the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Jan 2026
NZ-BeachTopo30: A national-scale and full-coverage 30 m beach topography dataset for New Zealand reconstructed by fusing ICESat-2 and Sentinel-2
Abstract. Beaches provide essential ecological functions and support socio-economic resilience, yet accurate mapping is hindered by systematic limitations in global Digital Elevation Models (DEMs). A critical challenge remains in the intertidal zone, where frequent tidal inundation creates extensive data voids, disrupting the continuity of coastal topography. To bridge this fundamental data gap, we present NZ-BeachTopo30 which is a national-scale and full-coverage 30 m beach topography dataset for New Zealand constructed by fusing ICESat-2 photon-counting altimetry with Sentinel-2 multispectral time series. Using DeltaDTM as a high-precision baseline for the stable backshore, we trained an XGBoost model on ICESat-2 control points and Sentinel-2 spectral-geometric features to reconstruct the missing intertidal topography specifically. SHAP analysis was further employed to interpret the physical driving mechanisms of these predictors. Validation against airborne Lidar confirmed that the dataset accurately recovers elevations in previously void zones with an RMSE of 0.94 m. By integrating these predictions with the DeltaDTM baseline, the final national-scale product achieves robust accuracy with an R² of 0.75 and an RMSE of 1.17 m. This targeted integration significantly expanded valid topographic coverage by 145.8 % from 79.9 km² to 196.5 km². It delivers the first spatially continuous and full-coverage beach topography dataset for New Zealand. This product distinguishes itself by seamlessly bridging the critical intertidal gap that disconnects land and sea in existing global datasets. Given the global availability of ICESat-2 and Sentinel-2, NZ-BeachTopo30 offers a scalable solution for worldwide applications and provides a robust foundation for inundation modeling and coastal management.
- Preprint
(9041 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 19 Apr 2026)
-
CC1: 'Comment on essd-2025-826', Robbi Bishop-Taylor, 24 Feb 2026
reply
-
AC1: 'Reply on CC1', nan xu, 14 Mar 2026
reply
Dear Professor,
We sincerely thank you for your highly positive and constructive comments on our manuscript. We deeply appreciate your time and expertise, especially given your foundational contributions to coastal and intertidal remote sensing. Your insights regarding tidal biases and environmental scope are incredibly valuable and have significantly helped us improve the depth and rigor of our discussion. Please refer to the attachment for our detailed point-by-point responses.
Best wishes,
-
AC1: 'Reply on CC1', nan xu, 14 Mar 2026
reply
-
RC1: 'Comment on essd-2025-826', Anonymous Referee #1, 23 Mar 2026
reply
Dear authors,
I have reviewed the manuscript “NZ-BeachTopo30: Bridging the Intertidal Data Gap by Fusing ICESat-2 and Sentinel-2”. Overall, this paper presents a novel and valuable dataset for New Zealand beach topography, filling an important data gap. The manuscript is generally well structured, and the methodology is clearly described. However, some revisions are still required to enhance the quality of the manuscript, and my specific comments are detailed as follows.
1 In the introduction, the authors discuss the characteristics and limitations of airborne LiDAR and spaceborne LiDAR together. I recommend discussing them separately to better distinguish between the two in the manuscripts.
2 I suggest the authors ensure the completeness of information for all datasets used, as I noticed that some data sources do not have corresponding URLs provided. This is essential to guarantee data accessibility and the reproducibility of the methodology.
3 Sentinel-2 composite images serve as the foundation for beach topography modeling in this study. Hence, I recommend that the authors provide a more detailed explanation of the meaning of different quantiles derived from Sentinel-2 optical imagery.
4 I suggest the authors supplement the paper with an explanation of the applicability of the SMOGN algorithm to regression tasks, or add relevant clarifications.
5 I suggest the authors also provide the latitude and longitude information for the beaches corresponding to panels a-f in Figure 5.
6 It is suggested that the authors add relevant discussion and outlook in the discussion section regarding whether the proposed approach can be applied to the inversion of muddy tidal flat topography in the future.
Citation: https://doi.org/10.5194/essd-2025-826-RC1 -
AC2: 'Reply on RC1', nan xu, 14 Apr 2026
reply
Dear Reviewer 1,
Thank you very much for all comments and suggestions on this manuscript, and we have revised the contents and answered the questions item-by-item (the black font corresponds to the Questions and the blue font corresponds to the Answers). Please find the detailed responses in the attached file.
Best wishes,
-
AC2: 'Reply on RC1', nan xu, 14 Apr 2026
reply
-
RC2: 'Comment on essd-2025-826', Guoping Zhang, 24 Mar 2026
reply
Dear authors,
The study titled “NZ-BeachTopo30: A national-scale and full-coverage 30 m beach topography dataset for New Zealand reconstructed by fusing ICESat- 2 and Sentinel-2” presents a novel and robust methodology for generating topography for beaches and then develop a useful and national-scale dataset across New Zealand. Please find my revision suggestions below.
- I suggest adding one sentence at the end of the second paragraph in the introduction to summarize the current status and limitations of remote sensing techniques for beach topography reconstruction. Moreover, The first sentence of the third paragraph in the introduction should be revised to provide a general overview of mainstream DEM datasets, so as to improve the logical coherence of the paragraph;
- I suggest that ICESat-2 track information should be added in Figure 1 to better illustrate its spatial distribution across the study area;
- In Figure 2 workflow diagram, I suggest that the authors explicitly indicate that airborne LiDAR was used for validation; otherwise, it could easily be confused with the spaceborne ICESat-2 LiDAR data also employed in this study;
- The authors used composite remote sensing imagery of different quantiles to represent states at various tidal levels and claimed that this method can achieve denoising effects. I recommend the authors provide more quantitative evidence to demonstrate the rationality of this approach used in this study;
- I suggest the authors add boxplots for the X-axis and Y-axis corresponding data in panels a and b of Figure 14 in Section 5.2, to enable a more intuitive comparison. In addition, In Figure 14, it is evident that DeltaDTM systematically omits low-lying areas of beaches as well as regions adjacent to the ocean. I suggest that the authors emphasize this point in their discussion;
- I suggest that the authors add a discussion in this section addressing potential limitations of this study. For example, the use of OSM data to delineate beach extents may introduce errors inherent to the baseline data itself. Beaches undergo long-term evolution, and their temporal changes are also critical. Additionally, the 10 m spatial resolution may be insufficient for accurately representing narrower beaches. In summary, I would like to see the authors expand the discussion to include these considerations in the manuscript；
- This study presents a useful and large-scale dataset that fills an important gap in coastal elevation data for New Zealand. Hence, I suggest that the authors provide a more systematic explanation regarding the advantages, significance, and potential application scenarios of the data.
Citation: https://doi.org/10.5194/essd-2025-826-RC2 -
AC3: 'Reply on RC2', nan xu, 14 Apr 2026
reply
Dear Reviewer 1,
Thank you very much for all comments and suggestions on this manuscript, and we have revised the contents and answered the questions item-by-item (the black font corresponds to the Questions and the blue font corresponds to the Answers). Please find the detailed responses in the attached file.
Best wishes,
-
RC3: 'Comment on essd-2025-826', Anonymous Referee #3, 27 Mar 2026
reply
This manuscript by Wang et al. presents NZ-BeachTopo30, a 30 m beach topography dataset for New Zealand that fills intertidal data voids in DeltaDTM by fusing ICESat-2 photon-counting altimetry with Sentinel-2 multispectral composites via XGBoost. The problem is well-motivated — intertidal voids in global coastal DEMs are a genuine and widely recognised obstacle for coastal hazard assessment — and the dataset itself is a welcome contribution. The production workflow is clearly described, the validation framework is multi-layered, and the sea-level rise demonstration nicely illustrates the practical value of improved intertidal coverage. I have only a few minor comments.
The manuscript positions NZ-BeachTopo30 as "the first spatially continuous and full-coverage beach topography dataset for New Zealand." The national-scale application and the specific integration strategy with DeltaDTM are clearly valuable. That said, the ICESat-2 + Sentinel-2 + ML approach for coastal elevation estimation has been demonstrated in several publications, and XGBoost for structured regression is well established. The authors might consider framing the novelty more precisely around the data product and the fusion design rather than the methodology itself — this would make the contribution stand out more clearly on its own merits.
The 4,392 valid ICESat-2 pixels span 340 of 1,576 beach units (~22%). The transferability analysis in Section 4.4 is a real strength of the paper, and the modest degradation in accuracy (RMSE 0.93 → 1.02 m) is reassuring. It may be worth noting that this assessment can only be carried out where airborne Lidar is available, which itself tends to be concentrated in more accessible or higher-priority coastal segments. A brief acknowledgement of whether performance on remote, unsampled beaches might differ would give readers a more complete picture.
The internal model evaluation (Fig. 3) against ICESat-2 train/val/test splits usefully demonstrates model consistency, while the independent Lidar comparison (Fig. 4) provides the more informative accuracy diagnostic. It might be worth giving slightly more prominence to Fig. 4(f) — the accuracy specifically within the newly predicted intertidal voids (R² = 0.61, RMSE = 0.94 m) — since this is the metric most directly relevant to the paper's core contribution. The combined-product metrics (R² = 0.75, RMSE = 1.17 m) include the high-quality DeltaDTM backshore pixels and could, on their own, give readers a somewhat optimistic impression of the ML prediction component. Relatedly, the Lidar DEM was resampled from 1 m to 30 m via mean aggregation; on steeper beach faces this could introduce representativeness differences that inflate apparent disagreement. A brief note on this point would be helpful.
The SHAP analysis is well executed and adds useful transparency to the modelling. The main findings — NIR reflectance correlating positively with elevation, distance-to-coast tracking the cross-shore gradient — align with established coastal remote sensing understanding. The analysis is valuable precisely as confirmation that the model has learned physically sensible relationships. The authors might consider framing it in those terms rather than as a discovery of "physical driving mechanisms," which could slightly overstate what the interpretability exercise reveals.
Finally, the conversion from NZVD2016 to EGM2008 orthometric heights (lines 188–190) is an important step but is described only briefly. The difference between these two vertical reference surfaces varies spatially across New Zealand. It would be helpful if the authors could indicate the approximate magnitude and spatial pattern of this offset, so that readers can judge whether residual datum inconsistencies contribute meaningfully to the reported error budget.
Citation: https://doi.org/10.5194/essd-2025-826-RC3 -
AC4: 'Reply on RC3', nan xu, 14 Apr 2026
reply
Dear Reviewer 1,
Thank you very much for all comments and suggestions on this manuscript, and we have revised the contents and answered the questions item-by-item (the black font corresponds to the Questions and the blue font corresponds to the Answers). Please find the detailed responses in the attached file.
Best wishes,
-
AC4: 'Reply on RC3', nan xu, 14 Apr 2026
reply
-
RC4: 'Comment on essd-2025-826', Guoping Zhang, 14 Apr 2026
reply
I think the authors have answer all questions well. The manuscript has meet the requirements of ESSD and can be published.
Citation: https://doi.org/10.5194/essd-2025-826-RC4
Data sets
New Zealand National-Scale Beach Topography Dataset (30 m): A Fusion of ICESat-2 and Sentinel-2 Yuhao Wang, Hao Xu, Nan Xu, Edward Park, Xuejiao Hou, Jiayi Fang, Zhen Zhang, Yongjing Mao, Huichao Xin, Chunpeng Chen, Yinxia Cao, Yifu Ou, Xinyue Gu, Wenyu Li, Xiaojuan Liu, Conghong Huang, and Qingquan Li https://zenodo.org/records/17785546
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 320 | 146 | 26 | 492 | 31 | 37 |
- HTML: 320
- PDF: 146
- XML: 26
- Total: 492
- BibTeX: 31
- EndNote: 37
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
I thank the authors for a fantastic paper. The remote sensing method is really impressive, and I look forward to inspecting the outputs in more detail. I had a few general comments and areas that I think could be clarified in the paper:
1) The method relies heavily on the use of Sentinel-2 percentile composites to serve as machine learning covariates that reflectance tidal variability. However, due to persistent tidal biases associated with sun-synchronous sensors, satellites like Sentinel-2 rarely observes the same tidal conditions at different locations along the coastline (see Figure 8 in Bishop-Taylor et al. 2018 https://www.sciencedirect.com/science/article/pii/S0272771418308783, and Figure 7, Fitton et al. 2021 https://www.sciencedirect.com/science/article/pii/S2352938521000355). These biases mean that, for example, an 80th percentile composite in one location may observe high tide conditions, while the same 80th percentile composite in another location may only observe mid-tide. How does your approach handle these tide biases, and how do they affect the large-scale consistency of your results? There is currently a single sentence touching on tide variation issues in the manuscript ("First, regional variations in tidal regimes can alter the spectral-elevation relationship, affecting intertidal height retrieval") but I feel it needs to be elaborated on given the importance of these biases for large-scale coastal remote sensing analysis.
2) The paper focuses primarily on open coast beach environments, which is a valuable niche for intertidal elevation modelling that has not seen as much research attention as more sheltered, tide dominated systems. However, I feel it would be valuable to also include either include an example of model performance across more extensive tidal flat environments, or include some discussion points about how/why these intertidal environments were excluded from the study.
3) The data package includes only data for the "NoData" voids in the DeltaDTM dataset. Are there plans to also provide a combined DeltaDTM + NZ-BeachTopo30 data as a single seamless topobathy DEM? Or is the expectation that downstream users will combine these datasets themselves?