CryoSat User Workshop 2

Satellite altimetry has been used to derive information about sea ice thickness in the Arctic already for several decades. In this connection the shape of the radar signal is used to identify leads, the open water between ice floes. Analysis of airborne altimeter data reveals that the waveform shape can additionally be used to identify different sea ice types. In this study we analyze signal waveforms from ESA’s CryoSat-2 satellite, to test the possibility of sea ice classification based on radar altimeter waveforms on an Arctic wide scale. We define six parameters to account for the difference in the shape of the radar waveforms obtained over First- and Multi-Year-Ice and find significant differences for several of these parameters. The Pulse Peakiness, Stack Standard Deviation and Leading Edge Width show the largest difference. These waveform parameters can thus be used to classify First- and Multi-Year-Ice over large areas of the Arctic Ocean. However, analyzing the spatial distribution we find some discrepancies compared to other retrievals of sea ice type. CryoSat waveform parameters have values typical for Multi-Year-Ice over large areas classified as First-Year-Ice. These areas are co-located with strong gradients in drift speed, indicating, that the radar signal is mainly sensitive to surface roughness. Potentially this information could be used to reduce biases in the freeboard retrievals and to improve estimates of sea ice thickness.

The main payload of CryoSat is a Ku-band pulsewidth limited radar altimeter, called SIRAL (Synthetic interferometric radar altimeter). When commanded in SARIn (synthetic aperture radar interferometry) mode, through coherent along-track processing of the returns received from two antennas, the interferometric phase related to the first arrival of the echo is used to retrieve the angle of arrival of the scattering in the across-track direction. In fact, the across-track echo direction can be derived by exploiting the precise knowledge of the baseline vector (i.e. the vector between the two antennas centers of phase) and simple geometry.

In order to monitor the performance of the CryoSat interferometer along the mission, in orbit calibration campaigns following the approach described in [1] have been periodically performed about once a year. This abstract is aimed at presenting our analysis of the accuracy (i.e. the closeness of the measurement to the true value) and of the precision (i.e. the closeness of agreement among a set of measurements) of the angle of arrival measured on the data acquired during the periodical calibration campaigns.

The end-to-end calibration strategy for the CryoSat interferometer, described in [1], uses the ocean surface as the known external target. In fact, the interferometer can be used to determine the across-track slope of the overflown surface and the slope of the ocean surface can be considered as known starting from the geoid. Denoting by β the across-track slope of the ocean and assuming that the knowledge error of the geoid slope is negligibly small, β can be compared with the across-track slope derived from CryoSat SARin Level1b products which results in β’=η(θ-χ) where η is a geometric factor, θ is the angle of earliest arrival measured by the CryoSat interferometer and χ is the baseline roll angle. By comparison of the expected across-track slope β and the measured across-track slope β’, the accuracy and the precision of the angle of arrival θ measured by the CryoSat interferometer can be assessed. It is worth noticing here that the accuracy of the measured across-track slope β’ is also dependent on the knowledge of the baseline orientation. As discussed in [1], starting from β and β’ a calibration function F(θ) for the angle of arrival can be computed.

In our analysis, the accuracy and the precision of the CryoSat interferometer have been assessed starting from the residual errors on the measured angle of arrival, that have been obtained after compensation of the calibration function F(θ). From this analysis it has been verified that the performance of the CryoSat interferometer are stable from the beginning of the mission. In particular:

The median precision of the angle of arrival has been verified to be about 30 mrad at 1Hz and this residual error has been mainly addressed to the phase noise on the SARin Level1b products, that in turns depends only on the instrument and on Level1 processing. This is in line with the expected performance of the SIRAL.

On the other hand, the median accuracy of the angle of arrival has been verified to be highly dependent on the accuracy of the baseline roll angle and on the different acquisitions (between few mrad and 80 mrad). This suggests that the accuracy of the measured angle of arrival can be mainly addressed to the knowledge of the baseline orientation, that in turns depends on the star tracker accuracy, on the conversion from quaternions to off-nadir mis-pointing angles and on calibration function F(θ).

Moreover, recalling that the calibration function F(θ) for the angle of arrival allows estimating the roll bias (the rotation error between the star tracker frame and the interferometer baseline), as a further result the roll bias of CryoSat along the mission has been measured. 

[1] Galin, N.; Wingham, D.J.; Cullen, R.; Fornari, M.; Smith, W.H.F.; Abdalla, S., "Calibration of the CryoSat-2 Interferometer and Measurement of Across-Track Ocean Slope," in Geoscience and Remote Sensing, IEEE Transactions on , vol.51, no.1, pp.57-72, Jan. 2013

A key issue that often hinders the establishment of accurate metrics for retracker performance over real-world surfaces, is the lack of suitable comparison datasets. The CryoVal Sea-Ice project has carefully selected a set of CryoSat-2 tracks for which there exist high-quality reference datasets to compare against. Here, we present a set of results obtained by applying a library of retracking algorithms to a set of those CryoSat2 SAR- and SARin-mode tracks over Arctic sea-ice. The quality of the results is compared relatively between retrackers, and absolutely against validated, in-situ and airborne measurements of height, and sea-ice freeboard. Recommendations are then made for the most effective algorithms for the retrieval of height from both ocean and sea-ice echoes.

The CryoSat-2 baseline-C data make it possible to detect some leads and polynyas and in some cases to measure freeboard. Nevertheless the processor still faces difficult cases. Indeed sea-ice areas are subject to a high spatial diversity which leads to a broad range of radar signatures. The complexity coming from the spatial diversity is emphasized by the strong temporal variability related to seasonal trends, local surface currents and very frequent extreme events. Some of the CryoSat Ice Processor limitations may be due to a poor understanding of the physics behind the SAR/SARin Level-1A to Level-2 products over the Sea-Ice domain.

Dedicated science oriented tools therefore need to be put in place to make best profit of both in situ data and imagery to thoroughly understand the signatures within the altimeter signals. Such insights could help to improve retrackers and refine the CryoSat Delay Doppler Processor (range zero-padding, burst weighting, echo discrimination, handling of side-lobe effects...).

Our work establishes a conceptual framework that eases and partially automates the analysis of CryoSat-2 measurements through the use of Sentinel-1 SAR imagery and other satellite missions over sea ice areas. The tool addresses several critical aspects: first the selection of closely co-dated satellite data for the area of interest (the results are very sensitive to this critical step which may prevent a full automation of the whole process), second the adequate positioning of the altimeter footprint within the image and third the identification of ice/snow classes that serves to improve the understanding of the physics in the signals and to confirm the detection of leads and polynyas.

The conceptual framework and its early implementation, that will be presented here, exploit SAR images segmentation and classification tools, CryoSat ground projected footprints and visual interfaces to permit a deep inspection of Individual Echoes, Delay Doppler Maps, SAR/SARin waveforms as well as retracker outputs, and help the consolidation of Sentinel-1 radiometry and polarimetry based classes.

Preliminary results are given for specific scenarios. In particular the Range Integrated Power parameters (Kurtosis, Skewness Standard Deviation, Stack Centre and mean waveforms) are analysed in conjunction with the SAR imagery.

The Baseline C update to the CryoSat2 processing chains brought improved algorithms, auxiliary models, and product formats. Here, the QA results for continental ice-sheets and Arctic sea-ice are compared with Baseline B, to illustrate the improvements these changes brought to product performance. The performance of the new land-ice retracker is compared against the existing ocean retracker available in Baseline B. The effects of the updated slope model on height retrievals over Greenland and Antarctica are investigated. Similarly, over Arctic sea-ice, the improvements brought by the new retracker for diffuse echoes in SAR mode are demonstrated. Finally, we investigate the evidence for any residual long-term drifts in the measurement performance.

We are experimenting with techniques and applications for fully focused coherent processing of radar altimeter echoes. Our experiments to date use CryoSat2 FBR SAR data, but our methods could also be used with similar data from Sentinel-3 or Sentinel-6.

From Geos-3 (1975) through Jason-3 (2015?), conventional altimeters employ incoherent processing, resulting in measurement footprint diameters on the order of 2 km [doi:10.1175/1520-0426(1989)006<0407:PCASLT>2.0.CO;2]. The Delay/Doppler [doi:10.1109/36.718861] or multi-looked SAR [doi:10.1016/j.asr.2005.07.027] algorithm used by the CryoSat, Sentinel-3 and Sentinel-6 missions employs coherent processing of only 64 echoes at a time (about 3.5 milliseconds of flight), narrowing the footprint in the direction along the track to about 300 m. This process should be called Doppler beam sharpening, not focusing, because 300 m is much wider than a Fresnel zone [doi:10.1109/LGRS.2011.2170052]. In contrast, our technique employs coherent processing throughout the time that a target is in view of the altimeter’s range window (more than 2 seconds) and narrows the measurement footprint to 1 meter along track. If the viewed scene contains a small, bright target, such as a transponder or a small water body on land or in ice, then we are able to fully focus on that target, and we can measure its extent in the along-track direction. If the viewed scene contains many incoherent reflectors, such as the open ocean, our technique yields speckle noise waveforms having a decorrelation distance of 1 meter along track. These can be incoherently averaged along track to optimize the trade-off between coherent and incoherent processing. This results in advantages over the CryoSat, Sentinel-3 and Sentinel-6 multi-looked SAR waveforms: our waveform increases the effective number of looks, resulting in more precise retrievals of geophysical parameters such as sea level, wave height, and backscatter. Our waveform also has a closed-form analytic solution for the expected power from a rough surface, so we can use standard retracking algorithms, with no need for numerical convolutions.

A fully focused coherent calculation requires that we account for the elapsed time between echoes with accuracy much better than one microsecond. The datation jitter in the CryoSat FBR SAR data product can be as much as 13 microseconds, and we have to correct for this. We also have to account for phase curvature within each burst. In contrast, the along-track Fourier transform used for the unfocused (300 m) calculation assumes that phase changes are only linear within one burst. Making a focused calculation over the Svalbard or Crete transponder shows that the signal to noise (SNR) is 31.6 dB when phase curvature is accounted for, but only 29.5 dB when phase variations are assumed to be only linear. After accounting for phase curvature, the RMS phase error in individual transponder echoes is 26 milli-radians, implying sensitivity to antenna motions of 45 micrometers. These results are at Ku band, so a Ka-band instrument such as KaRIN (SWOT) will need antenna positioning and datation more accurate by a factor of 3 to achieve the same SNR.

Fully focused coherent processing of data over transponders can yield estimates of datation bias and platform pitch bias. We hope that we may be able to use the process to distinguish between coherent and incoherent targets in a heterogeneous scene, such as in the coastal zone or at leads in sea ice.