the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Aug 2024
Seeing through the Sea with Satellites: Reconstructing Ocean Subsurface Temperature and Salinity with Satellite Observations
Abstract. In-situ measurements of ocean temperature and salinity are critical to ocean-related studies but are limited in space and time. Satellite retrievals provide high-resolution, globally-covered sea surface temperature (SST), salinity (SSS) and cannot directly measure the subsurface information., and height (SSH), but are limited to the ocean surface and cannot directly measure the subsurface information. Here we design a physics-informed algorithm that can reconstruct the vertical distributions of upper ocean temperature and salinity based purely on satellite observations. The algorithm stresses the tight ocean surface-subsurface coupling and the co-variability of ocean temperature and salinity. It is firstly tested with climate model simulations and then validated with actual observations by Argo floats, moored buoys and multiple ocean reanalysis datasets. The resultant satellite-based upper ocean temperature and salinity dataset has a global coverage, a high spatial resolution, and resolves ocean thermohaline structure from surface to 400 m. This dataset complements existing ocean subsurface products as an independent satellite-based observational dataset. The success of our reconstruction algorithm highlights a pressing need to maintain and advance the satellite observations of SST, SSS, and SSH. The reconstructed ocean temperature and salinity dataset can be accessed at https://doi.org/10.5281/zenodo.13145129 (Liu, 2024) and be used by researchers to study mesoscale ocean phenomena, assess the ocean heat content in various sea areas and etc.
- Preprint
(9604 KB) - Metadata XML
-
Supplement
(3260 KB) - BibTeX
- EndNote
Status: open (until 01 Jan 2025)
-
RC1: 'Comment on essd-2024-334', Anonymous Referee #1, 26 Sep 2024
reply
Review of the manuscript “Seeing through the Sea with Satellites:
Reconstructing Ocean Subsurface Temperature and Salinity with Satellite Observations” by Shizuo Liu and Shineng HuThe manuscript describes the application of a statistical methodology to reconstruct the vertical structure of temperature and salinity in the upper oceans (down to 400 m) starting from measurements of SST, SSS and SSH. The method is based on multivariate EOF computations applied to either climate model data or satellite and in situ interpolated products. The manuscript presents several major flaws and some questionable claims, both related to the originality and novelty of the methodology itself and in terms of relevance of the data and validation strategy that are presented (see detailed comments below). For these reasons, I believe that the manuscript requires quite substantial re-working and re-writing, and thus recommend rejection with a suggestion to resubmit it only after all major issues have been thoroughly addressed.
Major issues:
Limited Originality/Novelty of the methodology and inadequate reference to previous works
The use of vertical EOFs for projecting surface values at depth dates back several decades, with initial algorithms proposed by Carnes et al. in 1990 and 1994. These algorithms were also later adopted in operational systems (see Fox et al. 2002). In 2003, Pascual and Gomis introduced the concept of using multivariate EOFs to project surface information at depth, though their work was limited to geostrophic transport and did not address temperature (T) and salinity (S). During the same period, other techniques, such as the Gravest Empirical Mode (e.g. Mitchell et al., 2004; Meijers et al. 2004), were proposed and successfully tested. However, these methods were not referenced by the authors at all.
The approach based on multivariate EOFs was subsequently extended to jointly reconstruct temperature, salinity, and sea height (SH) vertical profiles from surface data (Buongiorno Nardelli and Santoleri 2005). That work initiated a series of studies that effectively applied the technique to satellite data either limiting to the purely statistical approach (e.g.: Buongiorno Nardelli et al., 2012, 2017) or including simplified dynamical information (e.g. Yan et al., 2020, 2021). The technique proposed in the present manuscript basically reduces the original multivariate T-S-SH EOF reconstruction to bivariate T-S EOFs, followed by the projection of surface height onto the identified modes. As such, it is misleading to present this as an entirely original and novel approach without referencing these foundational works.
Moreover, the list of machine-learning techniques presented is quite limited, and the authors' statements regarding their limitations seem biased and insufficiently justified. Additionally, many other relevant techniques have been proposed that are worth mentioning, such as those in Han et al. (2019), Buongiorno Nardelli (2020), Su et al. (2022), Pauthenet et al. (2022), Smith et al., (2024).
Input data use for the observational study
The authors train their statistical model using either monthly simulations from OGCMs or various observation-based datasets. However, it is unclear which space-time resolution they are targeting, especially since some input data are limited to a 1°x1° spatial resolution and monthly frequency (with the exception of SST data, which is daily). No explanation is provided on how the differences in grid resolution are handled when building the model. If the goal is to produce monthly reconstructions, the claimed benefits of the new product for mesoscale dynamics studies appear mostly unjustified.
Even more importantly, it seems the authors are unaware that the in-situ observational dataset they are using does not provide direct measurements but rather a low-resolution interpolation of sparse in-situ profiles from the global Argo drifting network. Consequently, it cannot be assumed that EOFs estimated from such a dataset would accurately capture dynamical modes beyond large scale seasonal signals. This limitation should be carefully considered when discussing the relevance and implications of their findings. Conversely, interpolated Argo data are always referenced to as “true” in situ measurements throughout the text, which is misleading and creates confusion.
Choice of reference datasets and validation metrics
When proposing new products, it is essential to carefully review existing similar products and demonstrate, through direct comparison, where (or whether) the new product offers improvements. Any purely data-driven reconstruction of the global 4D ocean state should be compared to well-established datasets like EN4 (Good et al., 2013) and ARMOR3D (Guinehut et al., 2004, 2012), which are publicly available, well-documented, and widely used by the scientific community.
Another reference that should always be kept into consideration is provided by (monthly) climatologies eventually estimated from the input data themselves (any new product should perform better than that).
The choice of the metrics for product validation is also quite important to ensure a robust and scientifically sound assessment. It has no sense to me that the validation of data at ¼° is carried out at 2°x2° just to increase some spatial correlations (e.g. fig.5, fig. 11,…). Similarly, monthly data should not be validated looking at annual statistics (e.g. standard deviation in annual average temperature, fig.6).
The only comparison with (just one) true observed timeseries is provided in fig.13. The choice of presenting separate values of each timeseries, however, seems not fully suited to appreciate quantitatively how accurate the reconstruction is. Maybe one could better sense relative performances looking at the timeseries of the differences between observed and modelled values (and also including synthetic metrics such as rms differences).
Methodological aspects
It is unclear why the authors normalize the profiles in input to the EOF dividing them by the standard deviation of each variable at the surface (and not, for example with respect to total standard deviation). This would likely lead to excessive weight given to variables that may display a higher variance at depth. This point definitely requires additional discussion.
The way the cost function is defined, and the need for additional weights are introduced, is rather unclear. From what I understand, the hypothesis is that SSH can be obtained as a combination of a triplet of height anomalies that are equivalent to three surface modes, as they are obtained by projection of SSH on the first three joint PC. These should likely be weighted exactly as the first three T/S vertical modes provided an analogous normalization is carried out.
Moreover, it is unclear why the authors decide to go for an iterative approach instead of directly solving the linear system associated with the 3 expressions that describe the truncated EOF reconstruction of SST, SSS and SSH (exactly as done in Buongiorno Nardelli and Santoleri, 2005). It is unclear what is the rationale of this approach, as well as the advantage of having two cost functions to estimate this “subjective” weights (defined such by the authors themselves).
Even after normalization, there's no clear explanation of how the model treats data across different depths. If the original layers from the model are retained without any modification, this could lead to unequal weighting of variability at different depths, which would posiibly introduce biases and/or inaccuracies in the analysis. The methodology for handling this depth-related variability or justification to ignoring it, needs to be clarified.
It's unclear how satellite-derived data is processed, particularly whether the data has been remapped to the same spatial grid and whether any averaging has been applied to ensure consistency in surface resolution. These details are important for understanding how the data aligns with the model's spatial structure and should be addressed to avoid any ambiguity about the data's integration. It is also crucial to allow reproducibility of the results.
References:
Carnes, M. R., Mitchell, J. L., & Dewitt, P. W. (1990). Synthetic temperature profiles derived from Geosat altimetry: Comparison with air‐ dropped expendable bathythermograph profiles. Journal ofGeophysical Research, 95(C10), 17,979–17,992.
Carnes, M. R., Teague, W. J., & Mitchell, J. L. (1994). Inference of subsurface thermohaline structure from fields measurable by satellite. Journal ofAtmospheric and Oceanic Technology, 11(2), 551–566
Fox, D. N., Teague, W. J., Barron, C. N., Carnes, M. R., & Lee, C. M. (2002). The Modular Ocean Data Assimilation System (MODAS). Journal ofAtmospheric and Oceanic Technology, 19(2), 240–252. https://doi.org/10.1175/1520‐0426(2002)019<0240:TMODAS>2.0.CO;2
Mitchell, D., M. Wimbush, D. Watts, and W. Teague, 2004: The residual GEM and its application to the southwestern Japan/ East China Sea. J. Atmos. Oceanic Technol., 21, 1895–1909.
A. J. S. Meijers, N. L. Bindoff, S. R. Rintoul, Estimating the four-dimensional structure of the southern ocean using satellite altimetry. J. Atmos. Ocean. Technol. 28, 548–568 (2011).
Pascual, A., and D. Gomis, 2003: Use of surface data to estimate geostrophic transport. J. Atmos. Oceanic Technol., 20, 912–926
Buongiorno Nardelli, B., & Santoleri, R. (2005). Methods for the reconstruction of vertical profiles from surface data: Multivariate analyses, residual GEM, and variable temporal signals in the North Pacific Ocean. Journal ofAtmospheric and Oceanic Technology, 22(11), 1762–1781. https://doi.org/10.1175/JTECH1792.1
Buongiorno Nardelli, B., Guinehut, S., Pascual, A., Drillet, Y., Mulet, S., & Ruiz, S. (2012). Towards high resolution mapping of 3‐D mesoscale dynamics from observations. Ocean Science, 8(5), 885–901. https://doi.org/10.5194/os‐8‐885‐2012
Buongiorno Nardelli, B., Guinehut, S., Verbrugge, N., Cotroneo, Y., Zambianchi, E., & Iudicone, D. (2017). Southern Ocean mixed layer seasonal and interannual variations from combined satellite and in situ data. Journal ofGeophysical Research: Oceans, 122, 10,042–10,060. https://doi.org/10.1016/j.rse.2015.04.025
Yan et al., A Dynamical‐Statistical Approach to Retrieve the Ocean Interior Structure from Surface Data: SQG‐mEOF‐R. J. Geophys. Res. Ocean. (2020), doi:10.1029/2019jc015840.
Yan, R. Zhang, H. Wang, S. Bao, C. Bai, Practical dynamical-statistical reconstruction of ocean’s interior from satellite observations. Remote Sens. 13, 1–18 (2021).
Han, M., Feng, Y., Zhao, X., Sun, C., Hong, F., and Liu, C. (2019). A convolutional neural network using surface data to predict subsurface temperatures in the pacific ocean. IEEE Access 7, 172816–172829. doi: 10.1109/ACCESS.2019.2955957
Buongiorno Nardelli, B. (2020). A deep learning network to retrieve ocean hydrographic profiles from combined satellite and in situ measurements. Remote Sens. 12. doi: 10.3390/RS12193151
Su, H., Jiang, J., Wang, A., Zhuang, W., and Yan, X.-H. (2022). Subsurface temperature reconstruction for the global ocean from 1993 to 2020 using satellite observations and deep learning. Remote Sens. 14, 3198. doi: 10.3390/rs14133198
Pauthenet et al., Four-dimensional temperature, salinity and mixed-layer depth in the Gulf Stream, reconstructed from remote-sensing and in situ observations with neural networks. Ocean Sci. 18, 1221–1244 (2022).
A. H. Smith et al., Reconstruction of subsurface ocean state variables using Convolutional Neural Networks with combined satellite and in situ data. Front. Mar. Sci. 10, 1–16 (2023).
Guinehut, S., Dhomps, A. L., Larnicol, G., & Le Traon, P. Y. (2012). High resolution 3‐D temperature and salinity fields derived from in situ and satellite observations. Ocean Science, 8(5), 845–857. https://doi.org/10.5194/os‐8‐845‐2012
Guinehut, S., Le Traon, P. Y., Larnicol, G., & Philipps, S. (2004). Combining Argo and remote‐sensing data to estimate the ocean three‐ dimensional temperature fields—A first approach based on simulated observations. Journal ofMarine Systems, 46(1–4), 85–98. https:// doi.org/10.1016/j.jmarsys.2003.11.022
Good, M. J. Martin, N. A. Rayner, EN4 : quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates. J. Geophys. Res. 118, 6704–6716 (2013).
Citation: https://doi.org/10.5194/essd-2024-334-RC1
Data sets
Reconstructing Ocean Subsurface Temperature and Salinity with Satellite Observations Shizuo Liu https://doi.org/10.5281/zenodo.13145129
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 357 | 48 | 33 | 438 | 24 | 6 | 8 |
- HTML: 357
- PDF: 48
- XML: 33
- Total: 438
- Supplement: 24
- BibTeX: 6
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1