Radar scattering from snow facies of the Greenland ice sheet: results from the AIRSAR 1991 campaign

In June 1991, the NASA/Jet Propulsion Laboratory airborne SAR (AIRSAR) collected the first calibrated multi-channel SAR observations of the Greenland ice sheet. Large changes in radar scattering are detected across different melting zones. In the dry-snow zone, Rayleigh scattering from small snow grains dominates at C-band. In the soaked-snow zone, surface scattering dominates, and an inversion technique was developed to estimate the dielectric constant of the snow. The radar properties of the percolation zone are in contrast unique among terrestrial surfaces, but resemble those from the icy Galilean satellites. The scatterers responsible for the percolation zone unusual echoes are the massive ice bodies generated by summer melt in the cold, dry, porous firn. An inversion model is developed for estimating the volume of melt-water ice retained each summer in the percolation zone from multi-channel SAR data. The results could improve current estimates of the mass balance of Greenland, and could help monitor spatial and temporal changes in the strength of summer melt in Greenland with a sensitivity greater than that provided by altimeters.<<ETX>>


INTRODUCTION
An understanding of the geophysical characteristics of polar icr sheets and snow masses and of their time evolution with changing environmental conditions is essential t o climate modeling. Tlir first examples of SAR observations of the Greenland ice sheet [l-41 have shown great promise for monitoring applications of the changing hydrology of polar ice sheets as a r~s u l t of atmospheric warming. Here, we report result from the first calibratrd tniiltichannel S A R observations of Greenland collected by AIRSAIL in June 1991. The goals of the Experiment were to record radar echoes from different melting zones of tlir Greenlalid ice sheet a.nd then t o relate thrse radar ohservations t o the snow a l~d fir11 physical properties. Fig. I shows the flight track of AlltSAlC; the locat.ion of 3 ice ramps where glaciologists recorded siiow stratigraphy, grain size, density, and temperature; and the 4 melting zones characterizing different degrees of melting of tlie ice sheet in t h e summer [SI.
AIRSAR operates a t C-, I,., and P~ band frequencies (5.F, 24, and 68-cm wavelengths), and records the complrte scatter^ ing matrix of each resolution element a t each frequency. The radar echoes are internally calibrated. T h e calibration was crosschecked using t h e radar responses of trihedral corner reflectors deployed at Crawford Point. prior t o flight. T h e rrsrilts iodicate a calibration acciirary better tha.n 1-2 d B for 8 2 30". For 8 5 30°, the C-hand radar cross-sections are underestiniatcd by a few dBs, whereas t h e polarimetric characteristics seem correct. Radiometric calibration errors inay be rauscd by processing artifacts due t o the shorter integration time, or t o errors in the predicted response from corner reflectors wlrich point more than 1.5" away from tlie radar lobe direction The system noisr level, dominated by thermal noise from the radar receiver, varies w i t h t h e receiver's gain sett,ings, frequency, polarization, and also with the range distance after range and azimuth compressions. The SNR was found to be large ovpr tlie percolation facies, but poor in the soaked-and dry-snow facies at large intidence angles. l'lie apparent increasr i n m g V for 0 2 GOo in Fig. 2 is due t o tliermal noise and not to wattering.

ANALYSIS OF RADAR SCATTERING
We examined SAR srenes collected i n 3 of the 4 melting snow zones of Fig. 1. The most striking observations are from tlir percolation zoiies, but interesting results are also obtainrrl in the dry-and soaked-snriw zones.
D r y -s n o w z o n e : Radar returns are low a t all 3 frequencies, and mostly diffuse at C b a n d , whicl, is consistent with volume scattering from dry snow grains. Lloilel prrrlictioiis basrd on Rayleigli scattering from a half-space of dry-snow agree with the AIRSAIL nie;~surrmrrrts a t C~b a n d (Fig. 2), but underrstiinates the radar returns at L-and P-band. The longer wavelrngths are probably sensitive to more deeply buried densrr snow layers of larger snow grains, with a reduced pore space. Additional scenes will he processed to detect eventual spatial variations i n snow properties within the dry snow zoiie.
S o a k e d -s n o w zone: Vidro imagery and 3.5 Inm photos taken during the A1RSAl1 flight show that areas of low harksratter in the C-band imagery (Fig. 3) I i a v~ also low radiances a t visiblr optical frequenrirs and likely correspond to arras of wet siiow accumulating in local surface depressions and surrounded by drier snow patches. C-band radar returns are low in wet snow because the snow layer i s smooth and highly reflective, and a r r high in dry snow becaiisc of volume scattering from snow grains. Spa: tial variations in radar backscatter a t L-band (not s l i o~~s ) arc different, a result uf the deeprr prnetration of the radar sigiials, revealing subsiirfacr features of Iiigh~r radar reflectivity among the wet snow patrhrs d e t e c t d at C~b a n d . V b a n d data (not shown) reveal Iiuried rracks and rrevitsses that are not visible at C-band.
We used the invrrsion teclinique of van Zyl e t al.
[i] to infer the dielectric ronstant cr and tlie rms liriglit h of the reflecting surface from the SAR scene of Fig. 3. The algorithm is valid for natural terrain dominated by surfare scattering, and uses 1111and VVpolarization at one frrquenry, preferably one for whicli X >> h. T h e results a t C-band are not satisfactory i n dry snow areas because volume scattering from s m w grains is falsrly i l lterpreted as rcsrilting from rough surface scattering. l h r c o r r e~ sponding pixt:ls ran be separated rising obi , > -28 dll. Pixrls for which 11 2 5 cm, which corresponds to snow-free bedrock, ari. also removed brcaiise the estimation of c, is no1 reliahlr. In the remaining areas, cr and h have consistent valurs across range (the inversion techniqiie is robust with incidence anglr variations): h = 0.8 f 0 . 3 cm, a smooth surface, and cT = 3.0 10.5, snow with a. liquid water content aboiit 1052% ( In addition, t h r inwrsion results reveal patrlies of standing water or thawed lakrs (er > 20), as verified from the video imagery.
These water forniations woiild be difficult to identify from C~ band VV or 1 ,~b a n d 1111 data alone. illiistrating tlie iiitrrest of polarinittric information. At I.-Iii~iid, the results are wriiplicated by t h e dreper priirtratiori of the signal. 6 , of "wet snow areas" is Y 3, hiit 11 F, cm is largrr than at, C-l,a.nd, suggrstiiig t1ia.t L b a n d signals p r n r t r a l e the shallow layrr of wrt snow and ill-teract with a rough interface of glacier ice ( E , = 3.2 (91). Thawed lakes appear more clearly than a t C-band.

T h e inversion results show t h a t SAR conld help m a p SIIOW
wetness in the Greenland soaked-snow zone, tliereby providipg valuable information for estimating siirfare fluxes and fresh water production over t h e ice sheet. C-band VV from ERS-1 c o d d separate wet snow from dry snow, but the addition of a t least O I I P polarizat.ion (e.g. (:-band 1111 from RADARSAT) rould perniit' a more reliable mapping of snow wetness.
Percolation facies: Umsually strong radar reflertivities and large circular (pc = o ;~~/ u~, , , wliere R a n d 1 , mean right ai111 left circular polarizations) and linear ( p~ = n f r v / o t H ) polarization ratios are recorded at C-and L~ band from the percolatioii facies. To t h e hest of o u r knowledge, no other terrestrial surface shows similar radar properties, and the only o t h r r known olrjects with similar exotic radar properties are the icy Galilean satellites of Jupiter [10][11]. T h e radar signatures from the percolatioil zone show a different behavior a t 68-cm wavelength (Fig. 4) and at small inridence angles, hut this is not surprising since the subsurface structures responsible for the unusual radar echoes from Grwnland and from the iry satellites probably are radically different. Nevertheless, t h e Greenland's percolation zone provides a uniquely arcessible. natural laborat.ory for studying exotic radar processes in a geologiral context. and for improving our understanding of radar scattering from distant icy objects.
Years ago, Zwally [12] suggested t h a t the low emission properties of the percolation zone could be diir t o volume sratlkring from ice features buried in the firn and created by summer rnrlt. Swift e t al.
[l] also noted iinusiial radar reflectivities from the perrolatioii facies, and suggestrd t h a t radars could estimate the density of these ire features. Rerent surfare-based radar observations a t Crawford h i n t [In] at 5.4 and 2.2 cni furtlrcr indicate t h a t most of the srattering takes place in the most recent annual laver of buried ice bodies, and t h a t surface scattering from the t,op 20 cm of the ice sheet dominates at small inridenre a.ngles. T h e exact scattering mechanism rrsponsible for t,he ,inusual echovs is still unknown, and littlr information is about t h e spatial distrihution and geometric characteristics of i r r lenses, layers, and pipes 1.5, 14-IF]. Ra,yleigli srattering docs not explain the radar propwties of t,lie percolation facies beca.iisc tlie radar cross-sections do not decrease as ,A4, aiid eveii largr snow Cohprent backscatter is capable of interpreting the radar p r o p erties from the icy satellites [I;] and is consistent wit.11 their geology and formation history [IO]. Coherent harkscatter occurs when electromagnetic waves traveling along timr-revrrsed paths constructively i n t r r a r t in t h r harksrattering direction, yielding enhanced radar reflectivities, and preserving the haiidediiess of circular-polarized signals (as opposrd t o specular rcflcctioiis). Coherent hackscatt.er only requires a low-loss nirdiiini, i.e. long photon pathlengths. High radar reflectivities arid circular polarization ratios greater than unity occur when the scatterers in the lossy medium are larger in size tliaii t h r radar wavelciigtli [18], and when t h e dielectric constrast is low [19]. Given typical sizes of ice layers and ice pipes [5, [14][15][16], and the relative tra.nsparency of dry, cold snow at microwave frequencies, coherent hackscatter from massive ice inclusions created by sumnier melt events rould explain t h e radar echoes from Greenland. Coherent backscatter however predicts 1191 t h a t pc should decrease with iucreasing 8, whereas AIRSAR shows thp opposite trend. This rlisrrepancy could be due t o strong backreflertion from wind crusts present in the t o p 20 cm of the ice sheet (131 which roiild yield small valurs of pc and p~ a t small incidenre angles. Since ice layers and ice pipes are anisotropic scatterers larger in size than t h e wavelength, we examined the high frequency solution of simple objects to determine whether unusual radar echoes can be predicted and better interpreted. We used dielectric cyliaders because a n exact analytical solution exists for computing their radar cross-sections [20]. In principle, highly depolarized signals could result from 180' phase shifts between IIH and VV, i.e. internal reflections within the icy cylinders. Although ice layers and ice pipes are not cylinders, it is a first-order approximation t h a t incorporates their dielectric properties, differences in size along perpendicular directions, and principal diniensioiis larger than t h e radar wavelength. From the numerical soliition for one cylinder, we generated the response from a distribution of cylinders by applying a rotation matrix of uniform dist,ribution in angle t o t h e scattering matrix from one cylinder, and incoherently added t h e returns from different cylinders. When the dielectric contrast between the cylinders aiid the matrix is low, for a large number ofa/X values, we found t h a t &HVI/* = 180", ILL > 0.5 and f i~ > 1 (Fig. 5 ) . By averaging the radar returns from a distribution of cylinders about 20-30 cm in size, we 01)tain /LL > 0.37, pc > 1.4, and radar reflectivities close t o 0 dll as observed for Greenland. T h e advantage of this modeling allproach compared t o the coherent backscatter theory is t o provide a more readily understood interpretation of the radar echoes and a model t h a t is easily invertible.

CONCLUSIONS
Monitoring t h e percolation facies using SAR could be of considerable importance. T h e Greenland percolation zone extends over a large range of latitudes and provides one of the most significant land bodies t h a t rould reveal climatic and hydrologic trends i l l t h e Arctic. T h e large contrast in radar backscatter bctween the percolation zone and t h e surrounding snow zones, and the fine spatial resolution of SARs should permit the fine monitoring of horizontal shifts in the boundaries of the perrola.tioii zone t h t probably no other instrument can detect. T h e percolation zone defines critical hydrological limits where melt-water re-freezes i n place instead of being re-distributed in the ocean waters as in the soaked-snow and ablation zones. Measuring the volume of melt.
water stored in the percolation zone each summer from SAR d a t a is essential t o improve current estimates of the mass balance of Greenland [16]. Our results are encouraging t h a t multi-cha.nne1 SAR instruments could estimate this ice-water volume. Furthermore, monitoring of t h e percolation facies with SAR could pro^ vide a more sensitive detection of spatial and temporal changes in t h e strength summer melting than t h a t provided by inonitoring changes in surface elevation measured by altimeters bccaiise melting in the dry snow zone will proba.hly first increase t h r arral extent of the percolation zone [lS].

A C K N O W L E D G E M E N T S This work is supported hy Ilr: R.
Thomas, Polar Researrh Program, NASA IIQ, and was carried out at.
the Jet Propulsion Laboratory, California Institute of Technolgy, undrr contract with thc National Aeronautics and Space Adrni~iirtrntioii.
13, 637-645, 1992.  Glacrol. 18, 195-215, 1977. [la] Jezek, K . , and P. Gogineni, Microwave remote sensing ofthe Greenland ice sheet, IEEE Geosr. and Rem. Sens. Soc. Newsletter, Dec. 1992. 114 Figure 2. Radar backscatter uo measured by AIRSAR a t C-band, a t 1111-, HV-, and VV-polarization vs B in the dry snow zone near the GISP 11 site. Model predictions (dotted lines) for scattering from a half-space of dry snow using the radar backscatter model of J.C. Shi, UCSB. The snow temperature is -lO°C, the ice fraction is 0.349 (snow density is 0.32g/cm3), the grain diameter is 0.2 f0.05 mm [6] and the air-snow interfare has a r m s height of 0.78 cm and a correlation length of 13.8 cm. is proportional to the square of ing rarely occiirs in t h e dry-snow zone; forms massive, buried, the length oft.lic cylinders, and to the density of cylinders; whereas ,rr and solid-ice inclusions in t h e perrolation zonr; saturates the snow pc only depend on a/A. The model results indicate that the ice-bodies are with liquid water in t h e soakrd snow-zone; and removes the sea-about 20-30 cm in diameter. I m i n length, with a density of 5 cylinders sonal snow cover and ablates the glacier ice in the ahlation zone. per square meter.