ HDF-UCD versions 1.0, 1.1, 1.2, and HDF-UCD documentation
By typing: extractlp lp08mi88.080_Pfndr_daily.hdf CLS
(i.e. extractlp "Daily Land Product Filename" "HDF object acronym")
a new HDF file called CLS.88080 from the file
lp08mi88.080_Pfndr_daily.hdf is produced. Run the program with no arguments and it
will list all options as shown below:
ACRONYM FILE TAG TYPE
CLS Land Classification SDS INT16
LST Land Surface Temperature SDS INT16
LAT LAT SDS INT16
LON LON SDS INT16
AST Scan Start Time SDS FLOAT32
ORB Orbital Elements SDS FLOAT32
For example: getorblp "Daily Land Product Filename" "HDF object"
"orbit #(1-16)";
(i.e., getorblp lp08mi88.080_Pfndr_daily.hdf CLS 05
produces a new HDF file called CLS05.88080 from the file
lp08mi88.080_Pfndr_daily.hdf.
The Land Surface Type Classification Algorithm by Neale et al. (1990) was derived during the calibration/validation of the SSM/I instrument. Data from other sensors, such as the AVHRR or Landsat Thematic Mapper, may be more appropriate for land surface classification due to their finer spatial and spectral resolution. However, the classification scheme for the SSM/I was developed to serve as an operational pre-selection criteria for application of the empirical parameter retrieval algorithms over land. Some empirical parameters are land surface temperature and snow parameters such as snow depth and water equivalent. An example of the use of this classification scheme is presented with the land surface temperature algorithms by McFarland et al. (1990).
The Land Surface Type Classification algorithms have undergone some improvements since the first version as described in Neale et al. (1990) and the SSM/I cal/val report (Hollinger et al. 1991). The changes were based on additional statistical analysis as well as the simulation of surfaces using a radiative transfer model (RTM) developed for the SSM/I (Vassiliades 1993). The improvements consist of changes in cut-off values for some of the channel combinations in several of the rules and, where needed, the addition of a condition to decrease the number of misclassifications as well as unclassified footprints. The changes are related to the presence of water bodies or water vapor in the atmosphere, both of which affect the brightness values in the 85.5 GHz channels. The original rules were too sensitive to water vapor in the atmosphere giving erroneous surface moisture classifications.
The other main area of change has been the classification of snow covered surfaces and the confusion of snow with other scatterers such as precipitation and sparsely vegetated surfaces. It is important to note that misclassifications occur between snow cover and precipitation during spring conditions. The rules classify some snow covered surfaces as rainfall or sparse vegetation/semi-arid conditions when in reality surface conditions may be a snow/soil mix or possibly a wet snow surface. In addition, the rules will tend to classify a densely vegetated area mixed with snow as precipitation over vegetation. It is also possible that footprints in the middle of very intense thunderstorms can be classified as snow cover. The only way to avoid the confusion is the implementation of a dynamic database, as suggested by Neale et al. (1990), in order to use the knowledge of previous overpasses for better classification accuracy. If multiple overpasses of the same area are viewed it becomes evident where the confusion between snow and precipitation are occurring. By improving the snow cover classification the algorithm developer has tried to minimize the misclassification of cold deserts and sparsely vegetated surfaces as snow by tightening the dry snow rule.
Due to the large footprint size of the SSM/I and the numerous possible combinations of surfaces within a footprint, misclassifications or sometimes footprints with no classification still occur with this set of rules.
One important aspect in the development of these rules described by Neale et al. (1990) is that the 85.5 GHz channel spatial resolution has been degraded to match approximately that of the 37.0 GHz channels. In this way, the value assigned to the 85.5 GHz concentric footprint with the lower frequency channels is an average of itself with the surrounding eight footprints.
Because the 85.5 GHz vertical polarization channel of the F8 SSM/I failed during the benchmark period, a second set of rules is used when the 85.5 V channel became degraded.
Dense vegetation (Class 1):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 <= 1.9
T85V - T37V >= -2
T85H - T37H < 7.5
Dense Agricultural and Rangeland Vegetation (Class 3):
T22V - T19V <= 4
1.9 < (T19V + T37V)/2 - (T19H + T37H)/2 <= 4
T85V - T37V >= -2
T85H - T37H < 7.5
Precipitation over dense vegetation (Class 4):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 <= 4
T85V - T37V < -2
Composite of water and vegetation (Class 2):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 < 6.4
T85V - T37V >= -2
T85H - T37H >= 7.5
T37V > 254
Composite of water and soil/wet soil surface (Class 6):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T85V - T37V >= 4.2
T37V - T19V >= -12.2
Precipitation over soil (Class 8):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T85V - T37V < -10.6
T85H - T37H < -6.2
T19V > 266
Dry snow (Class 14):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T37V - T19V < -7.8
225 < T37V <= 257
T19V <= 266
Wet snow (Class 16):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 < 4
T37V - T19V >= -1.3
T85V - T37V < 4.2
253 < T37V <= 266
T37H >= T19H
T85H >= T37H
T19V <= 266
Re-frozen snow (Class 19):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T37V - T19V < -7.8
T37V <= 225
Desert (Class 10):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 >= 19.7
T85H - T37H >= -6.2
T19V > 264
Semi-arid/sparse vegetation (Class 15):
T22V - T19V <= 4
10.5 < (T19V + T37V)/2 - (T19H + T37H)/2 < 19.7
T85V - T37V < 4.2
T37V - T19V < -1.3
T37V > 257
Medium density vegetation & dry soil surface (Class 9):
T22V - T19V <= 4
4 < (T19V + T37V)/2 - (T19H + T37H)/2 <= 10.5
-10.6 <= T85V - T37V < 4.2
T37V - T19V >= -7.8
For missing 85.5V Ghz channel:
Flooded conditions, standing surface water (Class 7):
T22V - T19V > 4
Dense vegetation (Class 1):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 <= 1.9
-1 <= T85H - T37H < 7.5
Dense Agricultural and Rangeland Vegetation (Class 3):
T22V - T19V <= 4
1.9 < (T19V + T37V)/2 - (T19H + T37H)/2 <= 4
-1 <= T85H - T37H < 7.5
Precipitation over dense vegetation (Class 4):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 <= 4
T85H - T37H < -1
Composite of water and vegetation (Class 2):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 < 6.4
T85H - T37H >= 7.5
T37V > 254
Composite of water and soil/wet soil surface (Class 6):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T85H - T37H >= 10.5
T37V - T19V >= -12.2
Precipitation over soil (Class 8):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T85H - T37H < -6.2
T19V > 266
Dry snow (Class 14):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T37V - T19V < -7.8
T85H - T37H < 10.5
225 < T37V <= 257
T19V <= 266
Wet snow (Class 16):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 > 4
T37V - T19V >= -1.3
T85H - T37H < 10.5
253 < T37V <= 266
T37H >= T19H
T85H >= T37H
T19V <= 266
Re-frozen snow (Class 19):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 >
T37V - T19V < -7.8
T37V <= 225
Desert (Class 10):
T22V - T19V <= 4
(T19V + T37V)/2 - (T19H + T37H)/2 >= 19.7
T85H - T37H >= -6.2
T19V > 264
Semi-arid/sparse vegetation (Class 15):
T22V - T19V <= 4
10.5 < (T19V + T37V)/2 - (T19H + T37H)/2 < 19.7
T85H - T37H < 10.5
T37V - T19V < -1.3
T37V > 257
Medium density vegetation & dry soil surface (Class 9):
T22V - T19V <= 4
4 < (T19V + T37V)/2 - (T19H + T37H)/2 <= 10.5
-6.2 <= T85H - T37H < 10.5
T37V - T19V >= -7.8
The algorithms used for the Pathfinder data sets follow the same rationale discussed by McFarland et al. (1990) and McFarland and Neale (1991) as to the selection of the channels used for the algorithms. The difference is that a much larger data set was used for the statistical regression which included data from different seasons as well as data from descending overpasses.
McFarland et al. (1990) used a gridding scheme to merge SSM/I footprints and average weather station data in a 0.25 degree lat/lon grid cell. In the corrected lat/lon center of each SSM/I concentric footprint was used to search for weather stations that fell within the 3 dB footprint of the 37 GHz footprint (approximately 33 km in diameter) for each overpass in the available data. The algorithm developer believes that this decreased the noise in the data sets.
The temperature values used in this effort consisted of the screen temperatures for the weather station with observation times of either 6 am or 6 p.m., the closest local times to the SSM/I F8 overpass. Though the screen air temperatures at 6 p.m. could be cooler than the surface temperature at the location for some surface types, the inclusion of the 6 p.m. data in the statistical regression for the new algorithms compensates for the error of applying the old algorithms to retrieve surface temperatures for usually warmer descending overpasses. Numerous comparisons between the old and new algorithms against "ground truth" data demonstrated that the new versions of the algorithms performed better overall for both ascending and descending overpasses.
The new version of the algorithms work with the new version of the land surface classification rules as described above. Algorithms were developed for several classes: dense vegetation; dense range land/agricultural vegetation; medium density vegetation/arable soil; wet soil surface; and sparse vegetation in semi-arid/desert areas. The equations and their coefficients are shown in Table 1. The statistics resulting from the linear regression procedures are shown in Table 2 and indicate a reasonable linear fit and the root mean square error (RMSE) values.
LST=C0 + C1*T19V + C2*T19H + C3*T22V + C4*T37H
CLASS 1 -- Dense VegetationLAND CLASS C0 C1 C2 C3 C4 1 -36.77 0.461 -0.148 0.544 0.317 3 -17.447 0.295 0.319 1.195 -0.711 6 37.716 0.178 -0.057 1.271 -0.493 9 1.866 -0.537 0.216 1.432 -0.068 10 & 15 34.973 -0.362 0.225 1.361 -0.303
Table 2. Results of Statistical Regression in the Development Of The New Land Surface Temperature Algorithms
Class F Value DF RMSE R-Square 1 675 562 3.20 0.826 3 3342 4288 4.04 0.757 6 2386 2486 3.93 0.793 9 9564 14931 4.99 0.710 10 & 15 1267 1787 4.74 0.739
Holinger, J.1991. ,DMSP Special Sensor Microwave/Imager Calibration/Validation, Final Report, Vol. II. Naval Research Laboratory, Washington, D.C.
McFarland, M. J., and C. Neale. 1991. Land Parameter Algorithm Validation and Calibration, DMSP Special Sensor Microwave/Imager Calibration/Validation, Final Report Volume II, Naval Research Laboratory, Washington, D.C.Chapter 9:1-64.
McFarland, M.J., R. L. Miller, and C. M. U. Neale. 1990. Land Surface Temperature Derived From the SSM/I Passive Microwave Brightness Temperatures, IEEE Transactions on Geoscience and Remote Sensing. 28(5):839-845.
Neale, C.M.U., M.J. McFarland, and K. Chang. 1990. Land Surface-Type Classification Using Microwave Brightness Temperatures From The Special Sensor Microwave/Imager, IEEE Transactions on Geoscience and Remote Sensing. (28)5:829-838.
NSIDC User Services
National Snow and Ice Data Center
CIRES, 449 UCB
University of Colorado
Boulder, CO 80309-0449 USA
phone: +1 303.492.6199
fax: +1 303.492.2468
form: Contact NSIDC User Services
e-mail: nsidc@nsidc.org