Convolutional neural network training dataset and results of AWI-ICENet1 retracker


This data set include the simulated and corresponding reference data in binary format used for the training of the AWI-ICENet1 retracker algorithm, which is a convolutional neural network (CNN). The simulation is carried out at 1000 randomly selected locations spread over the Antarctic ice sheet. At each location a reference waveform is simulated based on the local topography. This waveform is modulated using 95 different attenuation rates ranging from 1 to 20 dB (step width 0.2 dB). Finally 45 noisy waveforms are generated from each modulated waveform. Therefore, at each location 95*40=3800 waveforms are generated. The simulated data consist of a total of 3.8 Mio waveforms. The AWI-ICENet1 retracker is applied to the full CryoSat-2 time series. Monthly elevation change is estimated for Greenland and Antarctica and compared to estimates derived from ICESat-2. Here, we provide raster data sets of the elevation change, rates of elevation change and additional parameters such as correlation with backscatter and leading edge width as netcdf files. The CryoSat-2 elevation change products of AWI-ICENet1 are estimated for the time periods: 2011-2022, 2019-2022 for Greenland and Antarctica and 2011-2014 for Greenland only and are provided as monthly gridded georeferenced netcdf files. As input to the AWI processing we used the reprocessed Baseline_E Level_1B waveform product provided by ESA. Additional we provide the ICESat-2 elevation change products for 2019-2022 for Greenland and Antarctica as monthly gridded georeferenced netcdf files which are based on the same processing strategy. As input to our processing we used the ATL06.006 ICESat-2 data product provided by NASA.

We provide for each of the 1000 locations 3 different binary files.The first file named as e.g. ATT_ARR_000958.bin contains the 95 attenuation settings. This is a double precision file of size 760 Byte.The second file named as e.g REF_ARR_000958.bin contains the 95 reference retracking positions. This is a double precision file of size 760 Byte.The third file named as e.g. SIM_WF_NOISE_000958.bin contains the 3800 waveforms. A single waveform consist of 128 samples. This is a floating-point precision file of size 1945600 Byte. In addition a file named RANDOM_POINT_LIST_4326.dat is provided which contains the 1000 geographic locations (Id/Longitude/Latitude). This is a double-precision binary file of size 24000 Byte.The netcdf files are self explaining and include monthly raster data sets of various variables.The projection is EPSG:3031 for Antarctica and EPSG:3413 for Greenland. Pixel resolution is 5km.

Related Identifier References
Metadata Access
Creator Helm, Veit ORCID logo
Publisher PANGAEA
Publication Year 2024
Funding Reference German Research Foundation Crossref Funder ID 442929109 Advancing ice altimetry analysis to constrain the Antarctic Ice Sheet's response to climate change; German Research Foundation Crossref Funder ID 5472008 Priority Programme 1158 Antarctic Research with Comparable Investigations in Arctic Sea Ice Areas; Helmholtz Association of German Research Centres Crossref Funder ID ZT-I-0016 Artificial Intelligence for Cold Regions (AI-CORE); Helmholtz Association of German Research Centres Crossref Funder ID ZT-I-PF-5-9 DeepSAR
Rights Creative Commons Attribution 4.0 International; Data access is restricted (moratorium, sensitive data, license constraints);
OpenAccess false
Resource Type Dataset
Format text/tab-separated-values
Size 27 data points
Discipline Earth System Research
Spatial Coverage (-12.000W, -90.000S, 180.000E, 84.000N); Antarctica; Greenland
Temporal Coverage Begin 2011-01-01T00:00:00Z
Temporal Coverage End 2022-12-31T00:00:00Z