Last modified on Nov 28, 2023 17:06

Contributors: Frank Paul (University of Zurich), Michael Zemp (University of Zurich), Philipp Rastner (University of Zurich), Jacqueline Bannwart (University of Zurich)

Issued by: UZH / Frank Paul

Date: 02/08/2021

Ref: C3S_312b_Lot4.D2.GL.2-v3.0_Product_Quality_Assessment_Report_i1.0-area

Official reference number service contract: 2018/C3S_312b_Lot4_EODC/SC2

History of modifications

List of datasets covered by this document

Related documents

Acronyms

General definitions

This document¹ refers to the following versions of the datasets in the Climate Data Store (CDS):
Glacier distribution service: RGI v6

Scope of the document

This document is the Product Quality Assessment Report (PQAR) for the Copernicus glacier distribution service providing results of the quality assessment for the datasets generated by C3S. The related document for the glacier change service (which provides glacier elevation and mass changes) is presented in [RD1]. The regionally constrained datasets discussed here do not strictly refer to a specific version of the datasets in the Climate Data Store (CDS), but have been provided to the GLIMS glacier database. The global scale datasets available in the CDS (the Randolph Glacier Inventory, RGI) are extracted from the GLIMS database with a timing that is independent from the C3S schedule. We here present results for datasets in (a) RGIv5, (b) RGI v6, (c) the forthcoming RGIv7 and (d) as submitted to GLIMS for integration in future versions of the RGI. An overall assessment of RGI quality based on Pfeffer et al. (2014) is provided in the PQAD [RD2].

Executive summary

This updated version of the PQAR (v3.0) for the glacier distribution service presents in Section 2.4 the results of the quality assessment for five datasets to be included in the forthcoming RGIv7. As each new version of the RGI is a reprocessed dataset based on all of its earlier versions, we also present here the results of earlier assessments, sorted for the four categories mentioned above. Sections 2.1 and 2.2 present results for datasets included in RGIv5 and RGIv6, respectively. The datasets presented in Section 2.3 have been provided to GLIMS for possible inclusion in future versions of the RGI (e.g. RGIv8). Further details of all datasets are provided in the Table 1.

Table 1: Overview of the datasets with quality assessments presented in this document. The datasets 8 to 12 in the grey-shaded cells have been added for this version 3 of the PQAR.

1. Product validation methodology

The methods to determine product quality are described in Section 3.3.2 of the TRGAD [RD1] in detail and are thus only shortly repeated here. The created outlines are always visually inspected and manually corrected by the analyst before submission. This correction is required to achieve a product accuracy that is compliant with GCOS requirements (area better than 5%), as for example debris-covered glacier parts are not automatically mapped and can hide far more than 50% of the glacier area. It is this manual correction that introduces uncertainty to the product and thus needs to be assessed. The four methods we apply in C3S to determine product accuracy are:

(1) Overlay of outlines (for interpretation and visualization of differences),

(2) buffer method (extend and shrink all outlines by 1/2 or 1 pixel),

(3) independent multiple digitizing (at least 5 glaciers three times), and

(4) comparison with glacier areas derived from a higher resolution dataset.

From (1) to (4) the workload of the assessment increases and not always all methods can be applied. In particular, method (4) is seldom used as the related images have to be freely available and acquired around the same week as the satellite image. When manual corrections are sparse, also method (3) is not always applied. The methods selected for uncertainty assessment do thus vary with the region and the corrections applied.

2. Results of the quality assessment

2.1. RGIv5: European Alps

For the glacier outline dataset in the RGI5.0 covering the European Alps the following validation results were achieved: The overlay of outlines presented in Fig. 1 revealed that the TM-derived outlines from 2003 (yellow) could not be compared to the independent dataset from the 1999 Austrian glacier inventory (black lines) as this analysis obviously included large amounts of seasonal snow, i.e. glaciers were in general way too large in that inventory. On the other hand, Fig. 1b also revealed that a debris-covered tongue was not mapped in 2003 (white arrow) so that the size of this glacier is largely underestimated. This stresses the importance of always checking the new outlines against high-resolution images in Google Earth or other tools / map servers and correcting them as appropriate.

Figure 1: Two regions with glaciers in the Hohe Tauern mountain range (Austria) as seen in a Landsat TM false-colour composite from 2003 with the outlines derived from Landsat (yellow) and the 1999 inventory from Austria (black) derived from aerial photography. Panel a) shows Olperer Ferner (O) in the Zillertaler Alps and b) Habach Kees (H) in the Venediger Group. The arrow points to missed debris cover (from Paul et al. 2011).

A second quality assessment was based on a comparison of the automatically derived glacier outlines with the manually digitized outlines for three glaciers (Fig. 2). The digitizing was repeated independently five times to obtain a mean value and a standard deviation. As a third quality measure and validation, glacier outlines were also derived from a very high-resolution Ikonos image using manual digitizing. The related outlines are displayed in Fig. 2 as an overlay on the high-resolution image. Derived areas and related differences are summarized in Table 2. The results reveal that the standard deviation of the multiple manual digitization's is larger than the respective relative area difference, that the overall difference is small (-0.7%), and that differences for individual and very small glaciers (<0.05 km²) can be comparably large (here -10.5%). The comparison with Ikonos shows that the latter can also be the case for glaciers with a size of 0.1 km² (showing a difference of -6.4%). Interestingly, the mean value area difference is still small (-1.5%) as the automatically derived extents deviate in both directions.

Figure 2: Overlay of glacier outlines derived automatically from TM (black lines) and manual digitizations (coloured lines) on a high resolution Ikonos image (acquired in the same month, screenshot from Google Maps) for three glaciers in South Tyrol. a) Steinschlagferner and b) Schwemser Ferner (from Paul et al. 2011).

Table 2: Derived glacier areas and differences for the three glaciers shown in Fig. 2 for TM (automatically and manual) and for Ikonos (manual). The last row presents sums and averages (from Paul et al. 2009).

2.2. RGIv6: Karakoram / Pamir

The new inventory for the Karakoram / Pamir region is largely based on earlier work in the Glaciers_cci project (Fig. 3). In C3S we have enriched the dataset by supplementing information about debris cover (as separate polygons) and performing three different types of uncertainty assessment: (1) fixed uncertainty values applied to all glaciers (±2.5% for clean ice, ±5% for debris-covered ice), (2) the buffer method with different buffer sizes for clean (±1/2 pixel) and debris-covered glacier parts (±1 pixel), and (3) independent multiple digitization of outlines by all analysts for three difficult debris-covered glaciers. Methods (1) and (2) are applied to glacier complexes excluding overlapping areas as well as the boundary of clean and debris-covered ice. The multiple digitizing method (3) was required due to the high abundance of debris-covered glaciers in the study region. All participants manually corrected three times the outlines of three example glaciers from different regions with differing additional information being considered (e.g. coherence images and Google Earth imagery). The glaciers are of different size and contain a substantial debris-covered part, combined with difficulties of moraines, glacier confluences, and cast shadow.

Figure 3: The new glacier inventory and satellite footprints for the Pamir and Karakoram region (from Mölg et al. 2018).

According to Mölg et al. (2018), method (1) resulted in a total glacier area of 35'287 ±1944 km² for 23,000 glaciers >0.02 km², very similar to the uncertainty of method (2) with ±1948 km². The multiple digitization experiment (3) resulted in a ±13% standard deviation (averaged over all experiments). This value is rather high, but results from only one glacier (with an uncertainty of 33%) that is mostly in shadow and has a barely visible debris-covered part. Although this example reflects the mapping reality under challenging conditions with debris-covered ice in shadow, it is a worst-case scenario and only few of such glaciers exist in a larger inventory. Hence, the high uncertainty of this example has little impact on the overall uncertainty. The uncertainty of the other two glaciers was <5%. Figure 4 shows the raw and the corrected outlines from the four participants.

Figure 4: a) Raw classification of clean ice (yellow) for one of the example glaciers. (b) Overlay of glacier outlines from different analysts revealing the differences in interpretation.

2.3. GLIMS database

2.3.1. Novaya Zemlya (2013-2016)

We have created a new glacier inventory for Novaya Zemlya in the Russian Arctic from 9 Landsat scenes acquired in 2013 and 2015 (3 scenes each), 2016 (one scene) and 1998 (2 scenes), the latter only being used for removal of seasonal snow cover. The related inventory has been submitted to GLIMS and the methodological description and results analysis is presented by Rastner et al. (2017). As the region is dominated by ice caps, only few glaciers have a little bit of debris cover and manual editing was very limited. To determine the uncertainty of the generated outlines, we have thus applied the buffer method with a ±1/2 pixel margin. This gives an uncertainty of 0.5% for the northern part that is dominated by a large ice cap (or an area of 20,784.4 ±102.5 km²) and an uncertainty of 4% for the southern part that is dominated by numerous much smaller glaciers and ice caps (with a total area of 1,612.6 ±64.1 km²). Figure 5a shows an overview of the new inventory and Fig. 5b a comparison with the RGI5.0 for a small sub-region.

Figure 5: a) Outlines of the new glacier inventory for Novaya Zemlya (white) on top of the outlines from the RGI inventory (yellow). Both images are from Rastner et al. (2017). b) A close-up from the southern part of the island, illustrating differences in interpretation between the two inventories.

2.3.2. Franz-Josef-Land (2016)

A new glacier inventory has also been created for Franz-Josef-Land. In this case we used a Sentinel 2 scenes acquired on 12.9.2016 for most of the ten tiles (2 were from July 2016). An overview of the study region is presented in Fig. 6. This figure also illustrates that the ice caps and outlet glaciers are largely debris free, i.e. the debris-related corrections were minimal. On the other hand, despite the superb snow conditions, some seasonal snow remained outside of the ice caps and had to be corrected manually. Moreover, parts of some islands were covered by low-lying but semi-transparent clouds. Here glaciers were also corrected manually as the automated method missed these regions. Uncertainty was determined with the buffer method, using a ±1 pixel margin. Considering all glaciers larger 0.05 km² (and merging all ice cap entities) the total mapped area of 990 glaciers and ice caps is 12,363.5 km² ±86.5 km², i.e. the uncertainty is 0.7%. The close-up in Fig. 7 shows our new outlines from 2016 compared to RGI outlines from 2000-2010.

This overlay shows strong glacier retreat from the RGI extents (derived from satellite images acquired between 2000 and 2010) to 2016 in particular for marine-terminating glaciers. The overlay also reveals that drainage divides between glaciers are obviously at the wrong place compared to the relief. This is a result from the low-quality DEM that was used to derive it. For the new inventory we have used the high-quality ArcticDEM (resampled to 10 m spatial resolution) to derive drainage divides but the former divides as a principal guide to separate glacier entities (which is otherwise for ice caps somewhat arbitrary). This correction alone changed the sizes of individual glaciers largely so that a direct comparison with RGI extents is not possible at the glacier scale. It would still work at the level of complete islands, but here the uncertainty regarding the date of RGI extents would be large and area change rates difficult to compute.

Figure 6: The Sentinel 2 tiles used to create the new glacier inventory for Franz-Josef-Land. The inset shows glacier outlines with drainage divides derived from the ArcticDEM. The arrow points to Luigi Island (Fig. 7).

Figure 7: Overlay of glacier outlines from the new inventory (yellow) and the RGI (red) for Luigi Island.

2.3.3. Western Greenland (2016)

The new glacier inventory for western Greenland has been created from four Landsat 8 OLI scenes acquired in 2016 (3 scenes) and 2015 (1 scene). Glaciers were mapped automatically using the pan / SWIR band ratio to have outlines with 15 m resolution. Due to misclassification for glaciers in deep shadow or covered by debris, manual editing was applied to correct the outlines for hundreds of glaciers. Hence, the best method to determine uncertainty for this dataset is a multiple digitizing experiment (Paul et al. 2017). We have thus selected a most challenging region with glaciers in shadow and under debris cover and one person digitized 16 of them independently three times. As for the original inventory, a Sentinel 2 scene acquired in the same month is used for the corrections. An overlay of the digitized outlines is shown in Fig. 8 revealing a mean difference of 8 ±20%, the latter being the standard deviation of the relative area difference variability.

Figure 8: Overlay of corrected glacier outlines from the original inventory (green) and three independent manual digitisations (yellow, red and white).

2.3.4. Alps (2015/16)

We created a new glacier inventory for the Alps using Sentinel-2 scenes acquired in 2015 and 2016 (Paul et al. 2019). Due to the large number of debris-covered glaciers in this region and compilation of outlines from different analysts, we decided performing a multiple digitizing test according to Paul et al. (2013) to determine the uncertainty of the outlines. A related overlay of the outlines from all analysts is depicted in Figure 9. In the mean over all glaciers and analysts, we derived an area uncertainty of 3.4% but values for small glaciers were partly larger than 10%. It could also be shown that glacier extents generally increased a bit after consulting very high-resolution imagery. To a large part this was due to a more generous interpretation of debris-covered regions, in particular ice-cored lateral moraines. The consistency in the interpretation among the analysts was comparably high as for example the change to the very high-resolution imagery resulted in a larger change of glacier extents among all analysts. The overlay of outlines in Figure 9 reveals that the largest variability in interpretation occurred for debris-covered glacier parts.

Figure 9: Overlay of multiple digitisations from five different analysts (colour-coded) for a sub-region of the southern Swiss Alps (Otemma Glacier in the lower centre). The background shows the Sentinel-2 image (in false colours) used for the corrections.

2.3.5. Svalbard (2016/17)

We have created a new glacier inventory for Svalbard using Sentinel-2 images acquired on 31.7. and 2.8.2017 as well as a Landsat 8 scene from 19.8.2016. The inventory that is currently in RGIv6 has been compiled by Nuth et al. (2013), using about 40 different satellite scenes acquired between 2000 and 2010. Glacier outlines for this dataset had been digitized manually, resulting in generalization of fine spatial details and smaller glaciers (<1 km²) not being mapped. Hence, the development of these likely very sensitive climate indicators could not be followed. The new inventory overcomes these shortcomings but might also include ice or perennial firn patches that are not glaciers. Figure 10 illustrates the differences between the old (yellow) and the new dataset (red). In particular in the right part of the image several 'red only' glaciers can be seen, i.e. they are not included in RGIv6. For two other glaciers the 2017 extents are much larger than those from RGIv6, indicating that a different (less conservative) interpretation of their debris-covered parts has been applied. The figure also reveals that some drainage divides have been adjusted so that they match flow divides derived from the ArcticDEM. Real glacier change between the acquisition dates of the images used to create the outlines is comparably small, in particular smaller than due to the different interpretation.

Figure 10: Glacier outlines from RGIv6 (yellow) compared to the outlines of the new inventory from 2016 (red) for a small sub-region in northern Svalbard. The differences are due to a different interpretation rather than glacier growth.

2.4. RGIv7

It was agreed by the IACS WG on the RGI (https://cryosphericsciences.org/activities/working-groups/rgi-working-group/) that the next version of the RGI (v7) should have its temporal focus closer to the year 2000 and that regional quality shortcomings should be improved where possible. Shortcomings include nominal glaciers (represented by circles of equivalent size) that should be replaced with real glacier outlines and correction of wrongly mapped seasonal snow, debris-cover or (frozen) water bodies. To help with this, the community was asked to provide datasets that had not been submitted to GLIMS so far when they fulfill the above criteria. More than ten new datasets were submitted and evaluated by the C3S team. Together with the previously identified shortcomings of RGIv6 (see TRGAD, [RD3]) it was decided which RGI regions or sub-regions will be updated in RGIv7 using either the new datasets submitted by the community or by creating new outlines in C3S. In the following, we present the quality assessment results of (a) the RGI6 (Section 2.4.1), (b) of the new datasets submitted by the community (Section 2.4.1), and (c) the corrections performed for both (a) and (b) in Sections 2.4.2 to 2.4.6.

2.4.1. Overall assessment of RGIv6 and applied changes

To determine which regions in RGIv6 should be changed, we analysed a couple of characteristics for all 19 RGI main regions and decided on how to proceed. This analysis revealed several regions that will stay as they are because (i) their quality was already good, (ii) they refer to the years around 2000, or (iii) they cannot be improved with scenes acquired closer to the year 2000 (e.g. due to missing data). Further, we identified regions that (iv) should be locally or regionally improved or (v) created because they are missing. The RGI regions without a change are 1, 2, 6, 9, and 10. The reasons for improving the other regions and the applied changes are listed in Table 3.

Table 3: Overview of the RGI regions (column 'RGI') that were changed, 'gl.' is 'glaciers'.

Overall, the call for data by the RGI WG provided seven new datasets that were evaluated for possible inclusion in RGIv7. We decided to use four of them (Ellesmere, Finnmark, Kamchatka, Pyrenees) and rejected the other three (Glacier NP, Alps, Peru/Bolivia). Additionally, we considered the recently created national inventories of Chile (Barcaza et al. 2017) and Argentina (Zalazar et al. 2020) for the new RGIv7 as they clearly surpass the quality of the existing datasets. Reasons for rejecting the other three were related to required corrections (that were estimated to be in the same range as correcting the original dataset) or too many missing glaciers. The Kamchatka dataset was rejected at first, but the authors provided later an improved version that was integrated.

2.5. Ellesmere (1999)

For Ellesmere Island we received a new dataset of glacier outlines referring to 1999 from the science community. We decided to integrate them into the new RGI because their drainage divides were much better and also corrected for an obvious small shift visible in RGI6. Closer inspection, however, revealed, that several glaciers (large and small) were missing and numerous rock outcrops had not been digitized, i.e. the mapped glaciers were too large. We decided to correct both drawbacks semi-automatically by a copy/paste operation of both missing glaciers and rock outcrops obtained by automated mapping of three Landsat ETM+ scenes (acquired also in 1999), but unfortunately this operation did not work perfectly so that we had to apply a large number of manual corrections, in particular by adding further rock outcrops (Figure 11). Also a few drainage divides were corrected using a flow-direction grid derived from the ArcticDEM, but this was a comparably small effort. In a last step, we integrated the revised dataset into the existing inventory that included more glaciers in the north. As the drainage divides for these glaciers were also of poor quality, we corrected them with new divides derived from the ArcticDEM.

Figure 11: The three Landsat ETM+ images used for correction of the glaciers and ice caps on Ellesmere Island. Black are the outlines from the provided dataset, glaciers in yellow were added and rock outcrops in white and red automatically or manually added, respectively. The inset shows a detail from the upper centre.

In total we edited about 2000 glaciers. The mapped glacier area in the provided dataset was too small by 92 km² from the missing glaciers and 170 km² too large from the missing rock outcrops. The total area before was 14575 km² and afterwards 14497 km². Overall, our edits corrected the glacier area by 1.8%. This might seem as a small correction at first glance, but for example the added rock outcrops have likely a strong impact on the calculated ice thickness distribution and thus on the total ice volume and its temporal evolution and sea-level contribution.

2.6. Baffin Island (ca. 2000)

For Baffin Island we had correct target dates of the outlines in RGI6, but recognized at closer inspection that obviously debris-covered parts were missing for many larger glaciers. As this has quite some impact on their size and other calculations (e.g. the global assessment of debris cover on glaciers by Scherler et al. 2018) we decided to correct these regions using satellite data from around the year 2000. Unfortunately, we detected during this correction, that numerous glacier outlines were much too large compared to glacier extents in 2000 and had to be corrected as well (Figure 12). Moreover, numerous smaller glaciers were missing and had to be added and drainage divides were often in wrong places. In the end, we basically checked all 7500 glaciers on Baffin Island and corrected several thousand. Considering the strong contribution to sea-level rise from this region (e.g. Zemp et al. 2019, Hugonnet et al. 2021), we think this correction was important despite the required high effort. Overall, the area changed from 19,197 km² to 18,777 km² with 287 km² newly mapped and 420 km² wrongly mapped in RGI6. The total wrong area is thus 707 km² or 3.7% of the original area. This also sounds like a small correction, but considering the wrong geometries and the partly rapid melting of small ice caps, the correction was important.

Figure 12: Two examples showing corrected glacier outlines (yellow) compared to outlines form RGI6 (black) for Baffin Island. Outlines were too large, missing or otherwise wrong.

2.7. Northern Greenland (1978)

The dataset covering the local or peripheral glaciers and icecaps on Greenland was created from Landsat TM and ETM+ scenes acquired between 1994 and 2008, but centered on the 1999-2003 period (Rastner et al. 2012). It was thus suitable for RGI7 and had not to be changed in this regard. However, as Landsat did not cover regions north of 80° N, the outlines of the glaciers north of it had to be taken unchecked from other available sources. This check has now been performed and the dataset was corrected using freely available orthoimages (Korsgaard et al. 2016) and the ArcticDEM. In Figure 13 we illustrate the differences between the two datasets. The previously mapped glacier area in this region has changed from 2442.5 km² to 3062.8 km² whereby 682.6 km² is newly mapped and 67.4 km² was wrongly mapped as glaciers. The combined omission and commission errors are thus 750 km² or 30% of the previously mapped glacier area. The number of glaciers changed from 133 to 375. We are aware that the aerial photography from 1978 is quite far away from the target year 2000. However, considering that there is likely only very limited area change in this region and the previous outlines were excluding many glaciers, the improvement is still large.

Figure 13: Overlay of corrected glacier outlines (yellow) with outlines form the original inventory (red) for a small region in northern Greenland. Several glaciers were not or only partly included.

2.8. Peru/Bolivia (1998)

Glacier shrinkage in Peru and Bolivia is strong and ongoing since a few decades (Rabatel et al. 2013, Seehaus et al. 2019). Accordingly, more recent glacier extents should be smaller than older ones. However, the RGIv6 glacier extents in this region referring to the time period after 2000 were much larger than extents in 1998 mostly due to wrongly mapped seasonal snow. The outlines also included wrongly mapped lakes and excluded many rock outcrops. Finally, the raster vector conversion has seemingly used the 'simplify polygons' option that resulted in sharp, triangular shaped glacier boundaries. We thus decided to improve the dataset by creating a new one using 16 Landsat 5 TM scenes acquired during the El Nino year 1998 where basically no seasonal snow was left outside of glaciers. In Figure 14 we show a comparison of the new outlines (uncorrected) with those from the RGI, illustrating the large overestimation of glacier extents in this region. Due to strong glacier shrinkage, all yellow outlines should be inside the white outlines. As the improvement is obvious from this comparison, we leave the quality assessment with this.

Figure 14: Overlay of corrected glacier outlines (white) with outlines form the original inventory (yellow) for a small region in Bolivia.

2.9. New Zealand (around 2000)

The glacier outlines In RGIv6 for New Zealand were from 1978 and thus quite far away from the target year 2000. Their quality was also not exactly known. A new inventory was compiled by Baumann et al. (2020) using Sentinel-2 data acquired in 2016. Also this dataset was quite far away from the target date. As promising Landsat ETM+ satellite images were available from ETM+ for around 2000, we decided using the new inventory from 2016 as a base to create a consistent year 2000 inventory by adding the larger extents from 2000 to the existing 2016 outlines by visual interpretation of the Landsat ETM+ panchromatic band (providing 15 m spatial resolution). The corrections worked in general well, but had to be applied to all 3000 glaciers of the inventory. Moreover, for the southern region in the Southern Alps of New Zealand, the available satellite scenes had clouds, seasonal snow or both. In Figure 15 we show the resulting changes in glacier extent for a small sub-region with three different images in the background. The scene used for most of the corrections of the glaciers in the northern part of the Southern Alps of New Zealand (from 13.4.2000) has no coverage to the left of the yellow line (left panel) and has thus to be replaced by other scenes. The scene from 20.4.2000 (middle) was locally cloud covered so that a third scene from 1.3.2002 (right) had to be used. However, this scene has remaining seasonal snow and was difficult to interpret. The resulting outlines were thus partly a bit of guesswork.

In the case a change could not be properly determined, we kept the 2016 outlines also for 2000. This mostly occurred for small completely snow-covered ice or firn patches. As also differences in the interpretation of debris-covered glaciers, avalanche cones, shadow regions and small perennial ice patches occurred (compared to the 2016 dataset), we decided performing a quality check by the providers of the 2016 outlines. This resulted in a long list of about 40 mostly small changes (at the level of individual pixels) that were addressed one-by-one and clarified. Overall, the year 2000 glacier extent in New Zealand was about 11.5% larger than in 2016, which is clearly outside the uncertainty of 4.3% given for the year 2016 inventory. However, we assume the area uncertainty of the Y2000 inventory is in the range 5-10%, in particular in regions with glaciers surrounded by seasonal snow, located in cast shadow, or being debris covered. The new Y2000 inventory also includes several ice or firn patches and avalanche cones that might contain ice but are not necessarily glaciers that flow.

Figure 15: Overlay of new glacier outlines for 2000 (green) with outlines from the 2016 inventory (pink) by Baumann et al. (2020) for a sub-set of the New Zealand dataset. The three panels show Landsat 7 ETM+ pan-chromatic bands acquired on 13.4.2000 (left), 20.4.2000 (middle) and 1.3.2002 (right), illustrating varying conditions of coverage (left, only the part to the right of the yellow line was covered by the scene from 13.4.2000), clouds (middle, partly hiding glaciers) and seasonal snow (right). The new outlines for 2000 were generated using all three scenes.

3. Compliance with user requirements

Many of the uncertainties and accuracy results reported in Section 2 are compliant with the GCOS requirements (better than 5%). Individual deviations can, however, be larger. This is not only true for very small glaciers (say <0.1 km² as the uncertainty increases towards smaller glaciers), but also for a wrong methodological interpretation such as missed debris cover or wrongly mapped seasonal snow fields. In such cases errors can be even larger than 50% so that the detailed uncertainty assessment presented above is more of an academic nature. Most of the datasets in the RGI have also a high accuracy, but in some mountain regions problems with wrongly mapped seasonal snow (Andes) or debris cover (High Mountain Asia) are still prevalent. The summary plot of area uncertainties in the RGI presented in Fig. 4 of the PQAD [RD2] reveals that individual glaciers should be larger than 1 km² to obtain a mapping accuracy better than 5%. Under favorable mapping conditions this might also be achieved for smaller glaciers (e.g. down to 0.1 km²), but in case of debris cover on a glacier (or seasonal snow hiding its perimeter), accuracy will in general be reduced.

References

Barcaza, G., S.U. Nussbaumer, G. Tapia, J. Valdés, J.-L. García, Y. Videla, A. Albornoz and V. Arias (2017): Glacier inventory and recent glacier variations in the Andes of Chile, South America. Annals of Glaciology, 58(75), 166–180.

Baumann S, Anderson B, Chinn T, Mackintosh A, Collier C, Lorrey AM, Rack W, Purdie H, Eaves S (2020): Updated inventory of glacier ice in New Zealand based on 2016 satellite imagery. Journal of Glaciology, 67 (261), 13–26.

Hugonnet, R., McNabb, R., Berthier, E., Menounos, B., Nuth, C., Girod, L., Farinotti, D., Huss, M., Dussaillant, I., Brun, F., and Kääb, A. (2021): Accelerated global glacier mass loss in the early twenty-first century. Nature, 592, 726–731.

Korsgaard, N.J., C.Nuth, S.A. Khan, K.K. Kjeldsen, A.A. Bjørk, A. Schomacker and K.H. Kjær (2016): Digital elevation model and orthophotographs of Greenland based on aerial photographs from 1978–1987. Scientific Data, 3, 160032; doi.org/ 10.1038/sdata.2016.32.

Mölg, N., Bolch, T., Rastner, P., Strozzi, T. and Paul, F. (2018): A consistent glacier inventory for the Karakoram and Pamir region derived from Landsat data: Distribution of debris cover and mapping challenges. Earth Systems Science Data, 10, 1807–1827.

Nuth, C., Kohler, J., König, M., von Deschwanden, A., Hagen, J. O., Kääb, A., Moholdt, G., and Pettersson, R. (2013): Decadal changes from a multi-temporal glacier inventory of Svalbard, The Cryosphere, 7, 1603–1621

Paul, F., Frey, H. and Le Bris, R. (2011): A new glacier inventory for the European Alps from Landsat TM scenes of 2003: Challenges and results. Annals of Glaciology, 52 (59), 144–152.

Paul, F., T. Bolch, K. Briggs, A. Kääb, M. McMillan, R. McNabb, T. Nagler, C. Nuth, P. Rastner, T. Strozzi, J. Wuite (2017): Error sources and guidelines for quality assessment of glacier area, elevation change, and velocity products derived from satellite data in the Glaciers_cci project. Remote Sensing of Environment, 203, 256–275.

Pfeffer, W.T., 18 others and the Randolph Consortium (2014): The Randolph Glacier Inventory: a globally complete inventory of glaciers. Journal of Glaciology, 60 (221), 537–552.

Rastner, P., Strozzi, T. and Paul, F. (2017): Fusion of multi-source satellite data and DEMs to create a new glacier inventory for Novaya Zemlya. Remote Sensing, 9(11), 1122; doi: 10.3390/rs9111122.

Rabatel, A., and 27 others (2013): Current state of glaciers in the tropical Andes: a multi-century perspective on glacier evolution and climate change, The Cryosphere, 7, 81–102

Seehaus, T., Malz, P., Sommer, C., Lippl, S., Cochachin, A., and Braun, M. (2019): Changes of the tropical glaciers throughout Peru between 2000 and 2016 – mass balance and area fluctuations. The Cryosphere, 13, 2537–2556.

Zalazar, L., L. Ferri, M. Castro, H. Gargantini, M. Gimenez, P. Pitte, L. Ruiz, M. Masiokas, G. Costa and R. Villalba (2020): Spatial distribution and characteristics of Andean ice masses in Argentina: results from the first National Glacier Inventory. Journal of Glaciology, 66 (260), 938–949.

Zemp, M., Huss, M., Thibert, E., Eckert, N., McNabb, R., Huber, J., Barandun, M., Machguth, H., Nussbaumer, S.U., Gärtner-Roer, I., Thomson, L., Paul, F., Maussion, F., Kutuzov, S. and Cogley, J.G. (2019): Global glacier mass changes and their contributions to sea-level rise from 1961 to 2016. Nature, 568, 382–386.

Related articles

• Page:
  ERB TSI: Product User Guide and Specification (PUGS)
• Page:
  ERB TSI: Product Quality Assurance Document (PQAD)
• Page:
  ERB TSI: Product Quality Assessment Report (PQAR)
• Page:
  ERB TSI: System Quality Assurance Document (SQAD)
• Page:
  ERB TSI: Algorithm Theoretical Basis Document (ATBD)