1. Introduction

In 1991, NASA began a coordinated program of airborne and spaceborne observations of the Greenland ice sheet. The primary objective of this program is to determine the mass balance of the ice sheet and its impact on global sea level (Reference Krabill, Martin and FrederickKrabill and others, 1995b. To assess the mass balance,three primary measurements are required: surface topography, ice thickness and ice velocity. Accurate measurements of surface topography provide an indication of whether the sections of the ice sheet are in mass equilibrium; repeated measurements yield a direct measure of variations in ice-sheet volume (Reference Krabill, Martin and FrederickKrabill and others, 1995a). Regional measurements of ice thickness combined with measurements of surface velocity can be used to estimate thinning or thickening rates through application of mass-continuity arguments (Reference PatersonPaterson, 1994).

Surface topography is determined to the 10cm level using an airborne laser operated on the NASA P-3 aircraft. Aircraft navigation is with a global positioning system (GPS ). Surface velocity is estimated by repeated GPS measurements of Stations fixed on the ice-sheet surface. Ice thickness and the internal structure of the ice sheet are measured using the ice-sounding radar, also on the P-3. The radar also measures the internal structure of the ice sheet and the physical properties of the glacier bed. The latter information is used to help explain the dynamic behavior of the ice sheet and, hence, in developing a hypothesis to explain any detected changes in ice-sheet volume.

In this paper we discuss the design and application of a new and improved ice-sounding radar, the improved coherent Arctic radar depth sounder (ICARDS. The system was designed to be flown in an aircraft or operated from the ice-sheet surface. It has been shown to be effective in acquiring ice-thickness data for ice thicknesses in excess of 3 km for cold ice. Perhaps more important, it has proven very effective for acquiring ice-thickness and internal-structure data on outlet glaciers located around the ice-sheet margin. Because ice wastage occurs through the outlet glaciers and around the margins of the Greenland ice sheet, data from this radar provide new information about processes occurring in the dynamically active ice-sheet perimeter.

2. Background

Radio-echo sounding (RES) has been used for the purpose of obtaining thickness measurements of glacial and sheet ice since the early 1960s, The origins of this approach can be traced to 1933 at Admiral Byrd's base, Little America, Antarctica, where the first indication that snow and ice are transparent to high-frequency radio signals was observed. An investigation by U.S. Army researchers, prompted by pilot reports of the uselessness of radar altimeters over ice, and suggesting the transparency of polar ice and snow in the VHK and UHF bands, led Waite and others to demonstrate in 1957 that a radar altimeter (the SGR 718 operating at 140MHz) could measure the thickness and other features of polar glaciers (Reference Waite and SchmidtWaite and Schmidt, 1961; Reference Evans, Drewry and RobinRobin, 1972). This observation resulted in one of the most important technical advances in glaciology, namely, the development and wide application of RES systems.

Following Waite's demonstration, Evans at Cambridge University's Scott Polar Research Institute (SPRI) developed the first of several VHF systems specifically for radio-echo sounding in 1963. Within the next few years several other research groups began developing and using RES systems, including the U.S. Army Electronics Laboratory (USAEL), the British Antarctic Survey, the Arctic and Antarctic Scientific Research Institute in Leningrad, the Geophysical and Polar Research Center at the University of Wisconsin, the U.S. Army Cold Regions Research and Engineering Laboratory, the Canadian Department of Energy, Mines and Resources, the Technical University of Denmark (TUD), Stanford Research Institute, and the Institut Geografii of the U.S.S.R. Academy of Sciences (Evans, 1967; Reference Evans and SmithEvans and Smith, 1969; Reference GudmandsenGudmandsen, 1969; Reference Weber and AndrieuxWeber and Andrieux, 1970; Reference RobinRobin, 1975a; Macheret and Zhuravlev, 1982). These systems, all short-pulse-type radar systems with operating frequencies of 30-600 MHz, successfully sounded ice sheets, ice caps and glaciers (both temperate and polar) in both Greenland and Antarctica. Both surface-based and airborne measurements were conducted. Thickness estimates from RES systems agree with those from seismic and gravity-based estimates (Drewry, 1975). Reflections from internal layers have also been observed since the first RES observations, and a variety of sources for these internal reflections have been identified, including layers of liquid water (Reference BamberBamber, 1987; Reference Davis, Dean and XinDavis and others, 1990), layering involving small permittivity changes due to changes in acidity (from large volcanic events) (Reference MillarMillar, 1981), changes in the size or shape of air bubbles within the ice (Reference Ackley and KeliherAckley and Keliher, 1979), and variations in ice-Crystal orientation and density (Reference HarrisonHarrison, 1973). Layers produced by the small permittivity changes represent isocrones and are useful in the interpretation of climate information (Gudmandsen, 1975; Reference Jacobel and HodgeJacobel and Hodge, 1995). In addition to measurements of ice thickness and layers, other features and characteristics observed include regions of bottom melting and freezing (Neal, 1979), ice-bottom sliding velocity (Reference DoakeDoake, 1975. glacier velocity (Reference Doake, Gorman and PatersonDoake and others, 1976), sub-ice lakes formed by pressure melting (Reference OswaldOswald, 1975), and bottom crevasses (Reference Jezek, Bentley and CloughJezek and others, 1979). Information on other ice parameters, including signal absorption (Neal, 1976), signal fading patterns (Reference HarrisonHarrison, 1971; Reference BerryBerry, 1975), propagation velocity (Reference RobinRobin, 1975b) and birefringence (Reference BentleyBentley, 1975; Reference HargreavesHargreaves, 1977; Reference Woodruff and DoakeWoodruff and Doake, 1979), have been obtained from RES data.

In the next generation of RES systems we see more specialization. For example, to reduce the reflections from nearby walls in sounding valley glaciers, the system frequency is increased to improve system directivity. A 620 MHz system succeeded in sounding Rusty Glacier in Yukon Territory, Canada, where a 35 MHz system and conventional seismic systems had previously been unsuccessful due to echo obscuration by the transmitted radio and seismic pulses and due to the proximity of the valley walls (Reference Clarke and GoodmanClarke and Goodman, 1975). Also, several low-frequency RES systems were developed specifically for sounding temperate glaciers where absorption losses arc significant due to higher ice temperatures and the presence of liquid water. Systems with frequencies of 1-32 MHz have been successfully used by numerous researchers to sound temperate glaciers (Reference StrangwayStrangway and others, 1974; Reference Watts and EnglandWatts and England, 1976; Reference Bjornsson, Ferrari, Miller and OwenBjornsson and others, 1977; Reference Sverrisson, Johannesson and BjörnssonSverrisson and others, 1980; Watts and Wright, 1981).

Techniques have also been developed for determining additional information on both the geography and roughness of subglacial terrain. Significant changes in the polarization of the returned echoes from land ice and an ice shelf indicate that tidal strain on crystal orientation at the hinge zone results in a large change in birefringence of the ice, suggesting that polarization could be used to distinguish floating from grounded ire (Reference Woodruff and DoakeWoodruff and Doake, 1979). A method for continuously monitoring the returned echo power from a subglacial ice/rock or ice/water interface can detect changes in the reflection coefficient and the absorption properties of the ice, as well as identifying the changing nature of the basal reflection properties along a flowline (Neal, 1976). As part of a physics experiment investigating a theoretical fifth force affecting Newton's inverse square law for gravitation, an extensive series of RES measurements were made around the Dye-3 complex in southern Greenland. Sounding data were collected along 124 radial lines, each about 5 km in length. This produced a map of bedrock topography with an uncertainty of < 5 m over most of the survey area (Reference FisherFisher and others, 1989). A precise grid pattern was flown over the summit region of Greenland with the TUD RES system to obtain both ice-surface and ice-bottom topography to accuracies of ±6 m and ±50 125m, depending on the bottom roughness (Reference Hodge, Wright, Bradley, Skou and VaughnHodge and others, 1990).

As enabling technologies emerged, RES systems became more capable. Digital data acquisition, signal processing and recording significantly improved system capabilities, not the least of which is dynamic range (Reference Clarke and GoodmanGoodman, 1975; Reference SivaprasadSivaprasad, 1978; Reference Wright, Hodge and BradleyWright and others, 1989). Walford and others developed a coherent RES system permitting the measurement of both the amplitude and phase of the received signal (Reference Walford, Holdorfand and OakbergWalford and others, 1977; Reference Walford and HarperWalford and Harper, 1981). Synthetic aperture radar (SAR) techniques were applied lo RES data by researchers from the British Antarctic Survey to produce two-dimensional maps of echo strength showing the grounding line of a glacier in the Antarctic Peninsula (Musil and Doake, 1987). The coherent Antarctic radar depth sounder (CARDS), the first RES system designed completely with solid-stale, computerized components, which is coherent and employs pulse compression to reduce peak transmit power requirements, was field-tested in Antarctica by researchers at the University of Kansas (Reference Raju and MooreRaju and others, 1990).

In the 1990s, a new generation of specialized RES systems has emerged to address specific glaciological questions. A miniature impulse RES system capable of operating from 1 to 200 MHz was developed and field-tested by Canadian researchers for sounding glaciers and ice caps (Reference Narod and ClarkeNarod and Clarke, 1994). Researchers from the University of Minister developed and fielded two ground-based 35 MHz RES systems: the first was a single-pulse system intended to measure the reflections from internal layering with high resolution; the second uses a burst transmitter designed to penetrate the sheet ice and observe the underlying bedrock (Reference Hempel and ThyssenHempel and Thyssen, 1993). Norwegian researchers applied SAR techniques to image the internal structure of the subpolar glacier Slakbreen in Spitsbergen, Svalbard, Norway, using range-gated synthetic pulse (stepped frequency modulation) systems operating from 5 to 20 MHz and from 320 to 370 MHz to investigate various geophysical phenomena (Reference Hamran and AarholtHamran and Aarholt, 1993). Researchers from the University of Alaska at Fairbanks used a 1.7 MHz impulse-type RES system to measure a cross-section of Taku Glacier, Alaska, and estimate the mass-balance flux (Reference Nolan, Motyka and EchelmeyerNolan and others, 1995).

Finally, the value of RES data depends on accurate knowledge of location. Therefore navigation and position-measurement techniques are of great significance to RES systems. When RES systems were first field-tested, navigation relied on observation by system operators, which is significantly hampered in regions without discernible landmarks. In 1967, navigation records, including a galvanometer showing aircraft heading, air temperature, static pressure and airspeed, were recorded along with airborne RES data. By 1971, an inertial navigation system, the Litton 51C, was used to annotate the RES data with position information (Reference Evans, Drewry and RobinEvans and others, 1972; Reference RobinRobin, 1975a). Position-measurement systems employing microwave signals and transponders at known geographic points have been used in ground-based RES experiments within limited areas with an accuracy of ± 10 m (Reference Clarke and GoodmanGoodman, 1975). By the late 1970s, LORAN and satellite navigation were available, providing less accurate yet affordable position information on a global basis suitable for RES applications (Reference Sverrisson, Johannesson and BjörnssonSverrisson and others, 1980). A Doppler navigator linked through a Tactical Air Navigation System navigation computer was used in 1983 to provide a continuous read-out of latitude and longitude with sub-kilometer accuracy (Drewry and Liestol, 1985; Reference Gorman and CooperGorman and Cooper, 1987). Currently, differential GPS data are collected with RES measurements, reducing positional uncertainties to +2-5 m (Reference Nolan, Motyka and EchelmeyerNolan and others, 1995).

Table 1 shows an updated summary of the characteristics of the various time-domain radar sets; it was fist tabulated by Reference Clarke and GoodmanGoodman (1975).

Table. 1.

An updated summary of various time-domain radars used in mdioglaeiohgy

3. System Design and Evaluation

The system described here is a modified and improved version of the radar system reported by Reference Raju and MooreRaju and others (1990). Figure 1 shows the basic elements of the radar system. It op-crates at a center frequency of 150 MHz with peak transmit power of about 200 W. The transmitter generates a 1.6 μs long pulse using a surface acoustic wave (SAW) expander. This pulse is coupled to the transmit antenna consisting of a four-element dipole array. An identical receive antenna collects the reflected signal. This signal is amplified using a low-noise receiver and compressed using a SAW filter to a pulse about 60 ns wide. A coherent detector generates in-phase (I) and quadrature (Q) signals from the compressed signal. Two identical A/D converters digitize the I and Q signals. These digitized signals are coherently summed, detected and stored on a disk for further processing. Important system parameters are provided in Table 1 (CARDS, U.S.A.).

Fig. 1.

Block diagram of ICARDS. Blocks shaded at a darker gray level were redesigned and rebuilt. The lightly shaded block was modified for improved performance.

3.2. Receiver

Figure 5 shows the five basic elements of the receiver: (1) front end, (2) sensitivity time control (STC) circuit,(3) driver amplifier,(4) compressor and (5) coherent detector.

The front end consists of a bandpass filler to eliminate spurious signals, and a limiter to protect the receiver form damage by strong signals. A fast GaAs SFDT switch blanks the receiver during the transmission period. The switch is followed by a low-noise amplifier with noise figure of 1.1 dB and gain of 32 dB. The output from the front end is supplied to the STC circuit Two GaAs voltage-variable attenuators and an amplifier are used to provide the STC. The amplifier between the two attenuators compensates for the attenuators'insertion losses.The STC circuit attenuation is variable between 0 and 35dB. An operator can adjust the turn-on and decay rate of the STC circuit with two panel-mounted potentiometers.

Fig. 2.

Block diagram of the radio frequency {RF) section of the ICARDS transmitter. Every block is enclosed as a module in a metal case.

The output from the STC circuit passes to the driver amplifier section dial contains a bandpass filter, two 32 dB gain amplifiers and a step attenuator. The filter eliminates any spurious signals introduced by the STC circuit. The step attenuator can be adjusted to provide attenuation between 0 and 63dB. The two amplifiers and attenuator provide a variable-gain driver amplifier. Using front panel controls, the driver amplifier gain can be set between 0 and 63dB. The driver amplifier output goes to the compressor section that consists of the SAW compressor with a 37 dB loss and a post-amplifier with a gain of 45dB to compensate for the SAW device insertion loss. The compressor output goes to the coherent detector that consists of I and Q, detectors, amplifiers and low-pass filters. The amplifiers are used to obtain conversion gain of l0dB from the I and Q detector sections. The low-pass filters, with cut-off frequencies of 9375MHz, eliminate aliasing introduced by digitization. The total maximum receiver gain, with minimal STC attenuation and step attenuator setting, is about l00dB. The radar is normally operated with a gain of 87dB when STC is at its minimum attenuation setting.

Fig. 3.

Transmitter waveform generation process.

3.3. Timing and control

Radar timing and control signals are derived from the pulse-repetition frequency (PRF) signal. This is generated from the 37.5 MHz TTL signal using a programmable divider. The divider can be programmed to provide any one of the four discrete frequencies: 1100, 2300,4600 and 9200 Hz.

Fig. 4.

Oscilloscope display of the RF-chirp waveform, showing a dean 1.6μs pulse width with no amplitude modulation.

We use digital delay generator (Analog Devices Inc., 1996) to obtain a short pulse for modulating the carrier signal. Figure 6a shows this circuit along with the waveforms. The PRF signal from the hex inverter (not shown) is supplied lo the trigger input of the delay generator, and the delayed output is supplied to the reset input to produce a narrow output pulse. The width of the output pulse is determined by the reset propagation delay, which is typically about 12 ns. The eight-bit digital word that determines the delay is set to provide minimum delay of the leading edge of the output pulse with respect to the trigger signal.

Figure 6b shows the circuit diagram for generating the transmitter blanking signal and corresponding waveforms. We use an integrated circuit (LS221) consisting of a dual monostable multivibrator with positive-transition or negative-transition trigger inputs. A negative-transition trigger input, derived from the hex inverter (not shown), triggers the multivibrator to generate an output pulse whose width is determined by Rl and Cl and can be adjusted by varying Rl. This output pulse goes to the negative-transition input of the second multivibrator. During the negative transition time the monostable multivibrator is triggered to generate the blanking pulse. The width of the blanking pulse is determined by R2 and C2, and it can be adjusted using R2. In this configuration the first multivibrator controls the start time, and the second multivibrator controls the duration of the blanking signal. While the output pulse width of the LS221 can be varied from a minimum of 35 ns to a maximum of 7 s, in our configuration the blanking-pulse width range can be varied from 700 ns to 4.2μs, and the start time of the blanking pulse can be varied from 460 ns lo 2.77 μs.

Fig. 5.

Block diagram of the RF section of the receive.

3.5. Antennas

The antenna assembly consists of two four-element dipole arrays, one for receive and the other for transmit. The transmit antenna is mounted under the port (left) wing, and the receive antenna under the starboard (right) wing of the P-3 aircraft. Each dipole is constructed with two hollow pipes of 1 in (2.54 cm) diameter and length slightly less than a quarter-wavelength, All four elements are connected to a power divider mounted near the antenna assembly, using equal-length 5 m long cables. The transmit pulse is supplied to the power divider input through 20 m long cables. The total of insertion loss in both cables and the power-divider loss is about 3.5 dB. Figure 8 shows a photograph of the transmit antenna mounted under the wing of the P-3 aircraft. The total computed two-way gain of the antenna, including focusing gain for 3 km ice, is about 27 dB (West and Reference West and DemarestDemarest, 1987). The wings of the P-3 aircraft tilt upward by about 7° which results in an antenna-gain reduction at nadir of about 6dB.

Fig. 6.

Circuit and timing diagrams for the transmitter modulator (a) and blanking-signal generator (b).

Fig. 7.

Black diagram of the digital system.

Fig. 8.

Transmit antenna mounted under the wing of the N426NΛ (NASA P3-B)aircraft.

4. System Evaluation

We evaluated the system performance by measuring its noise characteristics and impulse response. Receiver gain must be set such that noise from the coherent detector is higher than the quantization noise of the A/D converter by a factor equal to the signal-processing gain. This gain is the result of coherent and incoherent integrations. Typically we operate the system to obtain a signal-processing gain of about 30 dB, which can be achieved with various combinations of coherent and incoherent integrations. Normally we use coherent integration factors of 128 or 256.

?o measure the noise performance, we terminated the receiver input with a 50 Ω load, set the receiver gain at about 87 dB, which is the typical operating gain, and digitized the I and Q, signals. Figure 9 shows the results of this test. Between samples 50 and 900 the output noise is reduced to about one A/D count after 1024 coherent integrations. The return below sample No. 50 is caused by transi ents introduced by the receiver-blanking switch. During normal operation over glacial ice this part of the waveform is not digitized, because the start time for digitization is set at a much later time.

Fig. 9.

Test results with receiver input terminated with 50Ω load and the receiver gain set at about 87 dB.

We installed the radar on a NASA P-3 aircraft and flew over a relatively calm ocean off the coast near Wallops Island, Virginia, to measure the impulse response and system sensitivity. We set the end of the receiver-blanking interval and the beginning of the STC interval to about 1 μs. We also set the STC circuit to increase the receiver gain to its maximum value immediately after 1 μs. A variable attenuator was inserted between the receive antenna and receiver input. The aircraft flew at an altitude of about 15000ft (4500 m) to simulate free-space loss associated with a range of about 3.5 km. The variable attenuator was set to provide an insertion loss of 93 dB. Figure 10 shows the measured impulse response, along with an expanded view of the signal to illustrate the side lobes and the noise performance. The echoes before sample No. 100 are internal feed-through signals. The echo at about sample No. 690 is from the ocean surface. The signal-to-noise ratio is about 16dB. The total attenuation to reduce the ocean echo to make it 6 dB higher than system noise is about 103 dB. If we assume that the reflection coefficient at the water-ice interface is about 20dB higher than that for the ice-bedrock interface, the scattering and absorption losses in ice that this system can overcome are about 83 dB.

Fig. 10.

The measured impulse response along with an expanded view of the signal to illustrate side lobes and noise performance during an airborne flight test

4.1. Experimental results

Since 1993. we have collected a large volume of data on ice thickness as part of the NASA Program for Arctic Climate Assessment (PARCA). Reports providing first-order results with radio echogram for each flight-line, along with the ice thicknesses derived from these echograms, are available elsewhere (Reference Chuah, Gogineni and WohletzChuah and others, 1996a, b; Legarsky and others, 1997); here we provide a few sample results to demonstrate radar performance.

Figure 11a shows radar data collected along a flight-line passing within 2.6 km of the deep ice-core drill site in northern Greenland (Reference Dahl-JensenDahl-Jensen and others, 1997). The relatively flat bottom topography contributes to the smooth internal layers observed to depths greater than 2.6 km throughout the ice column. Echoes near the surface consist of reflections from dielectric discontinuities in the ice. The very flat echoes are multiple reflections between the aircraft and the surface.

Fig. 11.

Radio echogram (a) and an Λ-scope presentation of signal amplitude vs depth for one sample from the echogram (b) for radar data collected along a flight-line near the deep ice-core drill site in northern Greenland

Figure 11b shows the amplitude of one sample echo wave-form as a function of depth. Near-surface echoes are weak because the STC circuit was set to prevent receiver saturation from the strong primary and multiple echoes of the surface. The STC receiver gain is increased from its minimum value at about zero range to its maximum value for ranges greater than about 1 km. The ice-bottom return occurs at approximately range cell 810 and has a signal-to-noise ratio of about 26 dB. This demonstrates the radar is capable of measuring thicknesses greater than 3 km for cold ice.

Figure 12 shows radar data collected over the southern Greenland ice sheet during the 1997 field season. From the figure we can see several layers both near the surface and at depths of 1km and beyond. The variation in intensity between the surface and about 400 m depth is due to the blanking switch that protects the receiver from the strong surface echo. We also see significant variation in the bedrock elevation underlying the ice sheet; however, beyond about 1.8- 2 km the attenuation exceeds the sensitivity of the radar, and the bottom echo is lost. Many other interesting features appear in this image that are not present in the less complex area in northern Greenland. We see this in southern Greenland because the ice temperature is higher than in northern and central Greenland and consequently the absorptive loss is increased. We are exploring techniques to Improve the radar sensitivity in order to increase our penetration depth in these regions.

Fig. 12.

Radio eckogram data collected over the southern Greenland ice sheet during 1997 field season.

Fig. 13.

Radio echogram plot (a) and radar echo intensity (b) showing the surface and bedrock topography along Petermann Gletscher: Inset in (a ) illustrates the surface-elevation estimate based on isostatically compensated ice thicknesses.

Figure 13 shows surface and basal topography (top) and radar-echo intensity (bottom) along Petermann Gletscher in northwest Greenland. The calving front is located on the rightmost edge of the figure where the ice thickness is only about 60m. The ice thickness 125km upstream of the calving margin is over 1100in and the glacier rests on bedrock about 100 m below sea level. Moving seaward, the ice sheet thins and the bedrock elevation becomes increasingly negative. A sharp increase in echo intensity is associated with a deep trough located about 95 km upstream of the calving margin. We believe this may signify the presence of a small subglacial lake. The echo intensity begins to increase systematically about 8O km from the calving margin. We believe this represents a transition in basal properties as the grounding line of the glacier is approached. Based on echo strength, the location of bottom crevasses and a comparison of surface elevation, with the surface-elevation estimate based on isostatically compensated ice thicknesses (see Fig. 13a, inset), we find that the grounding line occurs about 70 km upstream of the calving margin and near the location determined using interferometric spaceborne SAR techniques (Reference RignotRignot, 1996).

5. Conclusions

We redesigned and rebuilt the University of Kansas radar depth sounder to make it fully functional. The system obtains high sensitivity for measuring ice thickness through signal processing and pulse compression. Results show that it provides excellent ice-thickness information over the cold ice, ice margins and outlet glaciers. We plan to collect additional ice-thickness data over the Greenland ice sheet during 1998 and 1999 to augment and supplement existing data. The radar is designed to display radar echograms for the operators in real time. The system is ready for routine data collection.

We plan further improvements to system performance by steering the antenna beam to eliminate gain reduction caused by the slanted wings. We also arc developing a new digital signal processing system to incorporate 12-bit A/D converters to increase dynamic range and store data in a coherent mode for implementing more coherent processing.