Close

Service Interruption

Data Set Documentation

NOAA-NASA SSM/I Pathfinder Daily Land Products Data Sets

Table of Contents

1.0 Introduction

This data set documentation contains information on the structure of the Hierarchical Data Format (HDF) daily land products files, instructions on how to get and install HDF on your machine and a description of three utilities: extractlp -- a program which extracts specified items from the daily land product file; getfiledesc -- a utility that pulls out the file description and prints it to the screen; and getorblp -- a program that extracts specified items and orbits from the daily land product file. Appendix 1 contains a brief discussion of the technical and scientific aspects of the data set, including pertinent references.

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.

2.0 Daily Pathfinder HDF Land Product File Structure

Each daily land product file has the following contents:

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.

Table 3. Data Columns by Orbit

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

For example, to get CLS for orbit 2 you would use an HDF utility to read the land classification object into a 1612 X 1040 array, then copy columns 66 through 129 into another array dimensioned 1612 X 64. The software, "getorblp", is described in section 4.0.

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.

2.1 Land Classification (CLS)

This HDF object contains the land classifications computed from the low resolution TAs. Each value is stored as a 2-byte integer. There is no scaling factor. Table 5 shows the land classification codes and the associated description.

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 snow
The 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.

2.2 Land Surface Temperature (LST)

This HDF object contains land surface temperatures computed from the low resolution TAs. The values are computed in degrees Kelvin. After multiplication by 10, the temperatures are stored as 2-byte integers. Surface temperatures can not be calculated for all land classification types (e.g., snow surfaces). For these situations, a flag value is used. See Table 4.

The HDF data type used to store the land surface temperature is DFNT_INT16. The HDF reference number is 3.

2.3 Latitude Values (LAT)

The latitude values are stored as signed 2-byte integers ranging from -9000 to 9000 (-90.00 to 90.00 degrees). The values are negative south of the equator and positive north of the equator. There is a value for every low resolution pixel. The array dimensions are 1612 X 1040; there are 1612 scans, each orbit has 64 samples per scan, and each orbits is separated by a delimiter column. The precision of the latitudes is to 0.01 degrees. Thus a value of 7524 means 75.24 degrees North latitude. Missing scans are denoted by the value -29999. Erroneous or mislocated scan latitudes have 200 subtracted from them before being stored. To retrieve an erroneous or mislocated latitude, divide the stored value by 100 and then add 200 to the result (see Table 6).

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.

2.4 Longitude Values (LON)

Similar to the latitude values described above, the longitudes are stored as signed 2-byte integers ranging from -18000 to 18000 (-180.00 to 180.00) degrees. The values are negative west of the Prime Meridian and positive East. Missing scans are denoted by a value of -18999. Erroneous or mislocated scans are not tagged since that information can be extracted from the latitude object. See Table 6 for stored longitude values. The HDF object is otherwise organized exactly as that for the latitudes described above.

2.5 A-Scan Start Times (AST)

The low resolution (A-scan) start times are stored in this HDF object. The times are stored in seconds of day (0.0 to 86399.9). Each value is stored as a 4-byte real number (floating point). Any missing scans are identified by -189.99. Any out of bounds values are flagged by 99999.9. See Table 6 for the stored AST values.

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.

2.6 Orbit Parameters

The Orbit Parameters object contains a set of classical orbital elements. These are derived from the two-line elements for the DMSP satellite and time tagged to 00:00:00 UTC. The file format follows.

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.

2.7 File Description

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.

An example of a file description is:
SSM/I Land Classification and
Land Surface Temperature
File ID = lp08mi88.080_Pfndr_daily.hdf
Satellite = F8
Julian Date = 88080 Beginning Orbit = 3868
Ending Orbit = 3882
Time Of First Scan (hhmmss) = 000001
Time Of Last Scan (hhmmss) = 235959
SSM/I PATHFINDER Software Version Number 1.0
MSFC File Structure Version Number 1.0
HDF Version Number 3.3
MSFC Tool Set Version Numbers:
		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.

2.8 Browse Images

Eight-bit raster images are included for the CLS and LST objects. Each image is composed of 720x720 pixels depicting both ascending and descending passes. The values are gridded with the last value being retained for the grid box, and then binned into codes for imaging purposes. Consequently, the images are better suited to visualization than to scientific study. A color palette is attached to each image for ease of viewing. One way to view the browse images is with the HDF Collage tool. The HDF object reference numbers for these images are 9 and 10 respectively.

3.0 Accessing the HDF Library and Tools

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 

4.0 How To Compile Programs extractlp, getfiledesc and getorblp

Three programs are included with this distribution. They are extractlp.c, getfiledesc_lp.c and getorblp.c. They run on Silicon Graphics work stations but should port easily to other platforms. One make-file is included with the distribution. It can be used to compile all three programs. To run the makefile issue the following UNIX command:
make -f Make.lp
With minor changes it should compile the utilities on your machine.

4.1 extract

The routine extractlp.c will be the most useful. This utility extracts an HDF object from the daily TA. It produces a new HDF file with the name of the HDF object selected.

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:

Usage: extractlp "filename" "list of HDF object"
This file contains the following HDF objects. Use the code when selecting an object to extract. You may select several codes at once, separated by a space.

      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
        

4.2 getfiledesc_lp

The getfiledesc_lp utility prints the text of the HDF annotation contained in the daily Land Product file. Its usage is getfiledesc_lp "Daily Land Product filename". The result should look like the example file description found in section 2.7.

4.3 getorblp

The getorblp utility extracts an orbit for any object from the daily land product file. It produces a new HDF file with the name and orbit number of the HDF object selected.

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.

Appendix 1 - Land Surface Products

Land Surface Type Classification Algorithms

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.

For all 7 SSM/I channels:

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
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

Appendix 2 - Land Surface Temperature Algorithms

The Land Surface Temperature algorithms were those developed by McFarland et al. (1990) during the calibration/validation effort of the SSM/I. However, those algorithms had some problems as discussed by McFarland and Neale (1991). One problem was a consistent bias retrieval towards lower temperatures because the data sets used in the regression procedures only included ascending (early morning) overpasses of the F8 instrument. Screen temperatures from the NOAA Cooperative Network of weather stations and not skin temperatures were used as the "ground truth" temperature values, due to the unavailability of Operational Linescan System (OLS) temperature values at the time.

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.

Table 1. Land Surface Temperature Retrieval Algorithms For SSM/I For Different Surface Types.

LST=C0 + C1*T19V + C2*T19H + C3*T22V + C4*T37H

LAND 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

CLASS 1 -- Dense Vegetation
CLASS 3 -- Dense Agricultural/Rangeland Vegetation
CLASS 6 -- Medium Density or Sparse Vegetation with Wet Soil Background
CLASS 9 -- Medium Density Vegetation/Arable Soil
CLASS 10 AND 15 -- Sparse Vegetation in Semi-Arid/Desert Regions

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

REFERENCES

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.

Contact Information:

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

Document Information:

Revision Date: 2 January 1997
Review Date: January 1997
Document ID: nsidc-0042
Citation:: This field is intentionally left blank.
Document Curators: NSIDC Writers
Document URL: http://nsidc.org/data/docs/daac/nsidc0042_ssmi_land.gd.html