Airborne Rain Mapping Radar (ARMAR)
(TEFLUN-B Field Campaign)

The ARMAR measurements were collected during the second phase of the TExas FLorida UNderflight field experiment (TEFLUN-B) which took place during the month of August and September 1998 and was focused on East Florida . The ARMAR instrument was flown aboard the NASA DC-8 aircraft under the direction of principal investigator Dr. Steve Durden of NASA/Jet Propulsion Laboratory.

Sponsor

Characteristics

ARMAR data is made available on anonymous ftp site for the priority days ( 225/Aug13, 238/Aug26, 248/Sept5) only. The ARMAR data set provided here is calibrated and earth located. since the data status is preliminary, before using this data in your publication, please contact the Principal Investigator Dr. Steve Durden (sdurden@jpl.nasa.gov).

The Files

The ARMAR data files are being archived in their native binary format. The typical data file sizes are 10 MB. Software is provided to convert the ARMAR data to NCAR's DORADE format.

File Naming Convention

Typical file sizes are around 10 megabytes, which corresponds roughly to 5 minutes of data.

Data Format

Format of Quick-look/Browse Product

Quick-look (browse) ARMAR images have been produced for a few selected flight lines during CAMEX3. Those files with gif or jpg extensions were produced during the experiment and provide annotation. Their names correspond to the year, month, day, and time (GMT) at the start of the image. Files with the sun extension are sun raster format and were produced after the complete data processing and first calibration. The vertical scale is approximately 15 km; the horizontal scale is the along-track time and is the number of pixels times 1.8 seconds. This was chosen as either 6 minutes or 30 minutes. The files name is the julian day and time followed by an underscore and a number. The number is 1 for reflectivity, 2 for doppler, 3 for LDR, and 4 for h-v correlation coefficient. The color bar at the end of the image corresponds to * reflectivity from 51 dBZ (white) down to 0 dBZ (blue) * Doppler from +17 m/s away (white) to -11 m/s toward (blue) * LDR from - 5 dB (white) down to -30 dB (blue) * rho from 0.64 (white) to 0.99 (blue)

Black areas are those with low SNR or no data.

Format of calibrated data

The format of the calibrated data is a sequence of headers. Each header begins with "#" followed by a capital letter unique to the particular type of header.

#V header: The format is #V, followed by 156 bytes, providing the processing software version numbers. Hence each file begins with first two bytes as #V.For CAMEX3 parts of the data were processed on the Sky processor with a single i860 and are version 28. Data were also processed on a new Sky system containing four i860s. The software on the new Sky processor is version 100.

#A header: a binary header (with first two bytes being #A) storing information for a single radar radial or ray,including antenna pointing, brightness temperature, radar data

#C header :containing DC-8 DADS navigational and aircraft data . The C-I headers are saved in ASCII just as they are received from the DC-8 DADS system. Each line is terminated by a carriage return and line feed. Because we do not modify the format of the DADS data, we refer the reader to the DC-8 DADS group at NASA DFRC for information on the format of these headers. A full set of DADS headers are recorded about once per second, and each header can appear at random intervals with respect to the radar data (A) headers.

An A-header begins with an 80 byte structure which describes characteristics of the data. This is followed by the radar data itself. The data contains several possible parameters: reflectivity in dBZ, velocity and spectrum width in m/s, HH-VV phase difference in degrees, and HH-VV correlation coefficient (unitless). These quantities were computed as floating point numbers and have been converted to 2 byte integers to save space. The floating point quantities were multiplied by 100 before converting to integers and can be recovered by converting the 2 byte integers to floating point and then dividing by 100. The exception is the HH-VV phase difference, which was mutliplied by 10 rather than 100. The data have been created on a Sun SPARCstation, which uses Big Endian ordering and IEEE format. The order in which the radar parameters are stored is as follows: all range bins of the first radar parameter (usually reflectivity), followed by all range bins of the second radar parameter, etc. A maximum of 400 range bins are possible. In Table 1 we list the contents of the A-header.

Table 2 shows the radar parameters that are found in the A-header for each possible value of the data type parameter.

As an example, suppose one has read the "#A" characters and then the 80 byte structure. Then from that structure one gets a data type of 3, and number of range bins of 310. From Table2, type 3 indicates that single polarization doppler data follows. So one must read, first, 310 words of dBZ values into an array, then 310 doppler velocity values, and then 310 values of doppler width. In all, this header occupied exactly (3x2x310+80+2=) 1942 bytes of mass storage (including the "#A" identifier).

The A-headers are loosely organized by the cross-track scans of the antenna positioner. Preceding each antenna scan is always an A-header of data type 8 or 9 which contains the noise floor mean and standard deviation. This noise floor information is applicable to all radar pulses of the subsequent antenna scan. In general, radar data within 5 dBZ of the noise floor standard deviation should be considered unreliable. There are typically about 20 radar rays (or A-headers) across a single scan of the antenna. Each radar ray is typically an average of about 250 raw pulses (the actual value can be found in the A-header structure, as shown in Table 1). Each radial is numbered across the scan, with the last radial in the scan being 1 (see Table 1).

The range from the aircraft to the first pixel of radar data is given in the A-header in meters. The range to subsequent pixels is computed as:

r_i = i * Delta t * 15 + r_0

where Delta t is the time between range bins as stored in the A-header in 100 nanosecond units. This variable normally has a value of 4, giving a pixel spacing of 60 m.

The specific antenna pointing angle of a given pulse is contained in the 80 byte structure, as indicated in Table 1. The starting and ending azimuth angles and elevation angle are with respect to the nadir oriented antenna positioner. Negative azimuth is to the left, and positive elevation is aft of the aircraft. The two azimuths define the cross-track swath, or window, over which the 250 (or so) raw radar pulses were averaged. Note that these parameters do not account for aircraft roll, pitch, yaw, or positioner mounting offsets. As an alternative to the antenna angle, the Cartesian unit vector given in the A-headergive the antenna vector as corrected for mount offsets and aircraft attitude. This vector is multiplied by 10000 to fit the integer format of the data storage. This vector places the antenna beam with respect to the aircraft ground track. This vector together with the heading, drift angle, latitude, longitude, and altitude as taken from the DADS headers will locate a given pixel in Earth coordinates.

Following tables provide the description of 80 bytes of header #A

.

The Companion Software

The DISC distributes read and conversion software that was provided by the data producers at JPL:

• read_arm.c Code to read ARMAR data and display it to the terminal
• arm2dor.c Code to convert ARMAR data from its native binary format to DORADE format

The ARMAR data was produced using a Sun workstation that loads the most significant byte of a two-byte integer first whereas a PC, for example, loads its integers little-end-first. Program read_arm.c has been written to detect and correct for this system difference. Program arm2dor.c at this time does not correct for this difference and may require a byte swap to run correctly on non-Sun implementations.

In addition to read software, number of other JPL documents are also available to assist users in reading and understanding ARMAR data. The JPL documents and text version of this DISC produced html document "README_tfb_armar.shtml" is also included in the distribution package. The DISC document is based on the JPL produced more comprehensive user guide "armar_users_guide.ps" and the document "quick.txt". These documents provide a description of the instrument, its operating modes; data collection, processing, and calibration; data quality assessment; and data format (special JPL internal format)and are available on line from the directory

ftp://disc2.nascom.nasa.gov/data/TEFLUNB/aircraft/nasa_dc8/armar/document.

Data Access

FTP Site

ARMAR Calibrated Data

ARMAR Browse Images

or directly via FTP at

ftp disc2.nascom.nasa.gov

login: anonymous

password: < your internet address >

cd data/TEFLUNB/aircraft/nasa_dc8/armar/

Points of Contact

ARMAR was completed in late 1991, and the first airborne testing was performed in May of 1992. Additional tests were completed in December 1992, and the system was deployed during TOGA COARE in the western Pacific in early 1993. The ARMAR data described here were collected during CAMEX3 in August-September 1998

Instrument Characteristics

Principles of Operation

ARMAR is a pulse compression radar. The transmitted signal is a relatively long, frequency modulated signal, referred to as a chirp. The ARMAR digital controller causes the chirp generator to produce chirps with the selected length, spacing, and start frequency. In normal operation, the chirp is a linear frequency modulated upsweep with 4 MHz bandwidth and amplitude weighting for range sidelobe suppression. The chirp start frequency is 60 MHz. A sequence of up to seven different chirps can be transmitted, with each chirp possibly differing in polarization and other parameters. The chirps are upconverted to 13.8 GHz and amplified by a high power traveling wave tube amplifier (TWTA). The amplified chirp is then sent to the antenna system. A small amount of power is sent directly to the receiver through a calibration loop. The TWTA is operated in the non-saturated mode to maintain the desired chirp amplitude characteristics.

The antenna system consists of a dual, linearly polarized scalar feed horn which illuminates a precision offset parabolic reflector. The signal is focused by the parabolic reflector and reflected by a flat mechanically scanned elliptical reflector which scans the beam +/- 25deg in the crosstrack direction. The reflector can also be pointed or scanned up to +/- 10 deg in the along track direction, if desired. Scanning can be done either in a bow-tie mode in which the elevation angle is changed across the scan or in a retrace mode in which the antenna is scanned with constant elevation angle from left to right and then rapidly re-positioned to the left. Both transmit and receive polarizations can be varied on a pulse to pulse basis, allowing a combination of like- and/or cross-polarization data to be collected.

The signal reflected from the rain is collected by the antenna and then amplified by a low noise amplifier (LNA). Following the LNA, the received signal is downconverted to the 70 MHz intermediate frequency (IF). Here, the signal is split into radar and radiometer signals. The radar signal is passed through a programmable attenuator before IF amplifiers and filters. In the final stage, the signal is downconverted to baseband (offset video, rather than I-Q) where it is digitized by a 12-bit analog/digital converter (ADC) at a rate of 10 MHz and recorded. The radiometer signal is acquired during a short time within each interpulse period after return from the transmitted pulse has reached zero. This signal is integrated in analog circuitry, sampled every 10 ms, averaged, and recorded. At the end of each antenna scan, or every few seconds in non-scanning operation, the radar enters a calibration mode in which the radar signal passing through the calibration loop, as well as radiometer noise diode and reference load measurements, are recorded.

The main control computer for ARMAR is a 486 Personal Computer (PC), which runs a Quick Basic program for overall system control. This computer serves as the operator interface, allowing the operator to start, stop, and re-configure the radar and display data. The radar configuration includes such parameters as transmit chirp length, chirp start frequencies, polarization(s), and antenna scanning parameters. A sequence of up to seven chirps with differing characteristics, such as IF and polarization, can be programmed. A VME 68000 computer is used for real-time control of the radar through a specially designed digital controller. The VME computer also controls the antenna scanner. The signal conditioner and formatter board in the PC receives the data from the 12-bit ADC and transmits it to a high speed Ampex DCRSi tape recorder.

Auxiliary data, including radiometer voltage, system temperatures, radar configuration, timing information, and aircraft parameters from the DC-8 data system, are recorded along with the radar data. The digitized voltages are recorded in binary, while the auxiliary data are recorded in the form of ASCII header strings. Radar data and auxiliary data are recorded as received.

The radar system is mounted in the cargo bay of the NASA DC-8 aircraft. The antenna beam is directed through an opening in the bottom of the DC-8 aircraft. A thin radome covers this observation port, and the entire antenna system is surrounded by a pressure box. The radar RF section is mounted on a plate which lays on top of the pressure box. This plate also includes IF, video, and ADC sections. A rack for the TWTA and other equipment is mounted in the cargo bay next to the pressure box. The system computer, tape recorder, and data processing system are mounted in a rack in the DC-8 cabin where the ARMAR operators sit.

TEFLUN-B/CAMEX-3 Data Collection:

• S - single polarization type 1 (HH, where H is E-field aligned with aircraft motion), no Doppler, 1 kHz PRF, (low data rate survey mode)
• D - HH-HV dual-pol (type 5), Doppler, retrace scanning, 4.6kHz PRF
• F - HH-VV dual-pol (type 4), Doppler, retrace scanning, 4.6 kHz PRF

The raw data from CAMEX3 are stored on 23 high density cassettes, labeled CA-CW. Each tape is divided into several records with the total tape holding approximately 2 hours of data. The exact duration varies depending on which configurations were used. There is a minimum of about 2 minutes in time between records on a tape because the radar is stopped and restarted. The first record on a tape is used for housekeeping purposes, so the data records begin with record 2. The high density tape record numbers are retained in the pre-processed data. Data files produced from tape CC, record 3, for example, are denoted, cc3_1, cc3_2, etc. One calibrated data file is produced from each pre-processed data file. However, at the calibration step, the file name is changed to reflect the UTC time of collection.

Each calibrated file is denoted by dddhhmm.ARM, where ddd is the Julian day, hh is the hour, and mm is the minute. Note that the times of the file names are approximate; data times should always be taken from the

#A header. Typical file sizes are around 10 megabytes, which corresponds roughly to 5 minutes of data.

Algorithm

ARMAR Data Processing:

The data processing system consists of a Sun workstation and a Sky Computers i860-based co-processor. The first step in the processing of ARMAR data is referred to as pre-processing and consists of pulse compression and averaging. The PC and formatter board are used to play the recorded data into the data processing system. The auxiliary information is sent over a serial line to the Sun computer, while the radar voltages from each pulse are sent directly to the Sky co-processor and correlated with a model of the transmitted signal to perform pulse compression. This correlation is performed by taking the FFT of each received pulse, multiplying by the reference function transform, and inverse transforming. As mentioned above, the raw voltages are real, offset video samples rather than I-Q samples; complex data (i.e., analytic signal) are generated by zeroing out the negative frequencies in the transform domain. The inverse FFT is done using one-half the number of points in the forward FFT. Averages of various second order statistics are then computed from the complex, pulse-compressed data. Which statistics are computed depends on the mode in which data were acquired. This process is computationally intensive. Several months of almost continuous operation were required for the data from CAMEX3.

In the second phase of processing, the pre-processed data are passed through a program which cleans up, calibrates, and time stamps all data. For reflectivity, the average magnitude squared is converted to the ratio of received to transmitted powers using calibration chirp data and laboratory measurements of the transmit path, receive path, and calibration loop attenuations. This ratio, along with the range, range resolution, and antenna pattern are used in the radar equation to calculate the equivalent reflectivity factor Z_{eq}. An estimate of the system noise floor is subtracted from the total power before conversion to reflectivity. This estimate of the noise floor is based on the radar data itself, using the signal in range bins with no rainfall. For Doppler mode data, the complex compressed signal is used in a pulse-pair algorithm to estimate the mean velocity and spectrum width.

The doppler velocity has been corrected for aircraft motion by subtracting the surface velocity from all range bins. The surface velocity is found by fitting the measured surface velocity for all beams in a scan to an exact equation for Doppler versus scan angle. The result is then used to compute the surface velocity for each beam. For observations over the ocean this should be more accurate than using the DC-8 DADS data. Note that over land with topography this method may be biased and the DADS data can be used in its place.

For dual-polarization modes, appropriate polarimetric quantities (e.g., HH-VV phase difference) are calculated, in addition to reflectivity and velocity. Radiometer voltages are converted to brightness temperatures using the noise diode and reference load measurements.

Data Quality Assessment

The following measurements test system calibration:

• Stability: The calibration pulse power has been examined over CAMEX3 experiment. This signal showed little variation, verifying that both the transmit and receiver chains were stable (to +/- 1dB). Note that any changes in this level are removed in the calibration procedure.

• Noise: The minimum observable reflectivity was calculated using observations. After subtracting noise, the system sensitivity is approximately 10 dBZ at 10 km range.

• Absolute calibration: Data was examined at the 10 degree cross section of the ocean under clear conditions. The average measured is approximately 7 dB, which is similar to cross sections measured by 14 GHz scatterometers and predicted by models at low to moderate windspeeds.This is lower than the average value in TOGA COARE (Durden et al. 1997) and could imply that the CAMEX3 reflectivities are biased low by about 0.5 dB.

• Relative calibration: Nadir-looking measurements of the ocean at HH and VV polarizations should produce the same backscattered power. We verified that this is the case for ARMAR to 0.1 dB.

• Polarization parameters: For mode D the cross-pol isolation is near 30 dB, based on measurements of the ocean surface at nadir in clear conditions. Most data were acquired in HH-VV mode (mode F), and the measured polarization parameters are the HH-VV correlation coefficient and the HH-VV phase difference. For the phase difference there is an offset of about 50 degrees due to the orthomode transducer, which provides different path lengths for each polarization. The correlation coefficient is normally very close to unity in rain and drops to 0.9 or less in melting ice. A few times were noted when the correlation coefficient is lower than expected (0.9 versus 1.0). This was noted, for example, on day 260 around 19:45 GMT and day 264 around 19:00. This is due to random changes in the start up phase of the chirp generator and is dicussed below under Doppler spectrum width.

• Radiometer calibration: The observed clear ocean brightness temperature were compared with known brightness temperatures under these conditions. The observed and expected are in reasonable agreement. The ARMAR radiometer brightness temperatures during CAMEX3 may be biased high by a few K.

• The analytical expression in Doviak and Zrnic (1984) and Monte Carlo simulations were used to derive the theoretical standard deviation for pulse-pair velocity measurements. These were compared with the residuals based on fitting a three term expression to the observed velocity versus radar azimuth angle. The expression has a constant term and terms proportional to the sine and cosine of the azimuth angle. This is an exact model, so the residuals of this fit are a good estimate of the velocity standard deviation. For the number of pulses typically averaged (approx 300), the velocity standard deviation should be 0.3-0.4 m/s. This is the observed residual in many cases, although the residual is often larger, particularly over land.

• some data were acquired with constant azimuth angle and variable elevation angle. The change in observed velocity was compared with the change that would be predicted from the change in elevation angle. The observed and predicted changes were within a fraction of a m/s.

• the theoretical Doppler spectrum width for 230 m/s ground speed is 4.5 m/s. Widths measured for the ocean surface during CAMEX3 were examined and found them to be 4-5 m/s, depending on the aircraft ground speed. As noted above the system occasionaly gets into a state where the chirp startup phase jumps randomly between two or more values so that some pulses in a sequence have one phase and some another. This broadens the Doppler spectrum and also affects the HH-VV correlation coefficient and phase. These parameters should still be useful in a qualitative sense when this condition occurs. The Doppler velocity should not be biased in any case. The user can recognize this condition by examining the surface return for large spectral widths( greater or equal to 5 m/s) and low HH-VV correlation coefficient (approx 0.9).

• the surface Doppler estimated from INS (DADS) data depends on knowledge of the antenna pointing. The accuracy of the DADS data makes this difficult, which is why the radar surface measurements are used instead. However, in some cases (over land with topography) it may be preferable to use the DADS data. To estimate the antenna mounting offsets, A model was developed which predicts the velocity that the radar would measure for the ocean's surface, given the radar antenna azimuth and elevation angles and the aircraft roll, pitch, and yaw. Parameters in the model were the unknown offsets in roll, pitch, and yaw of the radar antenna relative to the aircraft body, determined by fitting the model to the observed ocean surface velocity over an antenna scan. Roll and yaw offsets were found to be near zero degrees, while the pitch offset is 3.8 degrees. This pitch offset means that when the radar operator specifies an antenna elevation angle of zero, the radar antenna is actually pointing forward by 3.8 degrees (for a pitch angle of zero). A radar elevation angle of 3.8 degrees is needed to move the antenna to nadir. This offset was known when acquiring data, so the antenna was usually pointed close to nadir in the along-track (pitch) direction.

• the surface velocity, used to correct each beam for aircraft motion, is estimated by fitting the raw surface velocities to a three term equation, as discussed previously. In a few cases the surface velocity was aliased, causing this procedure to be in error. These cases can often be recognized by large differences between the observed and predicted Doppler. The user can recover the original radar velocities by simply adding the measured surface velocity to the data. Following dealiasing, the surface velocities can again be subtracted.

Recommendations

• add 0.5 dB to the reported reflectivities

• compute a new brightness temperature from the reported brightness temperature as

  T_{new}=1.08 T_{old}-23.7

Doviak, R. J. and D. S. Zrnic, Doppler Radar and Weather Observations. Academic Press, 103-107, 1984.

Durden, S. L., E. Im, F. K. Li, W. Ricketts, A. Tanner, and W. Wilson " ARMAR: An airborne rain mapping radar," J. Atmos. Oceanic Tech., vol. 11, no. 3, pp. 727-737, 1994.

Durden, S. L., A. Kitiyakara, E. Im, A. B. Tanner, Z. S. Haddad, F. K. Li, and W. J. Wilson," ARMAR observations of the melting layer during TOGA COARE," IEEE Trans. Geosci. Remote Sensing, vol.35,no.6, pp. 1453-1456, November 1997.

Tanner,A., S. L. Durden, R. Denning, E. Im, F. K. Li, W. Ricketts, W. Wilson, " Pulse compression with very low sidelobes in an airborne rain mapping radar," IEEE Trans. Geosci. Remote Sensing, vol.32, no. 1, pp. 211-213, 1994.

Durden, S., A. Tanner, W. Wilson, F. Li, E. Im, W. Ricketts, "The NASA/JPL airborne rain mapping radar (ARMAR),"Proc. 11th International Conference on Clouds and Precipitation, Montreal, August 1992, pp. 1013-1016.