the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Jun 2026
The Global TanDEM-X High-Resolution Coastline Product
Abstract. We present a globally consistent, high-resolution (10 m) coastline product derived from the TanDEM-X DEM product of the German SAR satellite mission TanDEM-X (DOI: 10.15489/entfdgi5es81). This vector product was developed with the aim of refining the TanDEM-X DEM, ensuring an optimal fit with this dataset. Additionally, the TanDEM-X coastline is a key component for deriving hydrographic data in the project HydroSHEDS v2.0 and thus defines the outer boundary of watersheds and rivers. To meet hydrological requirements, the TanDEM-X coastline was first extracted automatically and subsequently manually verified and improved. Since the coastline is a mapping feature located in the coastal zone, its position depends on the source data, the method used to identify the coastline, the time of acquisition, and the purpose of mapping. Due to this complex process and the lack of true reference data, a validation of the coastline product is challenging. Therefore, the quality of the TanDEM-X coastline is evaluated through statistical and qualitative comparisons in different globally distributed coastal areas and against multiple existing coastline products.
- Preprint
(1605 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2026-277', Anonymous Referee #1, 28 Jun 2026
- Coastlines are dynamic features; so the acquisition period of all input datasets should be clearly stated. Could the authors also provide information on coastline changes that may have occurred during this period and explain whether such changes were considered in the analysis?
- Page 4: I recommend adding a figure that illustrates the format of the delivered coastline product, specifically showing that it is provided as polygons outlining landmasses. The figure could also demonstrate the minimum 10 m spacing between consecutive polygon vertices.
- Page 5: Please provide further justification for using an external coastline as the initial input for coastline generation. In addition, include key specifications of the Copernicus DEMs used in the study, such as its spatial resolution and any other relevant characteristics.
- Page 5: Please consider referring https://gee-community-catalog.org/projects/s2coast/ and https://doi.org/10.1038/s41597-025-05180-9 as part of your reference datasets. These datasets provide global coastline products at spatial resolutions of 10 m and 2 m, respectively.
- Page 8: ‘The overall quality of the coastline corresponds to that of the TanDEM-X dataset’. This statement could be misleading, as only the spatial resolution of TanDEM-X data is directly inherited by the coastline product, but not necessarily the accuracy of the coastline extraction itself. I recommend revising the wording to more accurately reflect this distinction.
Citation: https://doi.org/10.5194/essd-2026-277-RC1 -
RC2: 'Comment on essd-2026-277', Gregor Luetzenburg, 21 Jul 2026
Review of The Global TanDEM-X High-Resolution Coastline Product by Warmedinger et al. submitted for publication at Earth System Science Data.
https://doi.org/10.5194/essd-2026-277
Summary
The manuscript presents the TanDEM-X high-resolution global coastline product, a vector dataset delineating the land-ocean boundary, derived from the TanDEM-X global DEM and refined through a combination of automated extraction and expert-based manual correction. The value of this contribution a globally consistent, high-resolution coastline product derived from a single, homogeneous SAR-based data source. The product's intended role in supporting hydrological applications such as HydroSHEDS v2.0, and its potential utility for a broad range of coastal and hydrological research communities, is well motivated.
However, the manuscript in its current form falls short of substantiating the quality claims it makes about the product. Most critically, no absolute horizontal accuracy is reported anywhere in the manuscript. This is, in my assessment, the single most important number missing from this work, without which a reader cannot judge whether the product is fit for their intended purpose. Related to this, several other sources of uncertainty that are essential to a rigorous data descriptor are not quantified or, in some cases, not mentioned at all: geometric/positional uncertainty from the extraction and vectorization workflow, operator bias introduced during manual correction, omission error, the influence of tidal stage at acquisition, and the effect of coastal change occurring across the multi-year (2010–2015) acquisition window on the internal consistency of a product presented as a single static dataset.
General comments
The abstract would benefit from following a clearer, more standard structure: (1) context/motivation, (2) what was done (data and methods, briefly), (3) key results with concrete numbers, and (4) significance/outlook. As currently written, the abstract focuses heavily on motivation and downstream applications (HydroSHEDS, DEM refinement) but omits several details a reader needs to assess the product itself, most notably the uncertainty estimates from the validation. Since this is a data descriptor, readers will expect the abstract to answer, at minimum: what was mapped, at what resolution and extent, how it was validated, and how accurate it is, ideally with at least one headline number for each.
The manuscript states the coastline product has a "spatial resolution of 10 m," directly equating this to the resolution of the underlying TanDEM-X DEM. This equivalence is not strictly valid and should be clarified or qualified. Raster spatial resolution refers to the pixel/grid spacing of the source data. A derived vector product, by contrast, does not inherit this resolution directly: its positional accuracy and level of geometric detail depend on the full processing chain used to extract it, including the threshold used to identify the land-water boundary, any smoothing, generalization, or simplification applied to the vector geometry, the manual verification and editing and the interpolation behaviour of the extraction method itself. As a result, the vector coastline could plausibly have effective positional accuracy coarser than 10 m (e.g., if extraction or smoothing introduces additional uncertainty) or, in principle, appear more detailed at some locations (e.g., where manual editing refines the automatically extracted line beyond the native pixel grid). I would recommend the authors either substantiate the 10 m resolution claim for the vector product specifically, ideally through a quantitative positional accuracy assessment, or rephrase the statement to make clear that 10 m refers to the resolution of the source DEM, with the vector product's own positional accuracy reported separately and explicitly.
The manuscript would benefit from situating the TanDEM-X coastline product relative to other established automated coastline mapping efforts, most notably Duan et al.'s S2Coast-2023 (https://doi.org/10.1016/j.rse.2025.115186), and Vos et al.'s CoastSat (https://doi.org/10.1016/j.envsoft.2019.104528). Both are widely used, openly available products against which readers will naturally want to benchmark a new global coastline dataset. I would recommend the authors include a direct quantitative comparison in the validation section.
Bridges and other man-made structures that visually resemble a coastline in the source data do not appear to be resolved correctly by the workflow. Please clarify how such features are identified and excluded (or corrected) during processing, and whether this is a systematic limitation the authors are aware of.
The manuscript does not report or quantify several key sources of uncertainty that a reader would need to assess product quality, including: absolute horizontal positional uncertainty, geometric uncertainty from the extraction/vectorization workflow, operator bias introduced during manual correction, and omission error (i.e., coastline segments missed entirely). Given that the conclusion states the product's "strengths and limitations are visualized in a comparison," this is not sufficient. A quantitative uncertainty budget is needed for the product to be usable by downstream applications.
The source data used for manual coastline correction, and the associated positional uncertainty of that reference data, is not described. Since manual correction is stated to be central to achieving "global consistency," the reliability of the reference material used for this step directly affects the reliability of the final product and should be reported.
The tidal stage at the time of SAR image acquisition is not mentioned. Given that coastline position, particularly in tidal flats, estuaries, and other low-relief coastal zones, can shift substantially with tide, this is a significant omission that should be addressed, either by reporting how tidal stage was accounted for (or not) or by discussing it as a limitation.
Source data were acquired between 2010 and 2015, yet the manuscript does not discuss how coastal changes occurring across this ~5-year window are handled or how they affect the internal consistency of a product presented as a single, static, "globally consistent" dataset.
Please provide a clear list of all files and formats delivered with the dataset, e.g., whether the coastline is available as both polygon and line GeoPackages, whether a raster version is also provided, whether the product is available as regional/continental subsets in addition to a single global file, and whether a file documenting acquisition time per coastline segment is included. This information is currently missing and is important for usability.
When comparing against other coastline products, how do the authors account for natural coastline variability between the TanDEM-X acquisition period and the acquisition period of the reference data? Without addressing this, apparent disagreement between products could reflect real coastal change rather than product error (or vice versa).
As currently written, the sections 4.1 & 4.2 primarily describe the validation methodology rather than presenting and interpreting results. Please restructure so that the actual comparison outcomes and their implications for product quality are discussed, separate from the methodological description.
Rather than focusing on a specific glacier (e.g., Petermann) and its individual retreat history, I would recommend a more general discussion of marine-terminating glaciers and floating ice tongues, and their particular challenges for coastline mapping. This would also be a natural place to reference dedicated efforts addressing highly dynamic marine-terminating glacier fronts, such as AutoTerm and Greene et al. 2024 (https://doi.org/10.1038/s41586-023-06863-2).
This is, in my view, the manuscript's main shortcoming: nowhere is an absolute horizontal accuracy value provided for the coastline product. This is the single most important number for a reader to assess whether the product is fit for their purpose, and its absence undermines the "overall high quality" claims made throughout (see also my comment on L463).
Specific comments
L9–10: Please add "https://doi.org/" in front of the DOI so it resolves as a clickable link.
L10: Delete the first occurrence of "TanDEM-X" in this sentence. It is redundant with "TanDEM-X DEM product" later in the same sentence.
L11: Please clarify which vector product is meant here: the coastline product itself? The sentence is also unclear on the direction of causality: the coastline is stated to be derived from the TanDEM-X DEM, yet this sentence states it was developed to refine the DEM.
L13: Please define HydroSHEDS v2.0 on first use or remove the reference if it is not essential here.
L14: Please clarify the mechanism by which a coastline product defines the outer boundary of watersheds and rivers. As written, this causal link is not self-evident to a reader unfamiliar with the HydroSHEDS workflow.
L15–17: This sentence does not add new information. It effectively states that a coastal feature is located in the coastal zone, and that mapped features generally depend on source data, method, acquisition time, and purpose. Consider removing it or replacing it with a more specific statement about what this means for the TanDEM-X coastline product itself.
L17–18: Please define "true reference data" and briefly explain why it is unavailable for coastline validation (e.g., no globally consistent ground-truth coastline exists at comparable resolution/accuracy).
L18–20: Comparison against multiple existing coastline products is reasonable, but as written this does not communicate what distinguishes your product's quality or accuracy from those alternatives. Please state the outcome of this evaluation.
L22–27: I agree with the general reasoning, but there are considerably more reasons why coastline mapping matters, and the manuscript undersells its own product by not engaging with this literature. Please see Vos et al. (2019, https://doi.org/10.1016/j.envsoft.2019.104528) for well-articulated reasoning on the importance of coastline mapping. More generally, I am surprised the manuscript does not include any comparison with Vos et al.'s CoastSat, given its prominence in the field.
L38: I still do not understand how the TanDEM-X-derived coastline improves the DEM. Please explain this mechanism in more detail.
L40: How is "well-established" defined here? I would recommend removing this term unless it can be substantiated.
L54–60: I would not use "coastline" and "shoreline" interchangeably. A shoreline is the exact, constantly changing line where water meets land, shifting over short timescales. A coastline is a broader, more permanent boundary marking the general limit of the landmass meeting the sea, encompassing the wider zone affected by coastal and marine processes.
L61: Here the coastline is correctly described as a dynamic feature changing over time and space, which is inconsistent with L55, where it is described as a one-dimensional feature without transitional or dynamic extent. Please reconcile this discrepancy by revising L55.
L63: Coastline change through deposition and erosion can also occur very rapidly. Please avoid implying these processes are inherently slow.
L63–64: Use either "deposition and erosion" or "depositional and erosional processes". The current phrasing mixes both forms.
L65: What about anthropogenic coastline change? See, for example, Sengupta et al. 2025 (https://doi.org/10.1016/j.ocecoaman.2025.107579).
L84: How should the "10 m spatial resolution" of a vector dataset be interpreted here? Is the implication that the mapped coastline lies within 10 m of the true coastline? If so, this should be stated explicitly and substantiated; if not, please clarify what the stated resolution actually means for the vector product.
L89: This does not appear to be accurate. I have observed many locations where vertices are spaced considerably further apart than stated.
L172–174: This sentence is unclear as written: "exceed a certain threshold... assigned to the land or water mask, respectively" does not logically follow, since a single exceedance condition cannot map to two different outcomes via "respectively." Presumably pixels above the threshold are assigned to one class (e.g., land) and pixels at or below the threshold to the other (e.g., water). Please rephrase accordingly.
L198–200: The 0.4 arcsecond resolution of the source DEM cannot simply be "transferred" to the vector coastline product. Raster resolution and vector positional accuracy are not equivalent, and the vectorization process (thresholding, morphological dilation/erosion, manual editing) introduces its own uncertainty that is not necessarily bounded by the original pixel size. Please either provide a quantitative positional accuracy assessment of the vector product to substantiate this claim, or rephrase to avoid implying a direct, lossless transfer of raster resolution to vector geometry.
L235: What is the "certain minimum distance"? Please state the specific value.
Figure 3: How were these sites selected? Are the red boxes shown to scale? Why do the sites appear in several distinct clusters rather than a more even global distribution?
L273: What are the "certain limitations" of the radar system? Please be specific.
L283–291: This reads more like a figure caption than manuscript text. Consider moving this information into Table 3 with a clear reference to the site locations (red boxes) shown in Figure 3. Also, why does Figure 3 show considerably more red boxes than are described in the text or listed in Table 3?
Table 3: What distinguishes Fjord 1 from Fjord 2, or Steep Coast 1 from Steep Coast 2? There also appears to be a strong bias toward Northern Hemisphere sites. Why are not all sites shown in Figure 3 listed in the table? Please also provide average values (mean and median) across all study sites, and clarify why standard deviation varies so substantially between sites despite broadly similar mean values.
L308–309: How much is "relatively high" and "comparatively short"? What does "detailed assessment" refer to specifically? Please provide concrete figures.
L312–328: Given the coastline paradox, comparing total coastline length across products has limited value. I would instead like to see a proper quantitative uncertainty assessment, specifically, what is the absolute horizontal accuracy of your coastline product? This is the key figure needed for readers to assess product quality.
L376: Replace "glacier tongues' bay" with "fjord."
L381: The grounding line is in fact visible in satellite imagery see for example Andersen et al. 2026 (https://doi.org/10.5194/tc-20-1589-2026).
L452: Technically, the provided DOI does not lead directly to the data, but only to the URL https://geoservice.dlr.de/web/datasets/tdm_coastline. I would suggest providing a DOI that resolves directly to the data rather than to a general webpage.
L463: "Overall high quality" is asserted without presenting the relevant key figures to support this claim. If I were to stand at any given point along the coast, how far would the TanDEM-X coastline be from the true coastline, and how confident are the authors in that estimate?
Citation: https://doi.org/10.5194/essd-2026-277-RC2
Data sets
TanDEM-X High-Resolution Coastline - Global, 10m TanDEM-X Science Team https://doi.org/10.15489/entfdgi5es81
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 153 | 52 | 8 | 213 | 6 | 11 |
- HTML: 153
- PDF: 52
- XML: 8
- Total: 213
- BibTeX: 6
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1