the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Sep 2026
Comparison of measurements from pressure-recording inverted echo sounders, moored ADCPs, and ocean reanalysis products in the North Equatorial Current region of the western Pacific
Abstract. With recent advances in data-assimilating ocean reanalysis products, mesoscale oceanic processes can be more realistically represented, increasing the need for regional performance assessments. Here, two widely used products, the Global Ocean Physics Reanalysis (GLORYS) and the Hybrid Coordinate Ocean Model reanalysis (HYCOM), are compared against two years of in situ observations from four pressure-recording inverted echo sounders (PIESs) and two acoustic Doppler current profilers (ADCPs) in the North Equatorial Current (NEC) region. While both products reproduce sea surface height (SSH) reasonably well, GLORYS shows closer agreement with the observations. However, their correlations are lower at the southern than at the northern sites located approximately 419 km apart, ranging from 0.84 (0.79) at the northernmost PIES site to 0.76 (0.76) at the southernmost site for GLORYS (HYCOM), consistent with the known degradation of satellite-measured SSH toward lower latitudes. The same trend also appears in the zonal current fields, with depth-averaged correlations (60–400 m) dropping far more sharply, from 0.75 (0.65) at the northern ADCP station to 0.36 (0.39) at the southern station. This degradation is largely associated with the reduced accuracy of satellite altimetry toward lower latitudes; assimilation of these errors can distort reanalysis SSH gradients and, consequently, zonal geostrophic currents. The relatively poor representation of intermediate-layer zonal currents at the southern ADCP station, particularly in GLORYS, may also reflect contributions from inaccuracies in the upper-ocean density structure. Although the accumulated density structure errors are comparable between the northern and southern NEC regions, differences in the relative contributions of the surface and baroclinic geostrophic components appear to explain the greater influence of baroclinic errors on the intermediate current in the southern region. In contrast, HYCOM shows vertically coherent anomalous flows throughout the water column, suggesting that surface-geostrophic variability is projected too deeply. Overall, GLORYS effectively reproduces the variability of the NEC region, though care should be taken of its limitations in interpreting intermediate-layer currents and variability affected by satellite altimetry errors, particularly in the southern NEC region.
- Preprint
(10764 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-5239', Anonymous Referee #1, 30 Oct 2026
-
RC2: 'Comment on egusphere-2026-5239', Anonymous Referee #2, 30 Oct 2026
General comments:
The authors compare two of the most widely used mesoscale-resolving reanalyzes, GLORYS12 and HYCOM GOFS 3.1, against two years (2012–2014) of data from four PIES and two ADCP moorings in the western Pacific NEC. Both products capture SLA reasonably well. Zonal-current skill drops sharply from the northern mooring M1 to the southern mooring M2. The authors attribute this to satellite altimetry errors (hump artifacts, degradation at low latitudes) assimilated into the products, as well as accumulated upper-ocean density errors that matter more where the surface geostrophic signal is weak. This work provides a genuinely independent dataset of mooring observations and shipboard CTD measurements for evaluating two of the most widely used reanalysis products in a region where the dynamics are complex to reproduce. The dataset and comparisons are both valuable to a broad community. I recommend publication after minor revisions to clarify and strengthen several arguments presented in the manuscript.
Specific comments
1.- Section 2.3 Methods: I know it’s pretty classical and widely used, but a short description of the EOF. I’m not suggesting authors should necessarily implement it here, but a complex version including both u and v can be considered. Just to point out, in Figure 3, standard deviation of v is of the same order than u’s.
2.- All eta’s RMSDs (4–7 cm) are at or below the stated PIES RMS error of 6.3 cm. A simple assessment of the errors in Δη gives around 9 cm (assuming uncertainties are independent among F sites), that makes difficult to have a sense of how bad Δη errors in Figure 7a. Not sure if a shade around the PIES Δη would make it easier. Perhaps a gray shading similar to that used in Figure 2.
2.1.- I also recommend to add a simple assessment of this uncertainty in the geostrophic estimations, to check if presented RMSDs exceed it.3.- I'm aware that this is not commonly done, as ADCP observations are often treated as ground truth. However, measurement noise associated with acoustic attenuation, low concentrations of suspended scatterers, and other sources of instrumental uncertainty can be estimated from the power spectral density (PSD) of the velocity observations.
You can find examples of this approach in these studies (https://doi.org/10.1175/JPO-D-14-0119.1, 10.1175/JTECH-D-24-0139.1 {see link for the full paper at the bottom}, https://doi.org/10.21203/rs.3.rs-11117134/v1).
The analysis can be performed separately for each ADCP depth bin or using bin-averaged velocities, depending on the application. Basically, the procedure involves identifying the flat, high-frequency portion of the velocity PSD and integrating it over the corresponding frequency range to estimate the variance associated with the white-noise component of the observations. This provides an estimate of the magnitude of the instrumental noise. As a short note, white-noise floor should preferably use the unfiltered hourly observations.
5.- I would also recommend including the RMSD normalized by the standard deviation of the reference ADCP velocity time series at each depth. This would provide a dimensionless measure of the reanalysis errors relative to the observed velocity variability, allowing a more meaningful comparison of model performance across depths and between stations M1 and M2, where the observed variability differs considerably. At least in the comparisons with ADCP.
6.- I recommend add a short discussion about the previous assessment of the effective degrees of freedom of the AVISO-CMEMS gridded altimetry products (early ¼ deg; https://doi.org/10.5194/os-15-1091-2019). The Allsat line in figure 2 is considerable smoother than \eta_{PIES}, I this might be reflecting the higher temporal resolution of the PIEs. The data could provide a sense of the actual temporal capability of gridded altimetry in this subregion.6.1. A quick assessment of how the Cr/RMSDs between PIES-derived SSH and satellite altimetry change as the PIES records are progressively smoothed would be useful. A frequency-dependent coherence analysis might also help identify the time scales at which the two datasets agree, but you could also keep it simple. The main objective would be to try to distinguish discrepancies arising from differences in effective temporal resolution from those associated with actual contamination of the altimetry data.
7.- Also in the discussions, the propagation of the hump errors via vertical projections is not addressed. The products use different schemes, but both use a vertical projection that might also propagate the hump errors into the baroclinic structure.
Technical corrections
1.- At the end of the results section, a summary table (r, RMSD, bias, std ratio, with uncertainties) would help a lot.
2.- It’s a minor detail, but in lines 135, authors mention that HYCOM have 40 layers, then, at line 42 they mention it has 41 layers interpolated to 40 vertical levels, so, perhaps in 135 they are talking about vertical levels?
3.- I'm not sure if it’s just me, but I find a bit difficult to see the cyan numbers in the axis tick and annotations in Figures 2 and 4-5 for example. Sometimes I also struggle a bit to see the lines in Figure 2 when the cyan line is over white background.Citation: https://doi.org/10.5194/egusphere-2026-5239-RC2 -
RC3: 'Comment on egusphere-2026-5239', Anonymous Referee #3, 03 Nov 2026
Reviewer expertise note: Firstly, My primary expertise is in marine chemistry, transient tracers (CFCs/SF₆), water-mass ventilation, and marine environmental baseline assessment, rather than in numerical ocean reanalysis, satellite altimetry, or the internal numerical cores of HYCOM/NEMO. My comments therefore focus on independent observational validation, statistical robustness, error attribution, data independence, and presentation. I recommend that the editor also seek a specialist review on the reanalysis/altimetry aspects, particularly the HYCOM thermobaric-instability discus
General assessment This manuscript compares two widely used eddy-resolving ocean reanalysis products, GLORYS12V1 and HYCOM GOFS3.1, against approximately two years of in situ observations from four PIESs and two ADCPs in the North Equatorial Current region of the western Pacific. It further examines SLA and zonal-current variability, and discusses possible causes of the discrepancies, including satellite-altimetry errors entering the reanalyses through data assimilation, differences in surface versus baroclinic geostrophic contributions, and upper-ocean density-structure errors. The study addresses a relevant regional problem. The use of PIES and ADCP observations as independent references is valuable, and the manuscript goes beyond a simple “product A versus product B” correlation exercise by attempting to diagnose the vertical structure of zonal-current errors and the relative roles of surface and baroclinic geostrophic components. The finding that performance degrades toward lower latitudes, and that intermediate-layer currents at the southern ADCP site are less well reproduced, is potentially useful for users of these reanalysis products. However, the novelty is moderate: PIES-based evaluation of satellite altimetry in this region has been reported previously (e.g., Jeon et al., 2018), and the present study mainly extends this to two reanalysis products. The observational basis is limited (four PIESs, two ADCPs, two years), and several key conclusions depend on statistical and error-attribution choices that are not sufficiently justified. In particular, the independence of the validation data, the statistical significance of correlation differences, the geometric representativeness of the cross-PIES SSH difference, and the causal attribution to hump artifacts require stronger support. For these reasons, I recommend major revision rather than acceptance in the present form.
Major comments
1. Independence of validation data and the role of Argo. The authors correctly note that Argo profiles are assimilated into both reanalysis systems and are therefore not strictly independent. Nevertheless, Section 4 relies heavily on Argo-based density and geopotential-height comparisons. The additional CTD dataset is useful, but only 41 profiles are used, concentrated in three shipboard campaigns. The manuscript should clearly separate independent validation (PIES, ADCP, CTD) from consistency assessment (Argo, satellite SLA), and the main conclusions should rest primarily on the independent data. A table summarizing the temporal and spatial coverage of the CTD profiles, and a discussion of whether the CTD-based conclusions are sensitive to campaign timing, would strengthen the analysis.
2. Statistical significance and effective degrees of freedom. The time series are daily averaged and 3-day low-pass filtered, so the effective number of degrees of freedom is much smaller than the nominal sample size. Correlation differences such as 0.75 versus 0.36 are central to the paper’s conclusions but are not tested for significance. Depth-averaged correlations should be computed using Fisher z transformation, with confidence intervals and, where appropriate, p-values. EOF modes should be tested for robustness (e.g., bootstrap, split-year analysis, sign consistency). Event-based composites (hump periods versus non-hump periods, NEUC events versus non-events) would be more convincing than whole-record correlations alone.
3. Geometric representativeness of Δη and the zonal geostrophic interpretation. The PIES array follows an oblique Jason-2 track, and M2 is not located on a purely meridional line between F08 and F12. The use of Δη = η(F12) − η(F08) as a proxy for the zonal geostrophic current at M2 therefore requires a quantitative justification. The authors should provide a table of station coordinates, distances, and azimuths; project Δη onto the meridional or along-track direction as appropriate; and estimate the sensitivity of the inferred current to this geometric choice. The error in the ADT gradient at low latitudes should also be discussed.
4. Causal attribution to satellite-altimetry hump artifacts is suggestive but not definitive. The evidence that hump artifacts are transferred into the reanalysis SSH fields is largely based on the fact that η_Re and η_Sat agree more closely with each other than with η_PIES during shaded periods. This is consistent with the proposed mechanism but does not establish causality. The authors should examine assimilation increments or innovation statistics if available, or at least perform a more explicit hump-period versus non-hump-period statistical comparison. The language should be moderated from causal to “consistent with.” Similarly, the HYCOM noisy-SSH/thermobaric-instability explanation rests on unpublished work and personal communication; it may be mentioned as a possible factor, but it should not be used as a central conclusion.
5. Error budget and the link between density errors and baroclinic current errors. The order-of-magnitude estimate that geopotential-height errors of 0.023–0.034 m over ~170 km correspond to ~0.039–0.057 m s⁻¹ is reasonable, but the propagation should be presented more explicitly, including the standard-deviation cases. A quantitative decomposition of the variance and error contributions from u_sfc and u_bs at M1 and M2 would make the argument that “the weaker surface geostrophic component at M2 allows baroclinic errors to matter more” much more convincing. The PIES total RMS error of 6.3 cm should also be included in the RMSD error budget.
6. Consistency of observational and reanalysis processing. The manuscript should clarify whether the reanalysis currents compared with ADCP are total currents or geostrophic currents, and how the geostrophic decomposition in Eq. (3) is used diagnostically. The interpolation method, temporal averaging, low-pass filtering, and DAC/inverse-barometer corrections should be described consistently for all datasets. Since HYCOM is 3-hourly and GLORYS is daily, the exact temporal matching procedure should be stated.
7. Overgeneralization of conclusions. The statement that GLORYS “effectively reproduces the variability of the NEC region” should be qualified by the observational limits: four PIES sites, two ADCP sites, two years, and specific depth ranges. The conclusion would be more defensible if phrased in terms of the sampled sites, period, and variables, rather than the entire low-latitude western Pacific NEC region.
Minor comments
1. Clarify at first use what “hump artifact” means and its typical amplitude.
2. State whether correlations are Pearson or Spearman and whether trends were removed.
3. Provide ADCP tilt, missing-data percentage, and depth-range justification in a compact table or supplement.
4. Explain how satellite ADT gradients were computed on their grid before comparison with 1/12° reanalysis fields.
5. Ensure EOF sign conventions are unified before comparing modes across datasets.
6. Check the data-availability statement and the Zenodo link; Ocean Science requires clear access to underlying data and, where possible, processing code.
7. In the Introduction, state more explicitly how this study differs from Jeon et al. (2018): is the novelty the joint PIES–ADCP evaluation of two reanalyses, or the decomposition of surface versus baroclinic error contributions?
The manuscript has a useful observational dataset and a diagnostically oriented analysis, and it addresses a relevant regional evaluation problem. However, the statistical robustness, independence of validation data, geometric representativeness of the SSH-gradient proxy, and causal attribution need substantial strengthening. I would be willing to review a revised version that addresses these points, but in its current form the manuscript requires major revision.
Citation: https://doi.org/10.5194/egusphere-2026-5239-RC3
Data sets
PIES-derived sea level anomaly data in the North Equatorial Current region Chanhyung Jeon https://github.com/PADOLab1/NEC_PIES
Argo float data and metadata from Global Data Assembly Centre (Argo GDAC) Argo https://doi.org/10.17882/42182
Global Ocean gridded L4 sea surface heights and derived variables reprocessed 1993 ongoing Copernicus Marine Service (CMEMS) https://doi.org/10.48670/moi-00148
Global Ocean physics reanalysis (GLORYS12V1) Copernicus Marine Service (CMEMS) https://doi.org/10.48670/moi-00021
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 310 | 154 | 67 | 531 | 68 | 62 |
- HTML: 310
- PDF: 154
- XML: 67
- Total: 531
- BibTeX: 68
- EndNote: 62
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript compares GLORYS and HYCOM with two years of PIES and ADCP observations in the western Pacific North Equatorial Current region. The observational dataset is valuable, but I find the scientific contribution rather limited. Most of the manuscript consists of comparisons of SSH, currents, and hydrographic structures using different observational datasets. Although these comparisons are useful, they are not organized around a sufficiently clear scientific question. The manuscript currently reads more like a regional reanalysis evaluation report than a study providing new physical understanding. My concern is not that such comparisons lack value, but that the manuscript does not clearly demonstrate what new scientific insight can be gained from them. I therefore do not think the manuscript is suitable for publication in its present form. A more substantial restructuring of the scientific question and analysis is needed, rather than a conventional revision. I recommend rejection of the current manuscript, although a substantially revised study addressing a clearly defined scientific question could be considered as a new submission.
Major comments
Lines 30–73:The main weakness of this manuscript is its limited scientific novelty. PIES-based SSH comparisons have already been conducted in this region, including Jeon et al. (2018). The present study adds comparisons with two reanalysis products and ADCP observations, but what new scientific understanding is gained? The authors show that both products reproduce SSH reasonably well but perform poorly in representing some subsurface currents. What is the broader implication of this finding? Does it mean these products should not be used for NEC studies, or only that they have limitations under certain conditions? I think the authors need to reconsider the central scientific question rather than simply presenting more comparisons.
Figs. 2–6: Most of these figures present separate comparisons of different variables. PIES is used for SSH, ADCP for currents, and EOF analysis for vertical structures. However, the physical connection between these results is weak. Given the simultaneous PIES and ADCP observations, the authors should investigate how SSH variability is related to the subsurface current structure, rather than treating these comparisons separately.
Fig. 7 and Figs. 8–9:The discussion attempts to explain current errors in terms of SSH-gradient errors and inaccuracies in the density structure, but the evidence is not sufficiently convincing. In Fig. 7, the proposed relationship between SSH errors and current errors is mainly qualitative. A direct quantitative comparison would be more useful. For Figs. 8–9, geostrophic current errors depend on horizontal geopotential height gradients, not simply on regional mean geopotential height biases. In particular, the conversion of geopotential height discrepancies into current errors in Lines 369–374 needs stronger physical justification. These analyses are central to the proposed error mechanisms and should be substantially strengthened.
The ADCP data are not publicly available, although they support several major conclusions.
Minor comments
Lines 279–282 and 425–450: The conclusion that reanalysis performance deteriorates toward lower latitudes should be treated cautiously. Only two ADCP stations are available, and their background current structures are different. The lower correlations at M2 may not be caused by latitude alone.
Fig. 6: The EOF methodology needs clarification. Were the mean currents removed? How were the modes normalized and compared between datasets?
Lines 91–94 and 134–142: The description of PIES measurements is difficult to follow and should be rewritten. Also, HYCOM is described as having both 40 and 41 vertical layers. Please clarify.
Lines 185–187: The double-prime notation is defined as either observation minus reanalysis or reanalysis minus observation. A consistent sign convention should be used throughout the manuscript.