The authors present a global dataset of landforms derived from a high resolution DEM. They propose new ways to identify plain areas and their transition to hilly and mountainous terrain. The novel way to do this by identifying core areas and including transition areas through a cost distance analysis yields results that seem visually quite accurate when the map is overlayed onto a relief background. Plain and higher relief areas are neatly differentiated. This type of information can be quite useful for geographical and ecological macro studies. The precise workflow does miss details to be reproducible. It is a pity that proprietary software was used and the workflow described in general terms only, which makes replication more difficult. The choice for some cut off values or thresholds (slope, elevation, accumulated cost) is not always clearly explained or motivated.

To classify the hilly and mountainous areas the authors propose a new approach as an alternative to a moving windows analysis that has documented limitations. Landform relief is not calculated with reference to the nearest elevation data within a (small) window, but expressed with reference to a regional baseline calculated by creating a TIN on the basis of the elevation at the border of a mountain range (i.e. where it transitions to plain). In addition, the baseline elevation takes into account the elevation of points along water courses within the mountain to create a baseline surface to act as reference for the roughness calculations. Thresholds are applied to the elevation differences calculated by subtracting the baseline elevation from the actual surface elevation. The lowest elevation differences are labelled hills, followed by low relief mountain up to highest relief mountain. This leads to a conceptual problem. In my opinion, when one talks about a mountain or mountain range such as the Himalaya as a landform, one considers the mountain as a whole, from the foothills to the highest summits as the landform "highest relief mountain". Similarly when talking about the Jura or the Vosges mountains, one would talk about low relief mountains, but not consider only the mountain summits to be low, but the whole landform down to the foot slopes as being the low mountain.

In some cases the transitions from different categories of mountain to hilly land is well captured in this approach, typically in ancient eroded landscapes with remnants of higher mountains. The dissected rolling hill landscape gets the label hills, while the remaining inselbergs are classified as mountain.

However, the story is very different in younger mountain areas such as the European Alps or Himalayas. If one looks at the GBLU map without legend overlaid onto a relief map, valley-like shapes appear very distinctly that follow  the actual valleys of these mountains. When looking at the legend, one sees that these are actually classified as hills. The same holds for flat valley bottoms inside the mountains, these are classified plains, even if they are long, narrow and sinuous.

In my conceptualization of a mountain, the mid slopes of high mountains do not pertain to the landform class middle relief mountain. They are mid slopes of a high mountain. Similarly, mountain valleys are not hills, just because the local surface elevation is below a certain threshold.

When I look at the methods and results of this paper, I think of the product as something like "Map of relief classes and relative (or regional) elevation zones", and I am convinced that this classification is useful for different scientific applications. Ontologically I don't think that the presented map units should be thought of as representations of landforms.

In summary, I commend the authors for what seems to be a very detailed and precise work and the product and the work that has gone into its production. Also, the results seem to be useful for certain research applications. I do not however agree with the authors that what is represented here are landforms, ontologically speaking.

The distinction I make here is further illustrated in the figure.

Figure: Upper transect: how I understand the current version of the GLBU. Lower transect: how I think landforms should be conceptualized in this context.

Regarding the data availability, the authors have presented the resources they developed on Zenodo. The files are easily accessible and useable in open source software. Files are presented in folders by 10 degree latitudinal bands, and it is quite easy to find a region of interest. All terrestrial areas of the world seem to be included in the data. There is a possible issue for global level use of the data in that it consists of many different tiles that need to be mosaiced, but this can be coded.

The validation is done against a number of similar products where one of the main differences is the resolution of the source layers (DEM), this product being based on very high resolution sources (~30 m at the equator.) The identification of plains seems to be more accurate than in any of the products with which it is compared.

Overall the manuscript is sufficiently concise, the language clear, although it could benefit from some minor edits here and there (see below). On several points the methods section should be developed a bit further to allow full replication of the work flow.

Overall the language is clear and very understandable, but some suggestions for minor improvements are given.

As said, in my opinion the layers presented in this work do not represent landforms. However I think that the classification of relief in plains and mountains with different values of elevation and relief intensity (roughness) can be quite useful for a series of environmental applications.

My recommendation would therefore to revise the title and some sections of the text where the product is labelled as a map of landforms and replace this with formulations that more accurately reflect what is shown, that is, not to speak of landforms but about a map of relief (roughness) and elevation classes (or something similar) instead. This would require rather limited changes to the text and figures.

Specific comments

35-36: I would add evolution or genesis to this list of research subfields of geomorphology

43: I would add that field work is an essential component of landform mapping (geomorphology)

46-47: there is a more recent product produced by Amatulli et al. that might be useful to refer to here: Amatulli, G., McInerney, D., Sethi, T., Strobl, P., & Domisch, S. (2020). Geomorpho90m, empirical evaluation and accuracy assessment of global high-resolution geomorphometric layers. Scientific Data, 7(1), 162. https://doi.org/10.1038/s41597-020-0479-6

56-58: However, as the authors stated, unsupervised classification based methods to perform higher-resolution global landform classification require an international team with knowledge of geomorphological development in a variety of climatic and physiographic settings. > do you address this?

69-70: not clear if this paper only object is to classify the shape or also something about the material (lithology) and / or genesis, / evolution. Methods and final product seem to be focusing on shape irrespective of material / genesis.

80: objective:  "to construct a global classification system for landforms that integrates geomorphological knowledge," : not clear where the geomorphological knowledge comes in in the method

82: typo: "high-resolutiojn" > high-resolution

99-100  "The first-level (L1) types are defined as ‘plain’ and ‘mountain’, reflecting the most fundamental morphological characteristics of landforms." If I understand it well, the first level distinguishes between plain and non-plain (i.e. hills and mountains), as all that is not plain is later subdivided into several classes of hills and mountains, not mountains alone.

102:  "This classification perspective aids researchers in conducting macro-scale studies"  This is indeed a valuable distinction

113:  "the area the missing from FABDEM"  >  the area missing from FABDEM

120:  "The following sections provide details that should allow users to reproduce our results."  : some more details would be needed to achieve this I think

123: Fig 1:  "accumulate slope " > accumulated slope?

 "Interecting with flat landforms"  > Intersecting with flat landforms

 "Eliminating fragement blocks"  > Eliminating fragment blocks

125: data preprocessing or data pre-processing (see figure, perhaps harmonize?)

130:  "data from latitudes below 70° are transposed onto the Behrmann projection, and the remaining data are transported onto the Lambert azimuth equal-area projection. " : suggested edit: Tiles between 70° N/S are reprojected to the equal area Behrmann projection, and the tiles polewards of 70° N/S to Lambert azimuthal equal-area.

132-133: this first sentence is more of a statement that would perhaps be better in the introduction. Starting this section with the second sentence works quite well.

140: Fig 2b  typo: "variant" > variant

147: how large must the continuous area of plain be to be considered a core area? I.e. how many contiguous pixels constitute a plain core area? Do you also apply a shape criterion, or can a very long area of contiguous plain pixels also constitute a core plain area?

148-150: it is not clear to me what the cost layer is in this calculation: elevation, slope, or something else? Same holds for 'cost' in Fig 2a.

149:  "The AS is calculated as the minimum cumulative cost of each position to the nearest landform core along a specific path"  Would it not be more precise to say: The AS is calculated as the minimum cumulative cost of each position to the nearest plain core along a specific path.

155-156: not clear to me how such an algorithm   achieves the most direct integration of geomorphological knowledge and expertise

160: does T2 have a dimension and a unit? 1500-2000, is that length in meters, or slope in degrees or something else?

161-162:  "but needs to be determined by integration with expert knowledge within different geomorphic regions".  Not clear if you state that this should be done or that it has been done, and if so how?

162:  "In some cases, it may exceed the recommended threshold range."   – not clear where and when

165-167:  "This novel method avoids the negative effect of local window analysis and is beneficial for maintaining the landform semantics for each block."  Visual inspection of a number of tiles indeed shows a neat identification of the borders of plains and their transition to hilly or mountainous terrain.

176-177:  "a method that fails to account for geomorphological semantics, and which therefore disregards the integrity of a mountain. " I would argue that the classification of L2 landforms proposed in this paper does just that. I do not see any landform concept reflected in the classes, and even less so in the map units corresponding to these classes. See general comments above

192:  "on basis of the plain boundary"  > on the basis of the plain boundary

192-193:  "To refine the representation of surface relief, we also take into account linear features representing the rivers. " I suppose you do not consider all rivers and streams to construct your TIN of mountain base. Rivers and streams go up to great altitudes. Which sections of mountain rivers did you consider to construct the TIN?

206: was there any reasoning behind the selection of these elevation bands? 0-1000, 1000-3500, 3500-5000 and >5000?

207-208: idem

277: Figure 7. Comparison between the GBLU and the Global Mountain Biodiversity Assessment (GMBA) projects. > Figure 7. Comparison between the GBLU and three mountain definitions presented on the Global Mountain Explorer (https://rmgsc.cr.usgs.gov/gme/)

278-279: this does not seem to be entirely accurate:  "We conducted a more detailed comparison for mountain regions using the Global Mountain Biodiversity Assessment (GMBA) (Snethlage et al., 2022) as reference data."  The three definitions are from three different institutions (WCMC, GMBA and USGS) but have conveniently been presented together on the Global Mountain Explorer (https://rmgsc.cr.usgs.gov/gme/). The latest mountain definition is the one by Snethlage et al (2002) which can be obtained from https://www.earthenv.org/mountains (scroll down to: Download the GMBA Mountain Definition v.2 here.)

337:  "fundamental role in supporting the identification of landforms that incorporates complex semantics."  > not clear what semantics means in this context

344  "influencinge community structure and function,"  > influencing community structure and function,

The authors introduce a global landform classification dataset (GBLU) that represents a significant advancement in resolution compared to existing global geomorphological data. Their three-levels classification system with 26 distinct landform classes demonstrates an approach to categorizing Earth's surface features. The use of 1 arc-second DEMs provides unprecedented detail at the global scale, and their methodology of combining geomorphological ontologies with key derivatives appears to effectively balance noise reduction while preserving important landform characteristics.

However, a notable limitation is the lack of a fully documented methodological scripting procedure (even an example code would be helpful) to enable complete reproducibility of the results. Several Python libraries, such as rasterio, pyjeo, xarray, and numpy, along with GRASS GIS modules, offer matrix filtering procedures and cumulative cost analysis that could facilitate the replication of the methodology in a more transparent way.

The full methodology (AS, TIN, SUI) is novel; however, several issues arise during the processing phase due to the absence of a computational scripting framework that would enhance the rigor of the geocomputation procedure.

Below are some geocomputation issues identified in the manuscript:

Data pre-processing

To reduce projection distortion, the authors state:

"Data from latitudes below 70° are transposed onto the Behrmann projection, while data above this threshold are projected onto the Lambert azimuthal equal-area projection."

This approach is reasonable; however, an overlap between the two projection zones is necessary to avoid border effects.

Methodology

Figures 1, 2, and 3 are well designed and effectively illustrate the methods. However, they are not supported by a scripting procedure that can be followed step by step. Additionally, several thresholds (e.g., Tas, Tss) are defined in the methodology but appear to be based on empirical, subjective decisions. It would be preferable to define them using statistical or mathematical criteria.

Figures 5–7 are well presented, but it would be beneficial to show the GBLU classification results alongside a transect, similar to Figure 3c, but using real relief data.

Due that the post-processing includes several aggregation/smoothing procedure do you really need to use a 1 arc-second DEM?

Would be more effective to use 3 arc-second MERIT Hydro in combination with the stream-network Hydrography90m to have a landform classification more in line with existing DEM-derived products?

Projection

The manuscript states: "Data from latitudes below 71° are transposed onto the Behrmann projection, while data above 69° are projected onto the Lambert azimuthal equal-area projection." However, WGS84 (World Geodetic System 1984) is a geodetic datum and can be represented using either a geographic coordinate system (latitude/longitude, expressed in degrees) or a projected coordinate system (e.g., UTM). The final tif files appear to be stored in the latter, but no specific explanation is provided in the manuscript.

Are the final tif files stored under two separate projections, or have they been homogenized into a single projection? Either approach is valid, but this should be explicitly stated in the manuscript and in the README.txt file available in the Zenodo repository.

Additionally, the processing appears to be done in 10° × 10° tiles. What happens at the tile borders? Is there an overlapping procedure in place?

tif files

The inclusion of tif file overviews (*.ovr) and a color table palette is appreciated, as they facilitate fast and visually informative rendering. However, it would be useful to include the code legend as metadata within the tif files themselves or at least document it in the README.txt file.

The .aux.xml files store statistical information about the tif files (e.g., mean, median). However, since the tif files contain categorical variables, this statistical information is not particularly useful.

I suggest increasing the grid tile size of the final tif files to 2° × 2° (or even 4° × 4°) to reduce the total number of files. This would simplify tile management, especially for large-scale downloads.

Dear Editor and Authors,

I read the paper “Global basic landform units derived from multi-source digital elevation models at 1 arc-second resolution”. There are some interesting aspects, but even if it a technical/data paper there is the need o improvements. Apart from the description of the methodology that is unclear, I think that there are many drawbacks in the paper that require a full restructuring of the work. First, the landforms classification is too simple and in no way reflects the complexity of landscapes. For example, the approach of Iwahashi et al. uses much more information, for example the texture of terrain (even if with a simplified index). The comparison with other methods is debatable both for the different rational behind some methods as well as for the different resolutions. You should at least apply those methods on the same DEMs you used with your approach. Here I suggest some references, to which I refer in the following more detailed comments.

Suggested references

Guth, P.; Kane, M. Slope, Aspect, and Hillshade Algorithms for Non-Square Digital Elevation Models. Transactions in GIS 2021, 25, 2309–2332, doi:10.1111/tgis.12852.

Fisher, P.; Wood, J.; Cheng, T. Where Is Helvellyn? Fuzziness of Multi-Scale Landscape Morphometry. Transactions of the Institute of British Geographers 2004, 29, 106–128.

Trevisani, S.; Guth, P.L. Terrain Analysis According to Multiscale Surface Roughness in the Taklimakan Desert. Land 2024, 13.

Minár, J.; Drăguţ, L.; Evans, I.S.; Feciskanin, R.; Gallay, M.; Jenčo, M.; Popov, A. Physical Geomorphometry for Elementary Land Surface Segmentation and Digital Geomorphological Mapping. Earth-Science Reviews 2024, 248, doi:10.1016/j.earscirev.2023.104631.

Lindsay, J.B.; Newman, D.R.; Francioni, A. Scale-Optimized Surface Roughness for Topographic Analysis. Geosciences (Switzerland) 2019, 9, doi:10.3390/geosciences9070322.

Guth, P.L.; Trevisani, S.; Grohmann, C.H.; Lindsay, J.; Gesch, D.; Hawker, L.; Bielski, C. Ranking of 10 Global One-Arc-Second DEMs Reveals Limitations in Terrain Morphology Representation. Remote Sensing 2024, 16, doi:10.3390/rs16173273.

Specific comments (A: author R: reviewer)

A:

Lines 67- 69 and also lines 72-74  “Nevertheless, higher DEM data resolution can be regarded as a double-edged sword, in that it at once provides the opportunity for landform mapping at a finer scale while at the same time increasing the challenge of reducing the noise effect (Jasiewicz and Stepinski, 2013) and maintaining the integrity of the identified landforms.”

R: I think that the referred problem of noise related to high resolution is a false problem. Apart from the ambiguity of the term “noise” (e.g., noise because of errors in the digital representation, or because you consider noise the fine-scale morphology?), multi-resolution approaches permit to analyze the landscape having control of the “noise” (independently from the interpretation). In addition, surface texture analysis should be an important component of landscape segmentation approaches (as Iwahashi et al.  or Jasiewicz and Stepinski, 2013) and can be particularly informative when computed at higher resolutions than global DEMs. Apart from the papers you cited I would consider the ones from Fisher Lindsay  and Trevisani  

A:

Lines 77- 79 “We focus on the classification of basic landforms that emphasizes morphological differences and, in so doing, we present the practical expression of landform ontology at the global scale that offers valuable insights into the Earth’smsurface structure comprising the constellation of landform types and their boundaries.”

Lines 80-82. “The objectives of this research are: (1) to construct a global classification system for landforms that integrates geomorphological knowledge, (2) to design a novel framework for global basic landform classification, (3) to develop an automated classification and mapping model for global landforms, and (4) to make available a comprehensive high-resolutiojn dataset of global landform units”

R:

I have the feeling that the stated objectives of the research are only partially covered. In regard to 1, I don’t see big integration with geomorphological knowledge. In regard to point 3, you are just mapping very simple aggregates of landforms (mountain, hill, plain) that do not represent the complexity of landforms. I think that the work of Iwahashi should be considered the starting point for new approaches, maybe considering additional geomorphometric derivatives. But just working with elevation, even if the algorithm could be interesting, does not seem a step forward and very useful practically. Finally, in regard to (4) I don’t think that term “high resolution” can be used with something derived from global DEM at 1 arcsecond resolution.

A:

Lines 91-100 

R: The motivations behind the derivation of the simple classification scheme are unclear and someway highly debatable. I don’t feel that it is a big deal to just subdivide between mountains, hills and plains. In addition, on the fuzziness of landforms perception and classification I surely would consider the work of Fisher et al..

A:

Lines 107-111 “In this work, the ‘Forest and Buildings removed Copernicus DEM’ (FABDEM) (Hawker et al., 2022) is the primary data for latitudes 60°S-80°N…”

R:

I would be more cautious or at least I would discuss more the selection of FABDEM instead of COPDEM, because some geomorphometric derivatives, are better represented in COP.

See for example Guth et al. In addition, another question is whether structures should be removed in urban landscapes or not.

A:

Line 117 “knowledge-guided framework….” 

R: how? I don’t see a relevant integration with expert knowledge.

A:

Line 119 “calculation of the mountain uplift index (SUI)” 

R: 

I feel that the name “uplift index” is ambiguous, it seems to imply some tectonic uplift. Moreover, see also later comment, it seems a local relief measure.

R: Line 121 What is “factor calculation” ?

A:

Figure 1, workflow and lines 128-130 “Meanwhile, due to the requirement of calculating landform derivatives, we determine the projection principles as follows: data from latitudes below 70° are transposed onto the Behrmann projection, and the remaining data are transported onto the Lambert azimuth equal-area projection. “

R: To work in a projected system is not a requirement but a choice. In every case if you project DEMs you should discuss all the related intricacies and approximations. See for example Guth and Kane.

R: Figure 2 and related caption. I think it is really difficult to understand how the AS works.

Also the description at lines 149 -160 is unclear to me: “The AS is calculated as the minimum cumulative cost of each position to the nearest landform core along a specific path…” 

How is computed cost? The cost of doing what? I don’t see how geomorphological knowledge enters in the method, it seems an heuristic approach.

A:

Lines 176-178 “However, commonly employed indices reflecting topographic relief are achieve using a window of fixed size such as 3×3, 5×5 pixels, or larger (Maxwell and Shobe, 2022), a method that fails to account for geomorphological semantics, and which therefore disregards the integrity of a mountain. Window size has a significant impact on results of relief calculation.”

R: but adopting multiscale approaches this issue can be resolved.

A:

Line 183 “In quantitative analysis, it is crucial to consider the underlying terrain of mountains to accurately assess changes in elevation.” 

R: unclear.

A:

Lines 185 “surface uplift index (SUI)” 

R: your index seems a local relief index on which there is a huge literature  (see for example Minar and cited reference therein…).

A:

Lines 188-189 “SUI considers the vertical elevation differences between the surface and the mountain base, which is more consistent with the human perception of mountain morphology.”

 R: The human perception is multiscale, so it just depends from the target of the analysis.

R: Lines 190-203. Not able to follow.

A: Lines 241-242 “Figure 4 shows the global landform classification results based on the abovementioned framework. This hierarchical dataset provides a more comprehensive understanding of the Earth surface” 

R: A more comprehensive with respect to which method? Or with respect to which reference dataset? Honestly the earth’s surface is a little bit more complex. Apart from the issues with deserts you mention, for instance big depressed areas or volcanic environments are not represented.

A: 

Lines 244-245 “The selected regions contain examples of the main landforms on Earth, as well as transition areas of different landforms.” 

R: Yes, in the selected regions there are interesting patterns, but your approach does not characterize/distinguish these.

A: “The abundant textural information provided by GBLU” 

R: I don’t see how your approach contains textural information in the sense of Iwahashi or Trevisani.

A: 259 “significant improvement achieved by applying GBLU is the increased detail in representing terrain features.” 

R: I see a very simple representation of landforms, but any indicator of patterns/texture is totally missing.

R:

Section 3.2 

This section has a lot of issues. You need to describe reference data (refdata) in the text not in the captions. Most importantly, it does not make too much sense to compare classifications performed at different resolutions or with different DEMs, given the different generalization levels of the landscape. Regarding Iwahashi you could apply the method to the same data you used in the analysis (if I’m not wrong it is implemented in SAGA). In addition, the method of Iwahashi et has been designed to take into account different aspects of morphology, including texture. It is not just based on elevation and slope.