General

Gillespie et al. present an extensive collection of ice thickness observations on Jostedalsbreen ice cap

obtained using ground penetrating radar (by scooter, by helicopter and by foot) between 2018 and

2023. They derive an impressive coverage of the ice cap with 90% of the area now being within

300 m of a data point. The point observations are used to calculate a distributed map of ice thickness

and bed topography based on a thickness inversion approach. Uncertainties are presented for both

the observations as well as the modelled thicknesses. The final ice volume of 70±10.2 km2 is closely

aligned with previously published estimates. The distributed bed and thickness maps are thought to

form a valuable input for future modelling studies of the ice cap as well as for predicting landscape

change, e.g. with regard to future lakes and possible break-up points of the ice cap.

I would like to commend the authors for their efforts to obtain such a wealth and density of

thickness observations which may well be unique for an ice cap of this size. A transparent account

of error sources and related uncertainties is evident throughout the manuscript and lends trust to

presented results. The manuscript is well written and clear, with appropriate figures and tables. The

provided data is easily accessible and well documented. All in all, the significant improvements in

coverage and accuracy compared to the previously available thickness observations of Jostedalsbreen

surely warrant publication in ESSD. I have only minor comments with the exception of one related to

the inverse modelling approach.

2 Specific comments

2.1 Inverse modelling

The inverse approach chosen is based on Huss and Farinotti (2012), a method which reduces glacier

shape to a flow-line and is often thought to not be the ideal tool for ice caps (e.g. its comparatively

low performance there is already discussed in paragraph 25 of the original publication, but also in,

e.g., Millan et al. (2022); Frank and Pelt (2024)). Previously published ice thickness maps of ice

caps based on this method (e.g. Farinotti et al., 2019) are not seldom characterized by unrealistic

bed shapes and issues at ice divides. Regarding the latter, the authors here aim to alleviate the

problems by resorting to an iterative procedure that involves interpolation of nearby GPR profiles and

”estimated thicknesses on ice divides based on local knowledge” - doesn´t sound entirely convincing.

Nevertheless, this strategy is helpful when looking at the output, yet there remain small inconsistencies

at glacier boundaries. Regarding the overall appearance of the bed shape and thickness distribution,

after plotting in QGIS, I find that it does not look realistic. For sure, the thickness observations are

well matched, but overall clear ”stripes” are visible that appear as if someone had drawn with a thick

pencil. So, my general question is: why did the authors settle for this approach and not for other

methods on distributed grids, possibly even one that is specifically designed for assimilating thickness

observations (e.g. Morlighem et al., 2017; Fürst et al., 2017; Jouvet, 2023)? Looking at the modelled

thickness field, I have doubts that the flux divergence would look somewhat realistic when put into a

distributed ice flow model. This could turn out to be an issue if, as suggested in the introduction, this

bed shape is used for prognostic simulations. In fact, is the final thickness distribution mass conserving in any way? For the plausibility of the location of future lakes, the unrealistic bed shape also doesn´t

help (although I am aware that there is always a relatively large uncertainty for such a product).

Besides this general comment on the choice of inverse method, I find that section 3.7 in parts is

unclear and should be revised (c.f. specific comments). Simply, while I understand that the method

has already been presented in other publications and hence some omission of details is justified, I

currently don´t find myself able to understand all steps taken. Not the least, more information on key

model parameters should be presented .

2.2 Other specific comments

L36f: Slightly re-formulate to make clear that future studies can use the outputs from this work for

climate change impact studies. This is not done here.

L40: There are also other reasons for glacier mass loss than atmospheric temperature increase (e.g.

increased ocean temperature for marine-terminating glaciers). Please re-phrase.

L 52: A newer version of the GlaThiDa with continuous additions of published thickness obser-

vations is available under https://gitlab.com/wgms/glathida. Consider referring to updated num-

bers from there as well. More generally, besides publishing the data on the repository given in the

manuscript, would you consider actually feeding it into the GlaThiDa?

L 169f: Here and in other places, the helicopter radar system is described as new/newly developed.

However, not much details regarding its novelty are given. So, what exactly are the novelties? And has

the system been thoroughly tested beyond what is mentioned in L 365ff? Also, what was the travel

speed?

L 274: You mention that you could infer information on glacier thermal regime from the collected

GPR data. Was this done systematically, and if so, would it be worthwhile adding more information

on that in the manuscript? Could be quite interesting!

L 434: The inversion method was originally developed for global applications and was relying on

some large-scale input products and coarse assumptions to infer parameters such as basal sliding, ice

viscosity and apparent mass balance. Was this global setup used to infer these parameters here as

well, or were they based on more local knowledge/input products (e.g. the mass balance observations

on Jostedalsbreen)? Also, in Huss and Farinotti (2012), one basal parameter per glacier was chosen,

not a longitudinal trend as stated here. Please state or provide a reference on how the basal sliding

distribution was derived.

L 444: Please state that this integrated form of Glen´s flow law assumes parallel flow and thus in

essence is the shallow ice approximation. In my view, this information is important for readers to be

able to evaluate potential errors of the model output.

L 449: Could you please comment on your choice of inferring ice thickness on a 10 m grid? There is

a physical limit to how much spatial detail one can obtain from a thickness inversion, at best horizontal

features of >1 times the ice thickness can be resolved (e.g. Gudmundsson, 2003). Isn´t the choice of a

10 m grid unnecessary? Or even, doesn´t it give the wrong impression that such small features should

actually be detectable in your thickness map? You later mention that you smooth the bed topography

(but not the thicknesses, why?) with a 50-100 m filter which, however, still in many places is lower

than one ice thickness.

L 463ff: Here you mention that the apparent mass balance gradient is optimized to reduce the misfit

to the thickness observations. In the next sentence, there is talk of two gradients that are subsequently

computed. Please clarify the discrepancy on the number of gradients. Furthermore, it is not clear

to me why the apparent mass balance is first optimized, and then calculated again. What exactly is

done when it is calculated after the optimization? Lastly, you assume a balanced mass budget over the

entire glacier unit. Instead of an assumption, isn´t that a necessity that follows from mass conservation?

L 473ff: Is the difference field smooth? It sounds as if that wouldn´t necessarily be the case, e.g.

one observation point may be well matched, whereas the neighboring one isn´t. How does that affect

the extrapolation? And why is the ice thickness distribution after this step not going through the

observations, if you apply point-wise corrections? And on a different note: I am somewhat confused

about the terms extra- and interpolation (here and in other places of this section). Please clarify what

you mean when you interpolate vs when you extrapolate, and between what or where to you do that.

L 479ff: Here I cannot follow any longer. I take that after step ii, you already have a thickness

distribution that is based on all available thickness observations. However, now again thickness ob-

servations in combination with model results are used for some more interpolation. Why and how is

that done? Where do you interpolate between? What happens within the buffer? Do you simply set

the thickness there to the observed one? And ultimately, is your final thickness distribution consistent

with the ice dynamics and apparent mass balance of the inversion? If not, what distinguishes your

approach from a statistical interpolation?

L 484: Please provide more information on what is meant by ”local knowledge” and how that

information was transformed to quantitative values.

L 516: Another way of assessing uncertainties would be to remove some of the thickness observa-

tions from the data set, re-run the inversion and test how the resulting thickness field compares at

those spots. Could that be worthwile?

L 594f: You mention that the old data set has considerable uncertainties in many places, and thus

you limit your comparison to an area which is more certain. I would argue that there is no reason

for that. Indeed, to get a full understanding of how much better your new data set is plus to allow

users of the old data set to judge how ”bad” the old data was, I think it would be meaningful to

undertake a full comparison. A simple scatter plot of old vs new thicknesses (at places where the two

are reasonably close together, or using your interpolated bed map) would already be helpful in my view.

3 Technical comments

L 26: I think adding a reference for the GlaThiDa is appropriate here, as not all readers can be assumed

to know what that is

L 62: Fürst et al. (2017) actually assimilates thickness observations, so should not be mentioned

here

L 74: remain, not remains

L 141 ”continues to this day”: consider rephrasing, e.g. ”continue to do so to this day”

L149: In contrast to other place/glacier names mentioned in the text, Bødalsbreen is not labeled

on the figures

L 629: Consider including also a comparison to Farinotti et al. (2019) and Millan et al. (2022)

4 Figures and Tables

Table 2: In the GlaThiDa, the attribute field for elevation is ELEVATION, whereas it is GNSS ELEVATION

here. For simplicity of usage, consider aligning with the GlaThiDa

References

Farinotti D, Huss M, Fürst JJ, Landmann J, Machguth H, Maussion F, Pandit A. 2019. A consensus

estimate for the ice thickness distribution of all glaciers on Earth. Nature Geoscience 12:168–173.

doi:10.1038/s41561-019-0300-3.

Frank T, Pelt WJJv. 2024. Ice volume and thickness of all Scandinavian glaciers and ice caps. Journal

of Glaciology :1–14doi:10.1017/jog.2024.25.

Fürst JJ, Gillet-Chaulet F, Benham TJ, Dowdeswell JA, Grabiec M, Navarro F, Pettersson R, Moholdt

G, Nuth C, Sass B, Aas K, Fettweis X, Lang C, Seehaus T, Braun M. 2017. Application of a two-

step approach for mapping ice thickness to various glacier types on Svalbard. The Cryosphere

11:2003–2032. doi:10.5194/tc-11-2003-2017.

Gudmundsson GH. 2003. Transmission of basal variability to a glacier surface. Journal of Geophysical

Research: Solid Earth 108. doi:10.1029/2002JB002107.

Huss M, Farinotti D. 2012. Distributed ice thickness and volume of all glaciers around the globe.

Journal of Geophysical Research: Earth Surface 117. doi:10.1029/2012JF002523.

Jouvet G. 2023. Inversion of a Stokes glacier flow model emulated by deep learning. Journal of

Glaciology 69:13–26. doi:10.1017/jog.2022.41.

Millan R, Mouginot J, Rabatel A, Morlighem M. 2022. Ice velocity and thickness of the world’s glaciers.

Nature Geoscience 15:124–129. doi:10.1038/s41561-021-00885-z.

Morlighem M, Williams CN, Rignot E, An L, Arndt JE, Bamber JL, Catania G, Chauch´e N,

Dowdeswell JA, Dorschel B, Fenty I, Hogan K, Howat I, Hubbard A, Jakobsson M, Jordan TM,

Kjeldsen KK, Millan R, Mayer L, Mouginot J, No¨el BPY, O’Cofaigh C, Palmer S, Rysgaard S,

Seroussi H, Siegert MJ, Slabon P, Straneo F, van den Broeke MR, Weinrebe W, Wood M, Zin-

glersen KB. 2017. BedMachine v3: Complete Bed Topography and Ocean Bathymetry Mapping

of Greenland From Multibeam Echo Sounding Combined With Mass Conservation. Geophysical

Research Letters 44:11,051–11,061. doi:10.1002/2017GL074954

Please refer to attached PDF.