| 16 Sep 2026
Integrated satellite monitoring and field validation of the periodically outbursting glacier-dammed lake Nedre Demmevatnet, Norway
Abstract. Glacial Lake Outburst Floods (GLOFs) present a significant hazard in warming alpine environments, but detailed, sub-seasonal studies of ice-dammed glacier lakes remain rare due to data scarcity. To overcome this challenge, we reconstruct nearly a decade (2016–2025) of drainage timings, outburst volumes, lake levels, and automated, machine learning-based sub-seasonal lake refilling cycles at the ice-dammed lake Nedre Demmevatnet (southwestern Norway), dammed by the glacier Rembesdalskåka, an outlet glacier of the Hardangerjøkulen ice cap. By integrating publicly available satellite and meteorological datasets, we evaluate remote sensing capabilities and establish an error budget for tracking a small and highly dynamic water body. We validate our spaceborne findings using field measurements, including water-level loggers, time-lapse cameras, a local automatic weather station, and high-resolution UAV photogrammetry, complemented by PlanetScope imagery. Combining Sentinel-1 and Sentinel-2 imagery, we constrain GLOF drainage windows to ± 2 days. Lake volume sensitivity tests revealed that while satellite outline errors are minor (2.6–4.2 %), using the static regional digital elevation model ArcticDEM (2014) causes a 25.5 % volume underestimation compared to our 2022 UAV bathymetry due to rapid ice-dam retreat and lakebed erosion. Crucially, the time gap between the last cloud-free satellite image and the GLOF introduces a relative daily volume underestimation of 1.7 %, which scales up significantly during cloudy periods. Our validated multi-sensor remote sensing approach enables valuable glaciological insights, revealing that GLOF timings shifted earlier by an average of 10 days, alongside shortened refilling periods over the past 9 years. Between 2016 and 2022, pre-GLOF lake levels reached 1236–1239 m a.s.l., closely matching the theoretical hydrostatic flotation threshold. In contrast, the 2023 event drained at approximately 1224 m a.s.l. – more than 10 m below this threshold – following a shortened melt period indicated by fewer positive degree days, possibly due to opening of subglacial channels through melt. Finally, we synthesize the workflow developed and tested on our case study into an operational framework that facilitates transferability to other rapidly changing ice-dammed lakes.
This study presents a detailed automated remote sensing workflow for using (mostly) freely available data to monitor ice-dammed lake height, alongside some field data to validate the method. This type of automation has significant potential for increasing our understanding of lake drainage dynamics, because it paves the way for time series analysis to be performed on the drainage dynamics of a much larger number of lakes than is possible using purely manual methods. The finding that temporal ambiguity in drainage dates can be resolved to +/- 2 days is very encouraging. Uncertainty in volume estimates is a difficult problem related to the availability of recent bathymetry data (a challenge for continuously retreating glaciers), but it should be noted that lake height data are well-resolved using this method, and height data are extremely useful (relevant for flotation drainage, and also in this case artificial drainage). As you will see in my more detailed comments, I think more could be made of lake height data, both in terms of the figures, but also in the conclusions. I very much enjoyed reading this work, which is well conceived and well written. I have some minor points on the technical side of the work, and some more substantive points that I think need some clarification on the interpretation and presentation of the data. Notwithstanding these queries, I think this is a very important and rigorous study which should be published.
With apologies for suggesting citing work I am a co-author on, I would recommend reading and citing a recent paper by Livingstone et al. (2026), which includes some satellite-based analysis of ice-dammed lake heights and volumes using a manual remote sensing approach similar, although less sophisticated, to that presented here. They also note changes in lake height over time related to glacier thinning, which is relevant to this study (see below). I am not aware of many studies similar to this, so it does seem like a useful one to cite in this paper!
The remote sensing method is very rigorous and is described well. The figures are mostly excellent and support the text. The data validation is thorough and demonstrates that the technique works very well, albeit with limitations on things like cloud cover, which simply cannot be overcome.
I found the technical aspect of this work to be very strong: I just have two queries related to this:
I have some more significant queries on the interpretation of the data, and how this is presented, particularly in section 5.3 and Figure 12.
I think section 5.3 needs some clarification. Figure 12 in particular presents a challenge to process interpretation, since it appears that melt becomes irrelevant to lake filling and drainage over time, and I don’t think this is fully explained. I suggest two changes to Figure 12 to aid interpretation and clarity in the paper: (1) replace the lake area data with lake height and volume; and (2) add the logger lake height data from 2023 and 2024. Lake height is important because of the artificial drainage and also flotation, so it would be very useful to see it plotted as a time series. Lake area is very difficult to interpret on its own, because the area of a lake doesn’t impact the drainage behaviour – rather it is the volume and height which do, and area scales non-linearly with height/volume depending on the bathymetry. This also applies to Figure 9, although it is more useful to show area here because that is what is being directly measured by the remote sensing methods, but I think it is important to show height/volume too. There are several mentions of height in the text of section 5.3 that suggest this is useful for the interpretation, but it is difficult to follow this and assess it without seeing it in a figure. It would also be useful to see the lake height data from the continuous monitoring in 2023 and 2024 alongside this since these data have much better temporal resolution that may help in understanding what happened in the June drainages in those years. As it stands, Figure 12 makes it look like there is basically no lake by 2025, which I find confusing.
Also in section 5.3, the interpretation of the increasingly earlier drainage trend as being related to warming and thinning (lines 667-70) doesn’t make sense, because the degree day modelling shows drainages occurring after little/no melt. The drainage mechanism transitioning from flotation to siphoning by the subglacial system around 2022 makes sense as an explanation for the lake draining more regularly at lower levels, but it isn’t clear how this is related to warming and thinning. The thinning part of the explanation is also contradicted in the finding that the flotation threshold didn’t change much due to the glacier retreating into deeper parts of the basin. I think this part needs some strengthening.
On lines 671-3 I am not sure which period you are referring to, and it isn’t clear exactly what you mean by there being almost no refilling. Do you mean to a particular level? Was there almost no lake? This is why it is important to show lake height on a time series!
Another difficulty I have in following this is that on line 716 you mention that the lake reached the flotation threshold in 2024. Figure 9 shows only small lake areas in this year, which presumably are scaled with much lower lake levels. The small lake areas are apparent quite close before the drainage date. Did the lake grow larger than the remote sensing was able to detect? What do the pressure logger data show?
Overall, I found the discussion in section 5.3 hard to follow. I don’t necessarily think your interpretations are wrong – much of the theory described is very plausible – I just found it difficult to reconcile with the data as presented.
Minor comments:
Using “M” instead of “× 106” in several places in the text is unusual. I would suggest using conventional scientific notation instead.
Line 121: Missing C in units.
Line 188: It is preferable to use the term ‘uncrewed’ instead of ‘unmanned’.
Line 195: Is this 80% horizontal overlap, i.e. between survey lines? I assume the vertical overlap (from one image to the next) was something similar?
Line 200: What was the Total Error of the surveys?