Dear Reader,

Thank you very much for your interest and constructive feedback!

To address your questions, as explained in the manuscript, both the point-based and polygonal datasets are distributed as spatial tiles following the grid scheme illustrated in the Zenodo and GitHub repositories (https://zenodo.org/records/16755044/preview/UGLC_tile_grid_map.jpeg?include_deleted=0). 

Specifically, the global tiling system is divided into105 tiles: 15 longitudinal steps and 7 latitudinal steps.

But, empty tiles that fall entirely within regions for which no available open landslide data source were available, are automatically excluded from the repository.

For this reason, regions such as China (where open, integrable landslide datasets were currently limited or unavailable) do not yet have data into the UGLC catalog, and consequently  corresponding tiles in the UGLC repository.

In future updates, if some open point or polygonal landslide data for China will become available and suitable for integration in the UGLC, new tiles will be generated for those areas and included in the corresponding catalog, and available on UGLC zenodo repository.

To ensure efficient data logistics, accessibility, and consistency with the heterogeneous and often discontinuous spatial availability of source data (as is the case not only for China, but also for parts of Russia, Central Africa, and South America), downloads are currently organized exclusively by tiles containing available data, rather than by geographic regions or national boundaries. If future UGLC releases will achieve a more homogeneous global coverage for both point and polygon datasets, we will certainly consider implementing alternative download options, such as by country or continent.

Regarding your observation about some landslide vectors appearing to coincide with road networks, we would like to clarify that potential geographic coordinate offsets or spatial misalignments are explicitly addressed in the point-based catalog through the ACCURACY attribute.

As described in the manuscript, this attribute provides an estimate of positional uncertainty (where possible) based on ancillary information (if available) from the original source catalogs and interpretable as proxies for geospatial error.

The uncertainty is indexed on a scale from 1 (minimum error) to 10 (maximum error), allowing users to account for and filter data according to their specific accuracy requirements.



We thank you again for your thoughtful comment and hope this clarification addresses all your concerns.

Sincerely,

The Authors

We sincerely thank the reviewer for the rigorous and conceptually deep evaluation of our manuscript. 

We particularly appreciate the emphasis on epistemological heterogeneity, which we recognise as a fundamental aspect of global landslide data integration.

We clarify that the UGLC does not produce a scientifically homogeneous dataset in an observational sense. Indeed it is based on the sampling strategy endorsed in each single original database. The novelty of the work is structural and ontological, not empirical. In the revised manuscript, we have explicitly clarified that UGLC is an organisational and semantic integration framework, rather than a homogenised dataset suitable for indiscriminate quantitative use. Importantly, we emphasise that the catalogue preserves the epistemological context of each source dataset through explicit provenance attributes (e.g., OLD_DATASET, OLD_ID, TYPE, PHYSICAL_FACTORS). Landslide records remain valid observations within their original acquisition context, and UGLC does not attempt to reinterpret or generalise them beyond their native meaning. To address the reviewer’s concern regarding potential misuse, we have introduced a dedicated section clarifying intended and non-intended uses. In particular, we explicitly state that:

• the catalogue is suitable for data discovery, exploratory analyses, and large-scale descriptive studies; 
• modelling applications require careful, user-driven filtering based on trigger type, landslide classification, spatial/temporal accuracy, and dataset provenance; 
• the dataset should not be used as a statistically homogeneous input without preprocessing. 

These additions are reflected in the revised manuscript ( line 44-51 section: “1. Introduction”; section: 1.2 “Emerging need for harmonised landslide data”; section:  1.3 Improving geospatial aspects of landslide catalogues”; line 384-390  section: “5.1 Analysis and statistical insights on UGLC”;  section: “5.2 Epistemological limitations and intended use”; section: “7.4 Landslide models and predictive applications”;  section: “8. Conclusion” )

About the semantic normalisation and methodological transparency, we have substantially expanded the methodological description and the normalisation process is now clarified as follows:

• Attribute harmonisation is driven by dataset-specific JSON lookup tables, available in the public GitHub repository; 
• These lookup tables were manually curated through a systematic, dataset-by-dataset analysis, ensuring semantic consistency while minimising interpretative bias; 
• The process is intentionally conservative, aiming to standardise existing information (e.g., resolving naming inconsistencies, typos, case sensitivity issues) rather than infer new attributes; 
• Interpretative steps were limited and applied only where sufficient contextual metadata supported them; 
• In the absence of reliable information, attributes are explicitly marked as ND or -99999, preserving the integrity of the original data. 

This clarification reinforces that the harmonisation process does not artificially homogenise the data, but instead regularises their representation.

These additions are reflected in the revised manuscript ( section: “4.3 Semantic standardisation and attribute harmonisation”;  section: ”4.5 Data normalisation”)

On duplicate detection and dataset overlap we have clarified that duplicate records are identified using a strict matching criterion based on: identical spatial coordinates and identical temporal attributes. When duplicates are detected, they are interpreted as informational redundancy across datasets. In such cases the most informative record is retained, to have fully identical records reduced to a single entry. This approach is conservative and may not detect near-duplicates, prioritising traceability and reproducibility over aggressive deduplication.

These additions are reflected in the revised manuscript ( section: ”4.4 Data merge”)

About the reliability, uncertainty, and completeness, we agree that these aspects require clearer clarification. RELIABILITY attribute is now explicitly defined as a relative confidence ranking, based on spatial and temporal precision not a quantitative uncertainty measure, and has not been independently validated as a statistical accuracy metric We also explicitly state that:

• no independent error estimation or uncertainty propagation is performed; 
• completeness is inherited from source datasets and is rarely explicitly quantified in the source datasets; 
• UGLC does not guarantee completeness and should not be interpreted as such. 

A formal disclaimer on data completeness has been added. 

These additions are reflected in the revised manuscript ( line: 185-197 section: ”3.2 Attributes structure and specifics”)

Regarding the missing tiles we thank the reviewer for identifying the issue. This was caused by a technical inconsistency in the tiling workflow and did not affect the master dataset. The issue has now been resolved, all tiles have been regenerated and verified and all polygon data are distributed as a global CSV to ensure full accessibility (Broken external links have been updated or removed where necessary).

On modelling claims we fully acknowledge reviewer’s concern and have revised the manuscript to moderate claims related to modelling applications. The text now clearly states that UGLC is designed to support large-scale analyses, but requires context-aware filtering and domain expertise for modelling purposes. These additions are reflected in the revised manuscript ( line 44-51 section: “1. Introduction; section: “5.2 Epistemological limitations and intended use”; section: “7.4 “Landslide models and predictive applications; section: “8. Conclusion”)

We thank the reviewer again for highlighting critical conceptual aspects. The manuscript has been substantially revised to improve clarity, transparency, and correct scientific positioning of the dataset.

We thank the reviewer for the constructive and insightful feedback, and for recognising the potential value of the UGLC dataset.

- On methodological transparency:

We have significantly expanded the description of the normalisation workflow. In particular, the semantic harmonisation is performed through dataset-specific JSON lookup tables, publicly available in the GitHub repository. These tables were manually curated through a systematic review of each dataset to ensure scientific consistency and minimal interpretative bias. The process focuses on standardising existing information, rather than inferring new attributes; while missing or unreliable information is explicitly preserved using standardised no-data values. 

Moreover, we also clarified:

• the duplicate detection criteria (exact match on coordinates and date); 
• conservative handling of spatial accuracy inference; 
• the preservation of every original geometries for both point and polygon datasets.

These additions are reflected in the revised manuscript (line: 180-197 section: “3.2 Attributes structure and specifics”; section: “4.2 Data conversion”; section: “4.3 Semantic standardisation and attribute harmonisation”; line: 265-269 section: “4.4 Data merge”; line: 323-333 section: “4.5 Data normalisation”)

- On reliability classification:

We clarified that the RELIABILITY attribute is a ranking system, not a probabilistic metric. It is designed to support data filtering based on practical thresholds reflecting typical geospatial precision levels. We also clarified how this metric is applied mainly to point data. While polygon data are considered spatially accurate by definition, but may or may not be temporally accurate. Consequently, polygons can have a RELIABILITY value of 1 or 2.

Regarding the treatment of temporal uncertainty when only partial dates are available every case was explained in (line: 310-321 section: ”4.5 Data normalisation”).

We also clarify that a global validation (e.g., via multi-temporal satellite imagery) would require a whole dedicated effort beyond the scope of this study, and this limitation is now explicitly stated in the manuscript. These additions are reflected in the revised manuscript ( line: 185-197 section: ”3.2 Attributes structure and specifics”; section: ”5.2 Epistemological limitations and intended use”)

- On dataset imbalance and bias:

We have added a dedicated subsection explicitly addressing:

• geographic imbalance; 
• bias related to trigger types (rainfall and seismic); 
• implications for statistical analysis and modelling. 

We clarify that these biases reflect data availability and reporting practices, rather than global landslide occurrence patterns. These additions are reflected in the revised manuscript (line: 384-390 and 405-410 section: ”5.1 Analysis and statistical insights on UGLC”; section: “5.2 Epistemological limitations and intended use”)

- On validation and scientific utility:

We agree that demonstrating utility is important. A direct comparison with existing global datasets (e.g., GLC, NASA COOLR) would be largely redundant, as these datasets are already included within the UGLC (e.g., GLC is incorporated into NASA COOLR, which is part of UGLC). Moreover, independent validation falls beyond the scope of this data paper and would require a dedicated effort. The original datasets were curated and published by their respective providers, often through peer-reviewed or institutional frameworks, and are therefore considered to have undergone prior quality control. To address the reviewer’s concern, we have added an exploratory analysis comparing UGLC point data with global rainfall intensity (NASA GPM-IMERG, 2000–2025) and seismic hazard (peak ground acceleration from the Global Earthquake Model – GEM, 10% probability of exceedance in 50 years, Vs30 = 760–800 m/s). This comparison highlights the expected spatial clustering of landslides in regions characterised by intense rainfall and seismic activity, and helps illustrate inherent biases in the dataset; as well as its practical and scientific utility. In addition, we have:

• clarified possible applications and their limitations;
• explicitly framed UGLC as a data infrastructure rather than a validated modelling dataset.

These additions are reflected in the revised manuscript ( line 44-51 section: “1. Introduction”; section: 1.2 “Emerging need for harmonised landslide data”; section:  1.3 Improving geospatial aspects of landslide catalogues”; line: 384-390 and 405-410 section: ”5.1 Analysis and statistical insights on UGLC”; section: 5.2 “Epistemological limitations and intended use”; section: 7.4 “Landslide models and predictive applications”; section: “8. Conclusion”).

Thanks to your suggestions the manuscript has been strengthened through improved methodological transparency, clearer justification of assumptions, and a more explicit discussion of limitations and bias.