EGU2020-20908
https://doi.org/10.5194/egusphere-egu2020-20908
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

A globally complete, spatially and temporally resolved estimate of glacier mass change: 2000 to 2019

Romain Hugonnet1,2, Robert McNabb3,4, Etienne Berthier1, Brian Menounos5, Christopher Nuth4,6, Luc Girod4, Daniel Farinotti2,7, Matthias Huss2,7,8, Ines Dussaillant1,9, Fanny Brun10, and Andreas Kääb4
Romain Hugonnet et al.
  • 1Université de Toulouse, CNRS, LEGOS, Toulouse, France. (romain.hugonnet@legos.obs-mip.fr)
  • 2Laboratory of Hydraulics, Hydrology and Glaciology (VAW), ETH Zürich, Zürich, Switzerland.
  • 3School of Geography and Environmental Sciences, Ulster University, Coleraine, United Kingdom.
  • 4Department of Geosciences, University of Oslo, Oslo, Norway.
  • 5Natural Resources and Environmental Studies Institute and Geography, University of Northern British Columbia, Prince George, British Columbia, Canada.
  • 6The Norwegian Defense Research Establishment, Kjeller, Norway.
  • 7Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Birmensdorf, Switzerland.
  • 8Department of Geosciences, University of Fribourg, Fribourg, Switzerland.
  • 9Department of Geography, University of Zurich, Zurich, Switzerland.
  • 10IGE, Université Grenoble Alpes, CNRS, IRD, Grenoble INP, Grenoble, France.

The world’s glaciers distinct from the Greenland and Antarctic ice sheets are shrinking rapidly, altering regional hydrology and raising global sea level. Yet, due to the scarcity of globally consistent observations, their recent evolution is only known as a heterogeneous temporal and geographic patchwork and future projections are thus not optimally constrained.

Here, we present the first globally complete, consistent and resolved estimate of glacier mass change derived from more than half a million digital elevation models (DEMs) generated or extracted from multiple satellite archives including ASTER, ArcticDEM and REMA. Combining state-of-the-art numerical photogrammetry and novel statistical approaches, we reconstruct two decades of glacier surface elevation change at an unprecedented spatial and temporal resolution. We validate our results by comparing them to independent, high-precision elevation measurements from the ICESat and IceBridge campaigns, as well as to very high resolution DEM differences from LiDAR, Pléiades, and SPOT-6. The elevation time series are integrated to volume changes for every single glacier on Earth and, by assuming an average density, aggregated to regional and global mass changes. We compare our revised glacier mass changes to earlier estimates derived from altimetry, gravimetry, geodetic and field data. As an illustration, our integrated geodetic mass loss over all Icelandic glaciers yields -8.3 +- 1.1 Gt yr-1 over the period 2002-2016 in agreement with a recent gravimetry estimate of -8.3 +- 1.8 Gt yr-1 (Wouters et al., 2019), known to perform well in this region. Both estimates are more negative than -5.7 +- 1.2 Gt yr-1, compiled from glaciological observations and geodetic data (Zemp et al., 2019).

Our global estimate of glacier mass change constitutes a new benchmark dataset that will help to: (i) assess present-day and future climate change impacts on glaciers; (ii) close the sea-level rise budget; (iii) assess the threat on water resources and (iv) facilitate research on natural hazards related to glaciers. Our results specifically provide a strong observational basis that holds a great potential to further our understanding of the multi-scale morphologic and climatic drivers of glacier mass change, essential to improve physically-based glaciological modelling and calibrate future projections.

How to cite: Hugonnet, R., McNabb, R., Berthier, E., Menounos, B., Nuth, C., Girod, L., Farinotti, D., Huss, M., Dussaillant, I., Brun, F., and Kääb, A.: A globally complete, spatially and temporally resolved estimate of glacier mass change: 2000 to 2019, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20908, https://doi.org/10.5194/egusphere-egu2020-20908, 2020

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 1 – uploaded on 06 May 2020
  • AC1: Discussion starter - Global glacier mass change 2000-2019, Romain Hugonnet, 06 May 2020

    In this exercise, we use spatiotemporal statistics to harness the potential of massive satellite stereo archives and globally assess glacier surface elevation change at high spatiotemporal resolution for the first two decades of the 21st century.
    Curious about the data, the mass processing, the statistics, the validation, the limitations? Or about our estimates of glacier change in a specific region, at a specific time? Don't hesitate to leave a comment and start a discussion!

    • CC1: Reply to AC1, ruitang yang, 07 May 2020

      Dear Romain,

               Thanks for sharing this great job. 

    • CC2: Reply to AC1, ruitang yang, 07 May 2020

      Dear Romain,

               Thanks for sharing this great job. Interesting results in the Karakorum. 

               I have a few questions concerning the interpolate and smooth methods of elevation change maps:

              How do you deal with the data void of the DEM? Did you use any interpolation method, e.g. local mean hypsometric method or any others mentioned in the McNabb and others (2019), or another one else?

               If using interpolation, which method has you used to smooth the elevation change map?  The maps showed in the presentation are smooth and beautiful.

               Best regards

               Ruitang

               P.S. It will be appreciated if you can share a snapshot of glaciers in the south-central of Alaska.

      • AC2: Reply to CC2, Romain Hugonnet, 07 May 2020

        Hi Ruitang,

        Thanks for you interest!

        We do not perform any spatial interpolation on the elevation maps displayed in the presentation. Those are the raw outputs directly extracted from our elevation time series.

        The local hypsometric method (McNabb et al., 2019) is used only to account for data gaps when integrating elevation change into volume/mass change.

        The smoothness of our elevation change series (and thus elevation maps) is due to several factors:

        - The outlier filtering method (with reference TanDEM-X, then with Gaussian Process regression), that harnesses the repeat temporal coverage to exclude unreliable observations

        - The interpolation method (again with Gaussian Process regression), that mitigates the seasonal signal (which is the main reason elevation differences on stable terrain are so close to 0), but can also enhance the vertical precision if repeat observations closeby in time can be used (e.g. 3 independent observations in 2010 with ~5-10m precision will actually yield an estimate with improved precision, maybe something on the order of ~3m).

        Hope this answers your question,

        Don't hesitate to send us an email with the specific location/period that interests you in Alaska !

        Best,

        Romain

  • CC3: Comment on EGU2020-20908, Bas Altena, 07 May 2020

    Dear Romain,

    A great effort you present here!

    In your slides you refer to a whole book, so it is difficult for me to deduce what you do with the filtering/fitting. Hence, I wonder what you do with the "seasonal" function. Is this a cosine, and if so, can the phase angle be changed (in general or local)? Is this low amplitude signal really needed, as the seasonal signal of snow cover might not be captured by ASTER? Lastly, coming back to the phase angle of the seasonal function, how do seasonal signals that are reversed perform in the fitting. For example, the work of Ian [D2576 https://meetingorganizer.copernicus.org/EGU2020/EGU2020-12528.html] shows a speed-up in winter in addition to a summer speed-up.

    Keep up the good work, best! bas

    • AC3: Reply to CC3, Romain Hugonnet, 07 May 2020

      Thanks Bas,

      A lot of points to answer!

      1/ Seasonal function: we only condition the temporal covariance of glacier surface elevation with a periodic component. The mean elevation in the time series is not directly conditioned by any seasonal function, but propagated independently from observations for each pixel based on available observations.

      More details: Prior to GP regression (can also be pictured as kriging), we identify different components in our covariance (by drawing temporal variograms from all our raw DEMs stacked in time). One of these component that we isolate is the periodic signal, that exhibits a variance of a classical periodic kernel, i.e. an Exp-Sine-Squared kernel. We estimate this kernel at a 1-year periodicity and a certain variance amplitude. When we apply GP regression, by looking for variance minima, the method will capture a periodic signal if it is significant in front of other components (nonlinear, local) and enough observation are available.

      2/ It is absolutely needed. As an illustration, it is one of the main reason our stable terrain looks this nice (very limited seasonal bias despite all these varying timestamps!). We do capture amplitudes of snow-cover on glacier all over the globe (brought out by comparison to ICESat/IceBridge), although those are still subject to large systematic biases (co-registration on snow/with snow). Lots of supplementary work required to exploit these aspects.

      I'm not sure I understand the last question completely, hopefully the other 2 points should already provide sufficient insights!

      Thanks again!

      Romain

  • CC4: Comment on EGU2020-20908, Douglas Hardy, 07 May 2020

    Romain et. al.,

    What a fantastic project! Thanks for your eagerness to share results. I would appreciate seeing results from the Andes, specifically Quelccaya Ice Cap (13.93°S / 70.82°W) in the Cordillera Vilcanota of Peru. I am interested in the longest period available, but 2010-2014 if only one.

    Thanks! Doug Hardy

  • CC5: QUICK QUESTION, Luca Maffezzoni, 07 May 2020

    Hi Romain! Excellent presentation. Just a quick question. in your analysis did you consider glacier mass change due to runoff or both runoff and dynamic mass loss?

    Thank you in advance for your reply.

    Luca

    • AC4: Reply to CC5, Romain Hugonnet, 07 May 2020

      Hi Luca,

      Thanks !

      I am not sure to fully understand the question. Maybe this can clarify:

      We assess the mass change as in classical DEM differencing "geodetic" methods, i.e.  by differencing the surface elevation of the glacier between 2 epochs into volume and converting to mass.

      In these approaches, we don't differentiate between the processes that lead to the volume change. All are included.

      I hope this helps!

      Cheers,

      Romain

  • CC6: Comment on EGU2020-20908, William D. Harcourt, 07 May 2020

    Hi authors, great job indeed! I was just wondering about timeline of data availability?

    Best wishes,

    Will

    • AC5: Reply to CC6, Romain Hugonnet, 07 May 2020

      Hi Will,

      I already answered by email, but I'll repeat it here in case others are curious as well:

      The complete dataset and results will be made available at publication. We don't know exactly when that will be. The work is finalized, so we hope it won't be too long!

      Best,

      Romain

  • CC7: Comment on EGU2020-20908, Fei Li, 20 May 2020

    Dear Romain,

    This is really a great work! I'm wondering if I can apply for the data for 8 glaciers in Tienshan, HAM region. The glacier information is following:

    Number Longitude(East) Latitude(North) RGIId
    1 88.32543524 43.83247214 RGI60-13.45233
    2 86.80970149 43.1097081 RGI60-13.45335
    3 86.80237639 43.11691971 RGI60-13.45334
    4 84.39218173 43.72932741 RGI60-13.47247
    5 77.08058731 43.043653 RGI60-13.08624
    6 88.35610762 43.78329611 RGI60-13.48211
    7 79.8943851 41.78013072 RGI60-13.43165
    8 78.1813488 41.82697931 RGI60-13.08055