| 12 Jun 2026
Temporal variability of measured vertical velocity over a canyon-incised slope in the Coral Sea
Abstract. Oceanic vertical velocity (w) is often neglected due to its small magnitude compared to horizontal currents. However, it plays a fundamental role in coupling the ocean surface with its interior, with major implications for a wide range of physical and biogeochemical processes. In this study, four ADCPs were deployed on fixed Eulerian moorings and collected w measurements over slightly less than one year, covering the water column from 50 to 410 m depth. Using a combined approach based on Fourier spectral analysis and Ensemble Empirical Mode Decomposition, the variance distribution was characterized. The spectral analysis highlights the fundamental anisotropy of the flow, as vertical and horizontal velocities exhibit distinctly different spectral properties. While horizontal currents are mainly constrained by topography, vertical motions are primarily driven by internal gravity waves and intermittent small-scale processes. Furthermore, a persistent w signal at synoptic scales shows that the velocity is not purely geostrophic at all temporal scales. The results show that the variability of w time series is primarily dominated by short timescales, on the order of a few days or less. The data are also strongly influenced by biologically driven diel vertical migrations, which impose a pronounced periodic signal. The overall dynamics of the study area, strongly influenced by a canyon-incised slope topography, is characterized by a prevailing weak downwelling regime (3−4 mm s-1), intermittently disrupted by short-lived (few days) strong upwelling events (exceeding the 90th percentile of 1-day moving averaged w data). Finally, our results suggest that the canyons could act as an efficient tracer sink (e.g., carbon), with vertical velocities reaching several mm s-1.
Robache et al. discuss vertical velocities measured at 50–400 m depth from four moorings over a sloping seafloor deployed for 11.5 months. They have a meaningful goal of trying to describe the various timescales at which vertical velocity varies. This has direct consequences for biology and biogeochemistry. The paper is organized well and the figures that are present are well designed.
Unfortunately, on account of the data contamination due to diel vertical migration (which the authors recognize but cannot sufficiently fix), I think the paper lacks clear results. Several attempts are made in the Discussion to interpret the results within existing theories, but not in a convincing way (more details below).
To turn this study into a publishable paper, my suggestion would be to simplify things. Spend more time presenting the data in its primary form (e.g., plots of velocity vs time) rather than abstract and derived quantities (e.g., IMFs from the EEMF or Q values). Also, lean more toward simple scaling analyses like that in Section 4.1 rather than the much more complex ideas in Section 4.2.
Listed below are comments separated into major and minor issues. Some of these elaborate on my points above.
Major:
A key selling point of the paper is that there is an 11.5 month dataset of vertical velocities at four moorings. Yet at no point is the whole record shown in a simple manner. Most of the figures are abstract quantities like PDFs, power spectra, cross-correlations, etc. It's difficult to interpret these without seeing the original w data. And certainly it makes it harder to provide a useful review. I would suggest adding a figure near the start of the Results that shows the whole record for at least one of the moorings. Something like the current Figure 2a but for all 11.5 months. You could break it into multiple rows to get a reasonable aspect ratio.
Another key selling point is the multi-time-scale analysis, but I'm skeptical of its value. Figure 6 (and line 265) says that 90% of the variance in the signal is captured by the lowest three modes. But these modes are the 24-hour variability and higher harmonics. Elsewhere in the paper, it's clearly shown and recognized that this variability is biological (diel vertical migration, DVM), not physical. I realize that you state somewhere in the paper that you have to use w_raw, not w_processed, for methodological reasons. But it makes the analysis seem worthless. You've identified that the biological signal is unwanted in this analysis, but then you've gone ahead and ignored this major caveat in a lot of your analysis.
When the harmonics are first discussed at line 232, it doesn't appear to be recognized that these are expected. Consider this: idealize the diel vertical migration as a negative dirac delta function in the morning and a positive dirac delta function in the evening, and then repeat this every day to create an idealized time series. The power spectrum of this time series would include a peak at 1/day and higher harmonics (i.e., the 12, 8, and 4 hour peaks identified here). To me, this seemed obvious as soon as I saw it. The fact that the authors don't seem to recognize this reduces my confidence in the rest of their interpretation of results involving power spectra.
The harmonic peaks are also mentioned for the quantity dI at line 243, but again they are not given their simple explanation.
How can the power spectra in Figures 4 and 5 extend below 0.001 per day? The record is only 356 days, which means the lowest frequency should be 1/356 = 0.0028 per day.
Related to the above comment, Figure 4's caption says spectra are 'smoothed using a mean per decade' (see also line 228). I do not understand what this means. I would've thought it is a moving mean with a window that has a log_10 width = 1, but that would smooth out all the detail, and it's clearly not what the graphs show
Figure 5. It's concerning that the w_processed spectra is above the w_raw spectra across a large band (all of f > 0.03 per day with the exception of the 1 per day peak and its harmonics). w_processed is meant to be w_raw with the effect of diel vertical migrations subtracted. How does this lead to an increase in variance for w_processed relative to w_raw for the range of frequencies in question?
240. Maybe it's okay, but it strikes me as odd to calculate a spectrum of speed V (i.e., a quantity that's always positive), when you could have just as easily calculated a spectrum for u_along.
247. How do you know that what remains after the DVMs have been removed is still a biological signal?
259. Figure 6 presents the Intrinsic Mode Functions from the Ensemble Empirical Mode Decomposition. These aren't things commonly used in oceanography, so I would suggest you need a figure prior to Figure 6 that demonstrates the decomposition. The reader cannot make sense of the relative variance of the IMFs if they don't understand what its components are. Although you define everything in Section 2.2.2, there are only formulas. A visual is needed as well.
Section 4.2. A lot of this section seems hand-wavy. Take lines 393–401, for example. You discuss a single number (the spectral slope β) in relation to several processes (three-dimensional energy cascade, geostrophic eddies, scales of energy injection, Heisenberg eddy viscosity, bottom boundary shear). Everything is mentioned too quickly without any real justification. The same kind of thing happens at lines 404–415 when you look to explain β=0.3–0.4
511–515. Again, there's too much hand waving going on. Take the sentence 'Such structures point toward the importance of secondary circulation, internal tides, baroclinic processes, and topographic interactions, extending beyond the classical Ekman framework.' That's a list of processes that incorporates too many things to be useful to the reader. This is soon followed by another long, wide-ranging list: 'Instead, they emerge from the interplay between topography, stratification, multiscale variability, and episodic processes.'
As most of Section 2.2 is descriptions of textbook statistical analyses (except perhaps 2.2.2), most/all of it could be moved to an appendix. A lot of it also seems unnecessarily detailed, especially for the standard methods used in oceanography like Fourier transforms, cross correlations, and moving averages.
Minor:
4. Replace 'slightly less than one year' with the actual length (11.5 months).
Figure 1. The labels for the arrows in the middle panel are too small to read
178. 'Data presentation' is a very vague heading. Replace with something more insightful
184. Instead of 'An example is shown in panel (a) of Figure 2', you can just add '(Figure 2a)' to the end of the preceding sentence.
188. As above, replace 'panels (b), (c), and (d) of Figure 2' with 'Figure 2b–d'
More generally, there are many places in the text where you use phrases like 'panel a' and panel b'. It would be simpler to just use Figure 1a, Figure 2b, etc. It possible to lose track of what figure you're referring to (e.g., at line 218), and having a reference to only panel a or panel b doesn't help.
Figure 4 caption. The vertical lines are described as being at the Nyquist frequency, which you've correctly defined in the caption, but why are they are at f ≈ 0.005 per day? The Nyquist frequency would just be the upper frequency limit of your power spectra, so there's no need to draw a line.
205. It's weird to say the 'NW, SW, and SE sites appear to be more affected' because there's only one other site. To say three of the four are 'more affected' seems more complicated than saying the other one is less affected.
209–211. Define what metric is being used for the uncertainties. Is it standard deviation? Is it 95% confidence interval?
209. It seems odd to say the NE site is 'larger and more variable' when its mean and uncertainty are nearly the same as the other sites.
212 and 502. 'persistent' seems like the wrong word. Maybe 'average' or 'net' describes the vertical motion? Persistent would imply it's always negative.
Figure 6a. This fit doesn't look that good to me. The real measure of how good this curve fits the data is whether it goes through the cloud of points for C2–C4, which it does not.
Table 1 caption. What is considered equal to 0?
231. Same comment as for line 205 about 'more' for 3/4 sites.
232. Peaks at higher frequencies are not 'subharmonics'. They're harmonics or higher harmonics.
232. 'f ≈ 10^-2.56', not 'f^-2.56'
242. The peak described as 'less pronounced' is still very pronounced
265. Related to the Figure 6a comment above, I disagree that the 'scatter plot follows the form of a power-law relationship'. Hence, I also don't see the value in fitting 'an asymptotic saturation model'
270. The R^2 of 0.96 is less impressive than it sounds. The data going into the calculation are certainly not independent given there is a cumulative sum involved
288. I'm unclear how you can meaningfully compute the cross-correlations for the highest mode IMF (i=11, top right squares in Figure 7). As stated in line 288, the associated time scale is ~1 year. But if that's the same length as your whole record, how is it meaningful?
296. Remind the reader that Q was defined back in Section 2.2.4.
328-339. This recognition of the ADCP uncertainty is welcome, but what about uncertainty in how vertical the ADCP is? For example, if the ADCP was tilted from vertical by 1°, then w = sin(1°)×u = 0.017u, which isn't necessarily negligible. Is this an issue, or did you ensure that verticality somehow?
433. K1 has a period of 23.93 h, not 24.96 h