| 22 Oct 2026
Rain-induced speed up events of a temperate surging glacier, AK, inferred from GPS measurements
Abstract. Temperate surge-type glaciers undergo periodic order-of-magnitude increases in flow speeds generally attributed to drainage system evolution, making them ideal sites to explore how the kinematic response of glaciers to rain events varies with subglacial drainage system efficiency. Here, we compare ten GPS-derived glacier surface speed records with modeled rainfall to assess short-term evolutions of the drainage system at the end of a surge (2020–2021) and during quiescent flow (2022–2024) at Sít’ Kusá, Alaska. Speed anomalies were identified using a short-term to long-term average ratio approach and all speed anomalies that followed large rain events were characterized by their duration, magnitude, and immediacy. During the surge, 9 of 16 large rain events (56%) caused speed anomalies, peaking at 7.5 m/d (116% increase) and lasting 66.5 hours on average. During quiescence, rain-induced speed-ups were less frequent (17 of 37 rain events (46%)), smaller (~1.5 m/d), shorter (~33 hours), but greater in percent increase (251%). Although the differences in speed characteristics generally indicate that the drainage system is more efficient during quiescence, speed-ups throughout the surge cycle suggest there is a persistently inefficient component of the drainage system. We find that particularly large and prolonged rain events can cause temporary slow-downs during the surge, but that speeds return to pre-event levels in less than two weeks as the drainage system reverts to its inefficient state. Our work supports recent analyses of glacier instabilities suggesting there isn’t a binary switch in drainage system configuration during surges. More broadly, our findings expand our understanding of glacier kinematic responses to rainfall, demonstrating that rain events can modulate speeds of both fast- and slow-flowing glaciers.
Summary
The authors present a very valuable dataset with an interesting research objective to investigate the correlation of precipitation events and glacier flow anomalies during both surging and quiescent phases of a temperate glacier. Before making any critical comments, I want to acknowledge the multifaceted physical challenge of obtaining GNSS measurements on a dynamic glacier in a climatically challenging environment, as well as the difficulty of geophysical signal processing as applied to GNSS data. I also understand the absence of local meteorological data and appreciate the need to make do with existing data products.
Below I outline the major issues, which largely relate to the methods and their presentation, with details provided as line-by-line comments in the annotated pdf. These comments are themed around wanting more robust assurance of signal-to-noise ratio in the GNSS records and therefore interpretability/interpretation of the data. Some of the comments can be addressed by a more fulsome description of the methods, while others will probably require some additional work.
I found the paper to be well written overall, with the text generally clear and free of errors.
General comments:
With the above appreciation for the difficulty of making and interpreting these measurements, I think significantly more detail is required in the main text, beginning with some description of the field installation of ICE vs ROCK GNSS stations. How were the station installed? How were they secured (e.g. tower bolted to rock for off-ice stations)? What was the condition of the instruments upon retrieval, for example, was station tilt measured? In my experience, station tilt can be a serious impediment to interpretation of glacier speed-ups. To address these details, I recommend adding a short subsection between Study site (2.1) and GNSS processing (2.2) to describe installation and related details. It would make sense if this section were to also include the first paragraph of the current 2.2 and Figure 2, since these arguably do not describe processing, but rather what the reader needs to know before processing.
Greater detail is also required in describing the processing workflow. I have questions about why the low-pass filtering and detrending were done separately, why the detrending was done with minimum instead of central values (seems to assume anomalies will always be positive) and the justification of the choice of filter characteristics (corner frequency) on the basis of the ROCK data. See details in annotated pdf.
The biggest criticism I have of the GNSS processing is that no uncertainty assessment appears to have been done (aside from that associated with choosing the filter characteristics using the ROCK data) and no uncertainties are reported on velocities. I would consider uncertainty assessment in some form a non-negotiable revision to the work that is needed to establish confidence in the interpretation of velocity anomalies. Data from the ROCK station provide an avenue to assess some of the uncertainty.
This is another hard problem, but the methods could be more robust and their presentation improved.
First, why was ERA5 chosen as the data source over other reanalysis products? This is a fine choice but should be explained. Next, why was only one ERA5 cell used for the downscaling (this is my interpretation from the figure), and moreover a cell that does not overlap any of the GNSS stations? I would assume the downscaling would have been tested on all cells overlapping the domain to examine the differences.
Please add a bit more information on the downscaling for those not familiar with the model used. What is the basis of the downscaling, e.g., physical versus statistical? Where did the precipitation lapse rate of 0.1/100 m come from? Is snow differentiated from rain? Here I think it is important to clearly state what source variable was used (i.e. was it precipitation or rain from ERA5?) and where/how in the process rain is differentiated from snow. Perhaps it never snow in the spatiotemporal domain? Whatever the case, please explain the procedure more clearly and any assumptions required.
Finally, I appreciate the comparison with station data (Yakutat), but it would make much more sense to me to execute the downscaling for Yakutat and compare to Yakutat, than to compare the apples of the downscaled precipitation over the glacier to the oranges of Yakutat obs. In so doing, the authors could more confidently attribute the temporal lag in the precipitation records currently compared to a true lag in precipitation between spatial locations versus a mismatch in phase between the reanalysis (ERA5) and station (Yakutat) data.
Lines ~140-180 describe a somewhat complicated process of identifying peaks in speed and peaks in rainfall and classifying events and then sensitivity. I wonder if a simpler approach might yield equally useful results, but require less ad-hoc manipulation of the timeseries. Why not just cross-correlate the precipitation timeseries and speed timeseries to identify peaks and lags in the cross-correlation function? If the timeseries are standardized, the amplitude of the cross correlation is a direct measure of sensitivity, assuming of course that precipitation is driving speed anomalies. The cross-correlation could be done on different segments of the timeseries individually to detect differences in lag which might provide some additional insight into the hypothesized drainage configuration.
Regarding the STA/LTA analysis, are these phase-preserving filters? I assumed initially they were, but then questioned this assumption when I saw the figure where the STA/LTA timeseries is clearly lagged relative to the original speed timeseries. Phase preservation seems important for the analysis since time lags between precipitation events and glacier acceleration are interpreted.
Details and technical corrections:
See annotated pdf for line-by-line comments.