Contributors: Lin Gilbert (University Leeds). Sebastian Bjerregaard Simonsen (Technical University of Denmark), Jan Wuite (ENVEO)
Issued by: University of Leeds / Lin Gilbert
Issued Date: 31/05/2020
Ref: C3S_312b_Lot4.D2.IS.1_v2.0_202001_Product_Quality_Assurance_Document_v1.0
Official reference number service contract: 2018/C3S_312b_Lot4_EODC/SC2
Note: This document provides the following three deliverables:
D2.IS.1-v2.0 Product Quality Assurance Document - Ice Velocity
D2.IS.3-v2.0 Product Quality Assurance Document – Gravimetric Mass Balance
D2.IS.5-v2.0 Product Quality Assurance Document – Surface Elevation Change
History of modifications
List of datasets covered by this document
Related documents
Acronyms
Scope of the document
This document is the Product Quality Assurance Document for the Copernicus Ice Sheets and Ice Shelves service. It describes the validated datasets and the methods used for validation.
Executive summary
The service addresses three essential climate variables (ECVs) by providing four separate products.
- Ice velocity is given for Greenland in product D3.IS.4
- Gravimetric mass balance is given for Greenland and Antarctica in product D3.IS.5
- Surface elevation change is given for
- Antarctica in product D3.IS.6.1
- Greenland in product D3.IS.6.2
We provide the validation methods and results from the CDR v2 for each dataset produced by the service. Validation is not performed on the brokered dataset, gravimetric mass balance, as it has been performed previously and documented. Relevant documentation is available at:
- AIS: http://esa-icesheets-antarctica-cci.org/index.php?q=documents
- GrIS: http://esa-icesheets-greenland-cci.org/index.php?q=documents
1. Validated products
1.1. Greenland ice sheet velocity – D3.IS.4
The ice velocity (IV) product assessment referred to in this document concerns the annually-averaged IV maps of Greenland derived from Sentinel-1 SAR data acquired from 2017-10-01 to 2018-09-30 (reprocessed from CDR v1) and 2018-10-01 to 2019-09-30 (both combined in CDR v2) (the latter is depicted in Figure 1). surface velocity is derived applying advanced iterative feature tracking techniques. The ice velocity map is annually averaged and provided at 500m grid spacing in North Polar Stereographic projection (EPSG: 3413). The horizontal velocity is provided in true meters per day, towards easting (vx) and northing (vy) direction of the grid, and the vertical displacement (vz), is derived from a digital elevation model (TanDEM-X 90m DEM; Rizolli et al., 2017). The product is provided as a NetCDF file with the velocity components: vx, vy, vz and vv (magnitude of the horizontal components), along with maps showing the valid pixel count and uncertainty (std). For further details, see the Harmonization Tables related document. For the product intercomparisons discussed in this document, we consider both the annually averaged map and the individual (6/12-day repeat) ice velocity maps on which the annual map is based. The individual maps are not provided as a product in C3S but are included here as an extra quality assurance.
Figure 1: C3S ice velocity map of the Greenland Ice Sheet based on Sentinel-1 data acquired from Oct 2018 to September 2019.
1.2. Gravimetric mass balance – D3.IS.5
The GMB products are brokered from the two ice-sheet CCI products and relevant documentation is available at:
- AIS: http://esa-icesheets-antarctica-cci.org/index.php?q=documents
- GrIS: http://esa-icesheets-greenland-cci.org/index.php?q=documents
1.3. Surface elevation change, Antarctica – D3.IS.6.1
The product is a netCDF file containing monthly gridded maps of the rate of surface elevation change over the Antarctic region, including ice sheet, ice shelves, ice rises and islands. Rates are given in m/year. The grid is based on a polar stereographic projection with central meridian 0E, standard parallel 71S, no false northing or false easting, using ellipsoid WGS84 (see National Geospatial Intelligence Agency, 2004). Its resolution is 25km by 25km.
The rates of change in each map are derived from the 5-year period centred on that map's timestamp. The timestamps are one month apart. The initial climate data record (CDR) contains maps centred on November 1994 to April 2016, and each monthly intermediate CDR (ICDR) adds one map. The ICDRS are accumulative, containing all previous data as well as the latest monthly map. The v2 CDRs and ICDRs carry on from v1, and incorporate reprocessed v1 data. As well as maps, timestamps and grid definitions, the product includes uncertainties on each rate, based on the combination of three uncertainty sources – input data, cross-calibration between satellites, and derivation of the rate of change from modelling. Flags are given for where rates have been produced, for geographical surface type, and for geographical regions of high slope. An example map is shown below in Figure 2 .
Figure 2: Example surface elevation change map from product D3.IS.6.1
1.4. Surface elevation change, Greenland – D3.IS.6.2
The surface elevation change product is available as a single NetCDF-file containing monthly gridded maps of the rate of surface elevation change over the Greenland ice sheet. Following glaciological conventions, the rates of elevation change are given in m/year. The data are posted in a polar stereographic projection (EPSG:3413), with central meridian 45W, standard parallel 70N, no false northing or false easting, using ellipsoid WGS84 (see National Geospatial Intelligence Agency, 2004) and a resolution is 25 km by 25km.
The rates of change in each map are derived from the 3-year or 5-year period centred on that map's timestamp. The timestamps are one month apart. The difference in time period is due to the enhanced capabilities of CryoSat-2 which enables the surface elevation change to be derived for shorter time periods. The initial climate data record (CDR) contains maps centred on November 1994 to April 2018, and each monthly interim CDR (ICDR) adds one map. The ICDRS are accumulative, containing all previous data as well as the latest monthly map. As well as maps, timestamps and grid definitions, the product includes uncertainties on each rate, based on the combination of three uncertainty sources – input data, cross-calibration between satellites, and derivation of the rate of change from modelling. Flags are given for where rates have been produced, for geographical surface type, and for geographical regions of high slope. An example map is shown in Figure 3.
Figure 3: Example of accumulated surface elevation change map produced from product D3.IS.6.2
2. Description of validating datasets
2.1. Greenland ice sheet velocity – D3.IS.4
In absence of ground-based validation data (e.g. GPS measurements), covering the same general time-period as the derived Sentinel-1 (S1) ice velocity (IV) product for Greenland, the product is evaluated against publicly available products covering the same area and time span. Although not a true validation, such an inter-comparison provides a good level of quality assurance, in particular in areas where little change is to be expected.
For product inter-comparison we utilize an independent collection of ice velocity maps derived from higher resolution TerraSAR-X (TSX) data, and covering the Greenland margins, as well as a Greenland wide ice velocity map, based on Sentinel-1 data. These IV maps were produced as part of the NASA 'Making Earth System Data Records for Use in Research Environments' (MEaSUREs) program (Joughin et al., 2011, updated 2018).
The TerraSAR-X-based data set consists of a collection IV maps covering the Greenland margins (including most of the major outlet glaciers) over the time-period from 1 January 2009 to 30 January 2019 (Figure 4, left and middle). The ice velocity is retrieved from repeat pass TerraSAR-X images (1 to 3 cycles) applying a combination of conventional InSAR and speckle tracking techniques (Joughin, 2002). Here we use release V1.2 of the data set, available through the NSIDC data portal (data set ID: NSIDC-0481, Joughin et al., 2011, updated 2019). Data files are delivered in GeoTIFF format at 100m grid spacing in North Polar Stereographic projection (EPSG: 3413). Separate files are provided for the x and y velocity components (in m/yr) along with corresponding (pseudo-)error estimates for both velocity components. The data set is occasionally updated, the latest available data is included for the assessment.
 |
|  |
Figure 4: Sentinel-1 IV mosaic showing locations of MEaSUREs TerraSAR-X derived IV maps used for product inter-comparison in black (left). The IV data sets cover the margins of Greenland and include most of the major outlet glaciers. The red circle indicates the location of C.H. Ostenfeld Glacier shown here as example (middle). Right: shapefile showing ice-free terrain in Greenland and used for the stable terrain test.
The Sentinel-1-based Greenland-wide data set is provided at 200 m and 500 m covering mostly the winter periods. The one used here covers the period 1 September 2017 to 31 May 2018 and is therefore largely overlapping with the 2017-2018 C3S product (CDR v1), spatially as well as temporally. We use version 2 of the data set, available through the NSIDC data portal (data set ID: NSIDC-0478, Joughin et al., 2011, updated 2018). The data set is occasionally updated, once the 2018-2019 map becomes available it will be included in the assessment.
Another method used here, for quality assessment of the ice velocity product, is the analysis of stable terrain, i.e. where no velocity is expected. This gives a good overall indication of the bias introduced by the end-to-end velocity retrieval including co-registration of images, velocity retrieval, etc. For masking the moving ice we use a polygon shapefile of the ice sheet and peripheral glacier outlines produced by Rastner et al (2012, updated 2018). This shapefile is derived semi-automatically from Landsat 5 and 7, using a band ratio approach (red/SWIR) with scene specific thresholds and manual correction of debris cover, seasonal snow, shadow, water (outside of glaciers), sea ice and icebergs. By inverting the ice sheet/glacier shapefile a land mask file is created as depicted in Figure 4 (right).
2.2. Surface elevation change – D3.IS.6.1 and D3.IS.6.2
Independent estimates of the rate of surface elevation change at discrete locations and over specific time periods are provided by the Airborne Topographic Mapper, a scanning laser altimeter flown on board aircraft by Operation IceBridge (Studinger 2014). Observation campaigns are conducted in the hemispheric spring in most years, and dataset production usually lags by a year. The validating dataset used by both hemispheric SEC products is the level 4 product, IceBridge ATM L4 Surface Elevation Rate of Change V001. This can be obtained free of charge on registration, from https://icebridge.gsfc.nasa.gov/. The surface elevation rate of change in this dataset is calculated from the change in elevations recorded at the same location between two overflights. Operation IceBridge will cease at the end of the Antarctic campaign in 2020. IceSat-2 will be investigated as a possible source of validation data.
3. Description of product validation methodology
3.1. Greenland ice sheet velocity – D3.IS.4
1) The quality assessment for ice velocity includes detailed validation with contemporaneous in-situ GPS data at various sites across the ice sheet and acquired by field teams of the Danish Programme for Monitoring of the Greenland Ice Sheet (PROMICE; Fausto and Van As, 2019) operated by GEUS in collaboration with DTU Space and Asiaq and the Institute for Marine and Atmospheric Research at the University of Utrecht, The Netherlands (IMAU; C.H. Tijm-Reijmer, Pers. Comm).
2) Inter-comparison with MEaSUREs TerraSAR-X-based data (data set ID: NSIDC-0481, Joughin et al., 2011, updated 2019): As a pre-processing step, the IV maps are first converted from m/yr to m/d, resampled to a 250 m grid spacing and the separate vx and vy velocity files are merged into a 2-bands GeoTIFF to match the native Sentinel-1 IV maps geometry and format. As glacier velocity can fluctuate significantly over short time periods, in particular on the downstream section of large outlet glaciers, we only compare datasets derived from single Sentinel-1 SAR repeat pass pairs (6/12 days only) and with a maximum time difference of 2 days between the respective start dates (master). We inter-compare both vx and vy components separately on a pixel-by-pixel basis (ignoring the vertical component of velocity).
3) Inter-comparison with MEaSUREs Greenland Ice Sheet Velocity Map (data set ID: NSIDC-0478, Joughin et al., 2011, updated 2019): Pre-processing is similar as described above. For the intercomparison the 500m product is resampled to the grid extend of the C3S product. The inter-comparison is done on both the vx and vy components separately on a pixel-by-pixel basis (ignoring the vertical component of velocity).
4) Stable terrain test: The results for the ice covered (moving) area are separated from ice-free (stable) ground. The masking is done using a polygon of the rock outlines.
As a measure of quality, we provide, statistics on the component-wise (e.g for vx and vy) mean, the standard deviation and the root mean square error (RMSE) of the residuals (defined here as Sentinel-1 IV minus MEaSUREs IV for the IV intercomparisons). Residuals larger than 1 m/d are excluded from the statistics as these contaminate the statistics and are in general extreme outliers that are easily filtered out in the IV products.
3.2. Surface elevation change, Antarctica – D3.IS.6.1
The validation method is a comparison of SEC results with all ATM measurements from the validating dataset that coincide in time and space. This is not a direct comparison to the product dataset, as the ATM overflights are highly irregular. Instead we use the underlying elevation change timeseries and analysis methods from which the product dataset was derived to produce results comparable to the ATM.
To make the surface elevation change product, first we divide the Antarctic ice region into a 25km by 25km grid. In each grid cell we use crossover analysis to derive a series of elevation change values (dh) over time from each of five radar altimetry missions – ERS1, ERS2, Envisat, CryoSat-2 and Sentinel-3A. The five timeseries are cross-calibrated and combined to produce a single timeseries of elevation change (i.e., dh vs t).
We can use the dh timeseries in each cell to derive the rate of change of surface elevation, ie dh/dt, at any point in time. In the product the rate is calculated over a 5 year window that advances in monthly steps, however, for validation we apply the same time periods as the ATM to the grid cell timeseries in which the ATM took a measurement.
We averaged the rates of elevation change computed from pairs of IceBridge overflights in 25km cells to match the spatial resolution of the satellite data. Similarly, we averaged the rates of elevation change, derived by linear least-squares fitting from the surface elevation change data, comparable to the IceBridge pairs in length of time and grid cell location.
The comparison was then restricted to where the root mean square of the IceBridge results and the standard deviation of the satellite results was less than or equal to 0.4 m/year, and the time period between overflights was at least 2 years.
The airborne data is sampled at 250m spatial resolution along flight-lines that preferentially sample fast-thinning ice. Because of this, its measurements tend to be biased to a higher value when compared to coarsely gridded data (Flament and Remy 2012, McMillan 2014). To take account of this, a bias factor is estimated using a high-resolution (1km by 1km) map of ice velocity (Rignot et al 2011). This bias factor is the velocity of the 1km by 1km grid cell containing the ATM measurement with respect to the average velocity of the 25km by 25km AIS cell containing it, averaged over all measurements used.
The validation results come from the comparison of the averaged satellite and biased airborne measurements. A map of the airborne measurement locations used, a scatterplot of the two components of the comparison and a histogram of their differences are shown, and the mean and standard deviation of the differences derived.
3.3. Surface elevation change, Greenland – D3.IS.6.2
The validation method is a comparison of surface elevation change results with all ATM measurements for the Greenland ice sheet. As the ATM L4 product provides elevation change estimates of repeated flightpaths, the locations and observational period are highly irregular. To ensure consistency in the inter-comparison we construct a record based on the C3S surface elevation product, which coincide with the ATM-L4 data both in space and time. The gridded 25 km solution from satellite altimetry (the C3S surface elevation product) is interpolated onto the individual ATM-observations in space by linear least-squares fitting.
This reconstructed dataset is only available for internal use in the validation effort but is derived solely from the NetCDF-files available at CDS.
The comparison is then restricted to where the root mean square of the IceBridge results and the standard deviation of the satellite results is less than or equal to 5 m, and we do not have any limitations on the time between ATM overflights. As the ATM data is sampled at 250m spatial resolution along flight-lines that preferentially sample fast-thinning ice, its measurements tend to be biased to a higher value when compared to coarsely gridded data (Flament and Remy 2012, McMillan 2014). Different measures have been taken to limit the bias, however for the Greenland ice sheet validation we note this possible bias but neglect any corrections for it.
4. Summary of validation results
4.1. Greenland ice sheet velocity – D3.IS.4
Four different tests were performed for product quality assurance:
1) Intercomparison of Sentinel-1 derived velocity with in-situ GPS measurements for 2017/18 and 2018/19.
2) Inter-comparison of individual 6-12-day repeat Sentinel-1 derived ice velocity maps, used to produce the annually averaged C3S ice velocity map, with contemporaneous MEaSUREs TSX derived ice velocity maps for selected major Greenland outlet glaciers.
3) Inter-comparison of annually averaged Greenland Ice Sheet ice velocity map for 2017/2018 only with MEaSUREs 2017-2018 Greenland-wide IV map (covering primarily the winter campaign).
4) Stable terrain test, providing insight on the performance of the ice velocity retrieval algorithm, by analysing the results in stable terrain.
Table 1 and Table 2 provide a statistical overview of all inter-comparison results for CDR v2. For both annual maps the GPS intercomparison shows excellent agreement with mean differences of only 2-3 cm/d and an RMSE of 5 cm/d. The intercomparison of ice velocity on selected outlet glaciers also show a high level of agreement for both annual products, with a negligible overall mean bias and an RMSE of 18 cm/d for the easting and 21 cm/d for the northing component based on sample sizes of 11.4 million and 9.6 million for the 2017/18 and 2018/19 maps respectively. For the ice sheet wide product intercomparison between MEaSUREs and C3S (only available for the 2017/18 map) the results indicate a negligible bias with an RMSE of 4-5 cm/d. Based on 1.3 million pixels the outcome of the stable ground test indicates for both annual maps on average a negligible mean bias with an RMSE of 1 cm/d for both easting and northing components.
Table 1: Summary of inter-comparison results for 2017-2018, CDR v1 (values in m/d).
Product | Reference | Pixels | dE | RMSE E | dN | RMSE N |
C3S CDR (v1) | GPS | 20 | 0.02 | 0.05 | - | - |
S1 IV maps | MEaSUREs TSX (select. glac.) | 11.4 M | - | - | 0.00 | 0.18 |
C3S (ice sheet) | MEaSUREs S1 (ice sheet) | 8.4 M | - | - | 0.00 | 0.05 |
C3S (land) | Stable Terrain | 1.3 M | - | - | 0.00 | 0.01 |
Table 2: Summary of inter-comparison results for 2018-2019, CDR v2 (values in m/d).
Product | Reference | Pixels | dE | RMSE E | dN | RMSE N |
C3S CDR (v2) | GPS | 16 | 0.03 | 0.05 | - | - |
S1 IV maps | MEaSUREs TSX (select. glac.) | 9.6 M | - | - | 0.00 | 0.18 |
C3S (land) | Stable Terrain | 1.3 M | - | - | 0.00 | 0.01 |
For further details please see related document, the Product Quality Assessment Report (PQAR).
4.2. Gravimetric mass balance – D3.IS.5
The GMB products are brokered from the two ice-sheet CCI products and relevant documentation is available at:
• AIS: http://esa-icesheets-antarctica-cci.org/index.php?q=documents
• GrIS: http://esa-icesheets-greenland-cci.org/index.php?q=documents.
4.3. Surface elevation change, Antarctica – D3.IS.6.1
Here, we provide a short summary of the validation results, for further details please see related document, the Product Quality Assessment Report.
Figure 5 shows the comparison results. All IceBridge Antarctic data available from the portal, as of November 2019, were initially considered. However, flights were concentrated on the West Antarctic Ice Sheet, Filchner-Ronne Ice Shelf and surrounding regions. In total 206404 points of comparison, covering 408 grid cells, were used. The cell-averaged resistant mean difference (ie excluding outliers of more than 3 sigma) was 0.081 ± 0.717 m/yr, with a correlation coefficient of 0.55. This is within the Global Climate Observing System (GCOS) target accuracy of 0.1 m/yr, as specified by the C3S project.
Figure 5: Validation results – D3.IS.6.1 Left: map of locations where comparisons could be made. Middle: Scattergram of grid-cell-averaged dh/dt (m/yr) from ATM and C3S, one point per cell. If the values agreed perfectly, they would lie along the red line indicated. Right: Histogram of differences of grid-cell-averaged dh/dt (m/yr), ATM minus C3S, mean of distribution marked by the red line.
Previously, less than half the datapoints and a third of the grid cells used in the v2 comparison were used in v1. This is because the new cross-calibration method in v2 has retrieved data where sampling was temporally sparser than in v1. In some cases, these data come from rugged areas where the initial altimeter measurements are more uncertain than usual, especially the Antarctic Peninsula. This explains the higher uncertainty in the collected validation results. Figure 6 shows a repeat comparison using only the cells used by v1, and this shows little difference. Note the narrower distribution of differences in the right-hand histogram.
Figure 6: Validation comparison, using v2 at v1 locations only. Black = v1, red = v2. Left: map of locations where comparisons could be made. Middle: Scattergram of grid-cell-averaged dh/dt (m/yr) from ATM and C3S, one point per cell. If the values agreed perfectly they would lie along the black line indicated. Right: Histogram of differences of grid-cell-averaged dh/dt (m/yr), ATM minus C3S, means of distribution marked by the dotted lines.
4.4. Surface elevation change, Greenland – D3.IS.6.2
Here, we provide a short summary of the validation results, for further details please see related document, the Product Quality Assessment Report. Figure 7 shows the result of the inter-comparison between the OIB and the C3S surface elevation changes. The monthly time-series of surface elevation change grids makes it possible to tailor the time-series to resolve the timespan of OIB repeat locations on the Greenland ice sheet. Based on more than 25k observations, distributed both in time and space, a median of bias of -3 cm/yr in relation to the OIB data are found. This shows the product compliance to the GCOS requirements.
Figure 7: Comparison of the rate of elevation change observed in the OIB data and the C3S Greenland surface elevation product; version 1 and 2. As the OIB level 4 data consist of data from all repeats of older flight paths, this inter-comparison is based on observations from 1993 flightpath until 2017. The upper-left panel shows the point-point agreeance, alongside the one-to-one line. The lower-left panel shows the complete distribution for all years, which is averaged in the right panel to show the spatial distribution.
References
Fausto, R.S. and van As, D., (2019). Programme for monitoring of the Greenland ice sheet (PROMICE): Automatic weather station data. Version: v03, Dataset published via Geological Survey of Denmark and Greenland. DOI: https://doi.org/10.22008/promice/data/aws
Flament, T. and F. Remy (2012). Antarctica volume change from 10 years of Envisat altimetry. Conference paper, International Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE International. DOI: 10.1109/IGARSS.2012.6351149
Joughin, I. 2002. Ice-Sheet Velocity Mapping: A Combined Interferometric and Speckle-Tracking Approach. Annals of Glaciology 34: 195-201. doi: 10.3189/172756402781817978.
Joughin, I., I. Howat, B. Smith, and T. Scambos. 2020, updated 2019. MEaSUREs Greenland Ice Velocity: Selected Glacier Site Velocity Maps from InSAR, Version 2. Boulder, Colorado USA. NASA National Snow and Ice Data Center Distributed Active Archive Center. doi: [https://doi.org/10.5067/JQHJUOYCF2TE|https://doi.org/10.5067/JQHJUOYCF2TE]. [Date Accessed: May 2020].
Joughin, I., B. Smith, I. Howat, and T. Scambos. 2015, updated 2018. MEaSUREs Greenland Ice Sheet Velocity Map from InSAR Data, Version 2. Boulder, Colorado USA. NASA National Snow and Ice Data Center Distributed Active Archive Center. doi: [https://doi.org/10.5067/OC7B04ZM9G6Q|https://doi.org/10.5067/OC7B04ZM9G6Q]. [Date Accessed: September 2018].
McMillan, M., A. Shepherd, A. Sundal, K. Briggs, A. Muir, A. Ridout, A. Hogg and D. Wingham (2014). "Increased ice losses from Antarctica detected by CryoSat-2." Geophysical Research Letters 41 (11): 3899 -3905
National Geospatial Intelligence Agency, 2004. https:earth-info.nga.mil/GandG/publications/tr8350.2/tr8350_2.html
Rastner, P., Bolch, T., Mölg, N., Machguth, H., Le Bris, R., Paul, F. 2012, updated 2018, The first complete inventory of the local glaciers and ice caps on Greenland. The Cryosphere, 6, 1483-1495. (doi:10.5194/tc-6-1483-2012)
Rignot, E., J. Mouginot and B. Scheuchl (2011). "Ice Flow of the AIS." Science 333 (6048): 1427-1430
Rizzoli, P., Martone, M., Gonzalez, C., Wecklich, C., Borla Tridon, D., Bräutigam, B., Bachmann, M., Schulze, D., Fritz, T., Huber, M., Wessel, B., Krieger, G., Zink, M., and Moreira, A. (2017): Generation and performance assessment of the global TanDEM-X digital elevation model. ISPRS Journal of Photogrammetry and Remote Sensing, Vol 132, pp. 119-139.
Studinger, M. (2014). IceBridge ATM L4 Surface Elevation Rate of Change, Version 299 1, Antarctica subset. N. S. a. I. D. C. D. A. A. Center. Boulder, Colorado, USA. DOI: 10.5067/BCW6CI3TXOCY
