The manuscript by Chen et al. presents a novel approach to river bedload monitoring using a phased acoustic array together with an accelerometer sensor built into a modified version of a Swiss Plate Geophone casing. The study aims to locate bed particles on the cover plate of the device and exploits the wave train content to estimate the particle size. The team presents laboratory aided tests of the principle and performance of the system and compares the results with earlier experiments using the “classic” Swiss Plate Geophone devices.

The study addresses a relevant and timely topic, suggesting indeed a possible step forward in the field of continuous bedload transport research, going beyond what existing approaches can deliver. The concept of using a phased array appears overall suitable to be enrolled for such a customised case. That said, I feel the study remains somewhat premature and incomplete in the methods and tests it presents and is likewise in a not very organised structure in terms of writing and figure design, which prevents the reader from clearly grasping the line of arguments.

I see a main unknown and potential flaw in the design to relate impact amplitudes with grain size while keeping the drop height for each particle somewhat constant. It is hard for me to imagine how under natural conditions there will be constant drop heights for particles of varying size. Or, the other way around, how would the approach successfully estimate the correct grain size if river bed particles hit the instrument with different fall velocities and overall differing horizontal velocity vectors? What about those particles that rather slide and roll instead of jumping? These fundamental questions are neither raised nor implicitly addressed by the study, neither in the experiment design nor in the argumentation.

In writing “somewhat constant” in the above paragraph, I wonder why the height of centre of mass of the particles above the plate was kept constant instead of keeping the full vertical drop height constant. This adds a further unnecessary (or unexplained) variable that changes the experiment results.

Another, major flaw arises from the inability to locate and quantify the effects of multiple particles being in contact with the plate. As nice as it is to isolate single impacts, under natural conditions the norm will be to experience multiple impacts during a bedload transport period, even during the time window for which the first pulse of an impact signal is collected. If it is impossible to reliably locate two or more sources at a time, it will also be impossible to extract reliable information on the particle size and hence the envisioned total bedload flux across the sensor plate. I might have missed some detail in reading the manuscript but from the results section (e.g. Fig. 7) is seems that the two synchronous impacts cannot be deciphered, as one would expect from the capabilities of a phased array setup, and as is discussed in the manuscript further down.

The text contains numerous grammatical glitches, equivocal wording occasions, unclear terminology, and cluttered graphics. I list some of those issues in my detailed comments but think the manuscript is not yet at a stage where a fine screen of such improvements makes sense. I suggest to pay due attention to improvements of the presentation quality of the study both in terms of writing and graphical art.

There are several occasions when the text mixes up theory, methods, results and/or discussion and hence requires reorganisation to be consistent. See details below for examples.

Many figures are awkward to read/interpret. Especially the insets of the reference data (Wyss et al.) make any comparison virtually impossible. Likewise, the semantics of the figures not always add up, which is when I would suggest to split figures to maintain consistent pieces of information rather than collections of fragmented information. See details for examples.

Detailed comments

l. 90-92, in the sensor description section it would help to give more detail on the device characteristics apart from the manufacturer and type ID, e.g. frequency range, sensitivity, digitizer sampling depth, and so on.

l. 108, 2.2 10^1 kg means 22 kg. Check values

Fig. 2, why are there so few validation drop locations and overall, why were not more locations used further towards the margin of the plate? It seem quite unlikely that only the central 2/3 of the plate would be impacted by moving bed particles, and the edge effects appear to be quite significant according to the study results.

The derivation/description of the beam forming technique seems to be or will at least be based on numerous references representing work of generations of scientists. Yet, there is only a single reference included. I suggest the authors give proper credit to the scientists that have built the physical concept, mathematical framework and perhaps computational implementation of the routine – unless all this has indeed been developed by the authors of this study.

Fig. 3 d, R_s needs to be defined

l. 169, the heading (simulation) does not depict what the text below refers to (construction details)

l. 171, 3D grid does not make sense, either it is a grid (2D) or a 3D structure (voxels, cube). Also, pixels is a grid-world term while nodes is a vector-based term. This (and other occasions) need to be corrected throughout.

l. 184, the equation provides rather trivial information, consider removing it

l. 194-196, this is more about phased array theory than actual methods employed in this study. Consider moving this to the introduction

l. 199-204, there are some references needed to support the statements in this section

l. 204-209, there is some rather trivial information that might be better placed in a general introduction (or just skipped) rather than being an actual topic of “Microphone elements arrangement”.

l. 213, why a sine wave and not the more likely to expect pulse signal? Can you explain/justify this? Will this decision not affect the validity of the test outputs with respect to the envisioned natural settings?

l. 214, why adding white noise? Can you justify, both why adding noise first of all and then also why specifically white noise? I would rather expect to see red noise or some derivative of it to better reflect “real” conditions.

l. 224, replace “centric” by “central”

l. 224, why indexing a repeat experiment run?

Fig. 6, it would help quite a bit if the panels a-d would indicate the locations of the sensors on the coloured matrices, instead of having to look for Appendix A to get that information.

Fig. 7, better split the figure into single impacts and multiple impacts. In addition, in (a) and (b) there are only two locations shown, in (e) there are three locations shown. Similar for c,d and f. Please resolve.

l. 309-310, I disagree with the statement that the locations agree well. This is just an artefact not a real two-source location. Actually we need proper tests here to show the real artefacts arising from the case of more than one impacting particle being relocated. What if two, three, four, … particles impact at the same time? What is the effect of different locations (up-down, oblique, different distances from center)? Here I only see two sources horizontally symmetrically impacting the plate.

l. 316, “indicating that the grain of the noises become weaker…”, this sentence does not make any sense to me

l. 324-326, this is interpretation and should not occur in the results section. Albeit, what have we learnt from that exercise after all? We already know that a phased array cannot locate two synchronous signals from two different locations. Also, what does that imply (later in the discussion) for a real world case when multiple impacts happen?

Fig. 8, any comparison with the inset data is not possible (also in figs. 9 and 10), please add those data points to the main chart, perhaps with clearly distinct colours. Perhaps add a small map that depicts the impact locations to prevent readers from searching back and forth to get the context right. Add the number of samples that are summarised in the box plots. Perhaps remove the x-axis line to make clear this is categorical data and not continuous, because at a first glance one could think the four box plots per diameter are a continuum of sizes. In summery, just make the job easier for readers to correctly interpret the figure.

l. 417-148, this should go to the methods section. In addition, why not just calculating the RMSE between real location and model derived location?

l. 422, “suggesting a good agreement”, first this is interpretation and second it is a weak statement that can and should be quantified.

l. 428, “boxes in green”, I see no green boxes at all

l. 429, what are these values (30 % and 10 %) based on? Are these arbitrarily chosen or based on some reference or some physical meaning?

l. 445-446, this sentence needs some context in which is shall be embedded. Currently, it hangs in the air with no logic

l. 466-470, but this is neither shown nor is this discussed in terms of its consequences for the ultimate goal, to get a reliable bedload flux estimate

l. 476-485, the discussion is weak here, it needs to go for more far reaching thoughts, beyond just interpreting possible reasons why this system deviates from earlier/other SPG devices. The title of the study tells us the new system improves the prediction of particle size for bedload measurements. Yet, this is not clearly presented.

l. 505-510, this part is fairly repetitive, it reiterates the results described earlier but lacks an interpretation

l. 513-536, this reads like methods description, certainly not discussion of results.

l. 534-538, this does not make sense. Ignoring friction any particle that falls from the same height would have the same impact velocity. This may leap back to my very above comment of using the centre of mass to define the fall height instead of the lower margin of the impactor.

l. 540-546, this is a weak interpretation approach, with many “may” and “might” terms but with very little rational arguments that would support the hypothesis. Either add experimental, reference or logical support or skip the attempt of interpretation.

l. 558-560, this is repetitive

l. 564-566, this is repetitive

l. 574, “the decrease of R_s tends to weaken”, what does that mean? I do not understand.

l. 582, “developed” would be quite a bit an overstatement, better use “presented”

The fundamental parts of the conclusion chapter are repetitive, giving a mere summary of details described above. A proper conclusion should step beyond the above presented material and ask “what do we know now, or what can we do now?” It shall feed back to the main research question and answer it.

l. 600, “allowing the spatial location”, yes but only when a single impact is isolated, which I think will be of little help under real world conditions.