the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 31 Jul 2026
Automated Homogenisation of Early Instrumental Temperature Data at the Global Scale Using the HCLIM Dataset
Abstract. Early instrumental observations provide a unique opportunity to investigate climate variability before the establishment of modern meteorological networks. However, their effective use is often hindered by incomplete records, uneven spatial coverage, and non-climatic inhomogeneities. The HCLIM dataset provides a recently compiled global collection of early instrumental temperature records, yet systematic evaluations of automated homogenisation methods for such data remain limited. Here, we evaluate three widely used homogenisation methods, CLIMATOL, PHA, and BART, applied to the HCLIM dataset. The assessment considers data retention following pre-processing, breakpoint detection in the homogenised series, and agreement with the Twentieth Century Reanalysis version 3 (20CRv3), using French and Southeast Asian station networks as representative case studies. The pre-processed datasets differ only modestly in overall structure, although BART retains fewer records (80 %) than CLIMATOL (96 %) and PHA (99 %) because of its stricter requirement for a minimum record length of 15 consecutive years. Consequently, BART preserves longer average record lengths and detects substantially more breakpoints, indicating greater sensitivity to potential inhomogeneities. Breakpoint occurrence varies considerably across regions and time periods, reflecting differences in station density and observational coverage. Comparisons with 20CRv3 show generally high agreement for all methods, with stronger consistency in the dense French network than in the sparser Southeast Asian network. Despite differences in data retention, breakpoint detection, and adjustment strategy, the homogenised datasets show strong agreement with one another and preserve the principal climatic signal. Notably, differences among the homogenisation methods are consistently smaller than the differences between the original and homogenised datasets, indicating that the primary challenge is not selecting a single optimal algorithm but reconstructing temporally coherent climate information from incomplete historical observations. The homogenised datasets produced in this study constitute HOM-HCLIM, a new global database of homogenised early instrumental temperature records that provides an important resource for climate reconstruction, reanalysis, and long-term climate variability studies.
- Preprint
(1243 KB) - Metadata XML
-
Supplement
(909 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3936', Jose A. Guijarro, 26 Aug 2026
-
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
Dear Jose Guijarro,
Thank you very much for your careful and constructive comments. It is especially nice to hear from you, and we greatly appreciate that you took the time to read and comment on our manuscript. Your detailed comments on CLIMATOL are particularly valuable and will help us improve the methodological accuracy and clarity of the paper. We will carefully address all the points raised in the revision.
Best wishes,
Elin, on behalf of the authorsCitation: https://doi.org/10.5194/egusphere-2026-3936-AC1
-
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
-
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
Dear Jose Guijarro,
Thank you very much for your careful and constructive comments. It is especially nice to hear from you, and we greatly appreciate that you took the time to read and comment on our manuscript. Your detailed comments on CLIMATOL are particularly valuable and will help us improve the methodological accuracy and clarity of the paper. We will carefully address all the points raised in the revision.
Best wishes,
Elin, on behalf of the authorsCitation: https://doi.org/10.5194/egusphere-2026-3936-AC1 -
RC2: 'Comment on egusphere-2026-3936', Anonymous Referee #2, 05 Oct 2026
- Overall evaluation
This manuscript presents an application of three automated homogenisation approaches, CLIMATOL, PHA, and BART, to 3,632 monthly temperature series from 125 countries spanning 1658–2021. The study compares method-specific data retention, breakpoint detection, temperature-series agreement, consistency with 20CRv3, and station-level RMSD. It also introduces HOM-HCLIM as a new global collection of homogenised early instrumental temperature records.
The work addresses a highly important problem. Early instrumental observations are indispensable for extending the observational climate record, but their use is complicated by incomplete temporal coverage, sparse reference networks, changes in instruments and sites, inconsistent observation practices, and insufficient metadata. The manuscript is therefore timely and potentially valuable.
The paper has several clear strengths global application to a large and historically important observational archive; comparison of three widely used approaches; explicit documentation of method-specific retention rates; investigation of dense and sparse station networks; reporting of breakpoint frequencies normalized by station-years; use of an external reanalysis as a consistency reference; public availability of substantial parts of the processing code; intended release of a reusable homogenised dataset.
However, the current evaluation demonstrates method agreement, modification magnitude, operational feasibility, and consistency with 20CRv3 more convincingly than it demonstrates homogenisation accuracy. This distinction is important because neither original HCLIM nor 20CRv3 is an error-free station-level reference. The manuscript itself acknowledges that HCLIM contains gaps and varying temporal coverage, and that 20CRv3 has coarse spatial resolution and considerable uncertainty during its early period.
The paper could become publishable, particularly if the authors: clarify the construction of HOM-HCLIM; moderate claims of accuracy and robust recovery; strengthen the evaluation with a controlled or withheld-observation experiment; redesign or fully explain the statistical comparisons; quantify reconstruction uncertainty; correct the verified internal inconsistencies.
Major comments
- The manuscript should distinguish method intercomparison from validation of homogenisation accuracy. The study evaluates the homogenised outputs by comparing them with the original HCLIM series; the other homogenised products; 20CRv3; station-level RMSD from HCLIM; selected regional and station examples.
These are informative comparisons, but none constitutes a fully known homogeneous target. High agreement with HCLIM indicates that the broad characteristics of the original series are preserved. It does not prove that non-climatic discontinuities have been detected and corrected accurately. Similarly, agreement among the three methods does not prove correctness because the products are derived from substantially the same observational series and reference networks. Agreement with 20CRv3 establishes external consistency, but 20CRv3 is not an error-free station-scale reference, particularly during the early instrumental period. See: https://www.sciencedirect.com/science/article/pii/S0022169422001603?via%3Dihub
The authors should distinguish explicitly among fidelity to the original observations; magnitude of homogenisation adjustment; agreement among methods; consistency with reanalysis; accuracy in detecting and correcting inhomogeneities. Unless an evaluation with known breakpoints or withheld observations is added, the term “performance evaluation” should be qualified as an intercomparison and consistency assessment.
I suggest the authors rewrite the statement as:
We compare the operational behaviour and consistency of three automated homogenisation approaches when applied to HCLIM. Because no complete and error-free reference exists for the early instrumental period, the analyses assess relative agreement and plausibility rather than absolute homogenisation accuracy.
- The manuscript’s largest methodological limitation is the lack of an experiment in which the target value or breakpoint is known. This does not necessarily invalidate the dataset-construction exercise because benchmarking studies already exist in the literature. However, it limits manuscript-specific claims that robust signals have been recovered and that particular breakpoint patterns reflect genuine inhomogeneities.
The authors should add at least one of the following.
- Artificial-breakpoint experiment
- Select relatively complete station networks.
- Introduce synthetic breakpoints of known timing and magnitude.
- Introduce realistic missing-data patterns.
- Run CLIMATOL, PHA, and BART without supplying the known answer.
- Evaluate detection rate, false-positive rate, timing error, adjustment error, trend error, and variance recovery.
- Withheld-observation experiment
- Remove known observations from otherwise complete records.
- Create gaps of different lengths.
- Reconstruct the withheld values.
- Compare reconstructed and observed temperatures.
- Stratify performance by gap length, network density, neighbour correlation, elevation, and region.
The second option is especially relevant because long gaps are reconstructed at Bourges and Hanoi.
- The construction of the final HOM-HCLIM data product is insufficiently explained.
The manuscript states that the homogenised datasets are combined to develop HOM-HCLIM. The code and data section then refers to “the best homogenised dataset.” However, the manuscript does not define how the final product is generated from the CLIMATOL, PHA, and BART outputs.
It is unclear whether the final dataset contains all three method outputs; averages the three outputs; selects one method globally; selects one method for each station or network; uses an ensemble median; selects the series with the lowest RMSD; applies a hierarchy where one method is unavailable; combines observed values with reconstructed values from different methods.
Add a dedicated Methods subsection titled Construction of the HOM-HCLIM dataset.
This subsection should explain what data variables HOM-HCLIM contains; whether all three method outputs are retained; how the recommended series is selected or calculated; how missing outputs are handled; how conflicting breakpoints are reconciled; whether values are observed, adjusted, or reconstructed; whether uncertainty or method spread is included; what “best homogenised dataset” means; how the final product can be reproduced from the code.
The released data should ideally include original HCLIM value; each method’s homogenised value; recommended value, if one is provided; observation versus reconstruction flag; breakpoint and adjustment flags; number of available reference stations; intermethod spread; quality category. Without this information, the central data product is not scientifically transparent.
- The methods are applied globally, but the main comparison with 20CRv3 is based on French and Southeast Asian networks, supplemented by Paris, Jakarta, and eight selected stations for RMSD analysis. The abstract appropriately identifies France and Southeast Asia as representative case studies. Nevertheless, several conclusions generalize broadly to the global HOM-HCLIM dataset.
France and Southeast Asia provide a useful contrast between relatively dense and sparse networks, but they do not represent all observational conditions across 125 countries. The global archive includes continental, maritime, high-elevation, polar, tropical, coastal, and isolated island stations. The manuscript also reports that BART could not be applied successfully to the Alaska and Pacific networks, demonstrating that applicability varies geographically.
The authors should either expand the detailed evaluation to additional representative networks; or qualify global conclusions and emphasize that performance is evaluated through selected regional case studies.
Additional evaluation regions should ideally include a dense continental network; a tropical sparse network; a high-latitude network; a Southern Hemisphere network; an isolated maritime network; a network containing exceptionally early records.
- Method-specific preprocessing confounds the comparison among homogenisation algorithms. The three methods do not operate on identical input datasets: PHA excludes only two records; CLIMATOL excludes 142; BART excludes 740; BART requires at least 15 consecutive years; PHA uses Network System A; CLIMATOL and BART use an auxiliary Network System B that is subsequently recombined.
Therefore, differences among the resulting products may arise from: the homogenisation algorithm; input-series selection; network definition; record-length requirements; available neighbours; parameter settings. The paper recognizes preprocessing differences but later discusses output differences primarily as differences among homogenisation methods.
The authors should separate two questions: Operational comparison: How do the methods perform under their recommended or feasible configurations?
Common-input comparison: How do the algorithms differ when supplied with the same stations, common temporal support, and comparable network boundaries?
The operational comparison reflects real use. The common-input comparison isolates algorithmic differences. Both are scientifically valuable, but they answer different questions. At minimum, the authors should repeat the primary comparisons on the station-period intersection common to all three methods.
- The manuscript reports 14,851 breakpoints for BART, 3,186 for CLIMATOL, 2,676 for PHA, and 11,888 from SNHT applied to HCLIM. It also reports normalized breakpoint rates per 1,000 station-years, which is a useful improvement over comparing raw counts alone. Nevertheless, the methods differ in retained stations, temporal coverage, detection resolution, network design, iterative procedures, and configurable settings. The manuscript itself states that the breakpoint frequencies reflect the adopted parameter configurations. A greater number of candidate breakpoints does not necessarily mean greater sensitivity to genuine inhomogeneities. It could also reflect different significance thresholds; shorter minimum segments; iterative redetection; over-segmentation; natural climatic variability; false-positive detections; different annual or monthly treatment.
The Abstract should replace:
“BART … detects substantially more breakpoints, indicating greater sensitivity to potential inhomogeneities.”
with:
“Under the adopted configurations, BART identifies substantially more candidate breakpoints than CLIMATOL and PHA, although the absence of a known homogeneous benchmark prevents separation of greater detection sensitivity from a potentially higher false-detection rate.” For every method, report: software version; temporal resolution; default and modified settings; significance threshold; minimum segment length; neighbour-selection rule; iterative stopping rule; treatment of closely spaced breaks; treatment of edge breakpoints; adjustment method. The analysis should also report breakpoint rates for the common station-period subset.
- The manuscript should not infer that BART’s additional breakpoints are genuine without direct evidence. The Results appropriately describe BART as producing higher breakpoint-detection rates. However, the Discussion interprets this as greater sensitivity to potential inhomogeneities and an increased likelihood of identifying subtle artificial shifts. The manuscript subsequently acknowledges that it may also increase false positives and over-adjustment. The balanced interpretation belongs in the Results and Abstract, not only in the Discussion.
Use neutral terminology such as: higher candidate breakpoint frequency; more active detection behaviour; greater adjustment activity under the adopted settings. Reserve “greater sensitivity” for a controlled benchmark where true breakpoints are known.
- Long-gap reconstruction must be evaluated separately from homogenisation. The paper presents long missing periods at Bourges and Hanoi that are reconstructed by the three methods. Reconstructing these gaps increases temporal completeness, but completeness does not establish accuracy. There is also a conceptual issue. Homogenisation and gap reconstruction are related but distinct tasks:
- homogenisation detects and adjusts non-climatic discontinuities;
- reconstruction estimates values where observations are absent.
A method may perform well at breakpoint correction but poorly at reconstructing a 40- or 50-year gap, or vice versa.
The manuscript should distinguish homogenisation from gap filling throughout; report the proportion of observed and reconstructed months; provide gap-length distributions by method and region; visually distinguish observations from reconstructed values; conduct withheld-observation tests; quantify reconstruction error by gap length; assign quality flags to reconstructed periods; avoid describing reconstructed values as equivalent to direct observations. Long reconstructed periods should be accompanied by uncertainty estimates or, at minimum, quality categories.
- The ANOVA and Tukey HSD comparison requires fuller methodological description and likely redesign. The authors compare the original and homogenised annual mean temperatures for 1895–2015 using one-way ANOVA and Tukey HSD. However, the manuscript does not state clearly what constitutes an individual observation in the ANOVA.
Potential observational units include: annual regional means; station-year values; station-level temporal means; pooled annual station values.
This distinction is essential because if annual values from the same stations and years are treated as independent groups, the analysis does not account for pairing, repeated measures, station effects, or temporal autocorrelation. If regional annual means are used, temporal autocorrelation still requires consideration.
The authors should specify exact observational unit; sample size in each group; treatment of missing values; whether identical stations and years are paired; whether annual values are regional averages; whether temporal autocorrelation was assessed; whether ANOVA residual assumptions were verified.
A more appropriate approach may be paired differences; repeated-measures analysis; mixed-effects modelling with station and year effects; block-bootstrap confidence intervals; generalized least squares with autocorrelation structure. The manuscript should report effect sizes and uncertainty, not only -values.
- The reported correlations with HCLIM are approximately 0.98–0.99, with small centered RMSD in the selected networks. These results demonstrate that the homogenised products remain similar to the original series in temporal variability and distribution. However, high similarity is expected because the products are derived from HCLIM. It does not establish whether: a true breakpoint was corrected; a false breakpoint was introduced; an adjustment had the correct magnitude; a reconstructed long gap is accurate; a long-term trend was preserved correctly.
The authors should describe the Taylor diagrams as measures of fidelity to the original series, rather than measures of homogenisation accuracy. The statement that these diagrams demonstrate preservation of the “principal climatic signal” is defensible if carefully qualified. The statement that they demonstrate correct homogenisation is not.
- RMSD from HCLIM quantifies adjustment magnitude, not method error
The manuscript correctly states in the Methods that RMSD measures how strongly each method modifies HCLIM. However, parts of the Discussion use lower RMSD in performance-oriented language, including statements that BART generally produces lower RMSD values and that PHA “performed best” at selected stations.
A low RMSD may indicate: preservation of the original record; fewer adjustments; smaller adjustments; failure to correct an existing inhomogeneity.
A high RMSD may indicate: aggressive or erroneous modification; necessary correction; extensive reconstruction; differences in temporal coverage.
Avoid ranking methods using RMSD from HCLIM unless an independent target is available.
Use wording such as: PHA produced the smallest modification relative to HCLIM at Calcutta rather than: PHA performed best at Calcutta.
Similarly, “BART generally produced the lowest RMSD” should not be presented as evidence that BART is more accurate.
- The manuscript identifies 20CRv3 as an external reference and acknowledges: its early-period observational sparsity; its 1° × 1° resolution; potential disagreement at high-elevation and coastal stations; member ensemble.
However, only the ensemble mean is used, and the nearest grid cell is matched to each station. The claim that elevation or coastal differences remain relatively constant through time requires evidence. Temporal variations in lapse rates, land–sea contrast, circulation, snow cover, urbanisation, or local exposure could alter station-grid differences.
Use “consistency assessment” rather than “validation against 20CRv3.”
The authors should consider: incorporating ensemble spread; calculating correlations and errors for each ensemble member; reporting uncertainty intervals; applying elevation correction where appropriate; testing bilinear interpolation or multiple nearby cells; stratifying by elevation and coastal proximity; separating early and later reanalysis periods; assessing temperature anomalies rather than absolute temperatures where appropriate.
The full ensemble is particularly important because uncertainty changes strongly over time and region.
- The manuscript computes correlations between STL-derived trend and deseasonalised components. Smoothed trend series are strongly autocorrelated, so apparent sample size can be much larger than the effective number of independent observations. The manuscript does not provide the STL settings or uncertainty intervals for the correlations.
Report: STL implementation and package version; seasonal window; trend window; robustness setting; handling of missing values; common comparison periods; effective sample size; autocorrelation-adjusted confidence intervals.
The evaluation should also include: anomaly RMSE; mean bias; trend difference; variance ratio; first-difference correlation; observed-period and reconstructed-period results separately. The statement that deseasonalisation “improves agreement” should be changed to “produces higher correlation” unless improvement is defined independently.
- Network System A is defined using station distribution, geographic proximity, and political boundaries. Network System B uses latitude, longitude, station-identifier naming conventions, and some Köppen-based criteria. These operational rules may be reasonable, but political boundaries and naming conventions do not necessarily represent climatological coherence. Conversely, useful reference stations may lie across national boundaries.
The authors should explain why each network is climatically coherent and provide diagnostics such as: interstation correlation; distance distributions; elevation difference; coastal exposure; network size; effective neighbour count; climate-zone composition.
A sensitivity analysis comparing the administrative network design with correlation-based or climate-zone-based networks would improve confidence.
- The manuscript reports overall retention of approximately: 96% for CLIMATOL; 99% for PHA; 80% for BART. These global percentages are informative but may conceal substantial selection bias. For example, losses may be concentrated among: short early records; tropical stations; high-latitude stations; isolated islands; high-elevation sites; data-sparse countries; records with limited metadata.
Global temperature distributions could remain similar even if particular regions or historical periods are disproportionately removed.
Report retention by network; continent; country; climate zone; century; record length; elevation; station density; coastal versus inland location. The manuscript should map excluded records and identify geographical gaps in the final product.
- Table 3 compares the highest and lowest mean, minimum, and maximum values across HCLIM and the preprocessed datasets. This demonstrates that several global record-holder stations remain included. It does not establish whether the broader distributions of extremes are preserved. The definitions of “mean,” “minimum,” and “maximum” also need clarification because the study is described primarily as an analysis of monthly mean temperature. The authors should: define each temperature variable; clarify whether minima and maxima are separate monthly variables; evaluate percentile distributions; compare the 1st, 5th, 95th, and 99th percentiles; assess station-level variance changes; assess regional tail distributions; distinguish preprocessing effects from homogenisation effects. The current analysis should be described as preservation of selected global extreme values, not comprehensive preservation of climate extremes.
- HOM-HCLIM is intended for climate reconstruction and reanalysis applications, but the manuscript does not describe value-level uncertainty. The reliability of each homogenised value likely depends on: number and quality of neighbouring stations; network density; length of temporal overlap; correlation with reference stations; breakpoint agreement; method disagreement; length of reconstructed gaps; metadata availability; early-period observational density.
Provide at least a quality-flagging framework. Suggested fields include directly observed versus reconstructed; homogenisation method; number of methods providing a value; intermethod spread; number of reference stations; mean reference correlation; gap length; breakpoint agreement; adjustment magnitude; confidence category.
If a single recommended estimate is supplied, structural uncertainty among methods should not be discarded.
- The Practical Implications section states that “cross-validation across methods reduces methodological uncertainty.” Comparing outputs from CLIMATOL, PHA, and BART is not statistical cross-validation because observations are not divided into training and independent validation sets. Moreover, agreement among methods does not necessarily reduce uncertainty. It may provide evidence of robustness, but the methods share underlying observations and may share reference-network limitations.
Replace “cross-validation across methods” with cross-method comparison; multimethod intercomparison; ensemble consistency assessment.
The authors could write: Agreement among independently implemented methods can increase confidence that major features are not specific to a single algorithm, although it does not by itself establish accuracy or eliminate structural uncertainty.
- Figure 10 sample size
The Results state that the correlation analysis covers 24 stations, whereas the Figure 10 caption says 20 stations. The correct sample size must be reported consistently.
De Bilt and Delft_Rijnsburg are confused
Table 4 reports:
- BART’s earliest breakpoint at Delft_Rijnsburg in 1727;
- CLIMATOL’s earliest breakpoint at De Bilt in 1760;
- PHA’s earliest annual breakpoint at De Bilt in 1865.
The Discussion instead states that De Bilt demonstrates convergence on a breakpoint “as early as 1727.” That is inconsistent with Table 4.
This sentence should be corrected or supported with separate metadata evidence.
Duplicated sentence
The France ANOVA paragraph states twice that no statistically significant differences were detected among the three homogenised datasets. One sentence should be removed.
Table cross-reference inconsistency
The text states that Table 2 summarizes exclusions and Table 3 provides examples of excluded stations. In the displayed manuscript, examples appear in Table 1, and exclusion counts appear in Table 2. All cross-references should be checked.
“Best homogenised dataset” is undefined
The data-availability statement refers to “the best homogenised dataset,” but the manuscript repeatedly concludes that no method is universally superior and does not define how “best” was established.
The authors should use the exact product name and explain the selection or combination procedure.
Original versus automated BART implementation
The Methods call all three methods fully automated, while the Introduction describes BART as combining automatic processing with interactive, expert-informed adjustments. The Results clarify that a fully automated implementation was used in this study and that this differs from the original semi-manual design. This distinction should be stated consistently from the beginning.
- The reference list requires careful checking. I recommend that the authors verify rather than merely reformat the following:
- the Venema et al. DOI appears malformed;
- the Valler et al. DOI contains an implausible numeric sequence;
- the Fleming DOI contains a duplicated DOI prefix;
- several entries lack complete article details;
- some in-text references use plural authors where the bibliography lists only a single author;
- Guijarro citations are inconsistent;
- WMO document numbers should be verified;
- Wikipedia and blog references should be removed unless scientifically indispensable;
- access dates should be standardized;
- typographical forms such as “Novembre” and “Octobre” should be corrected;
- “20crv” in the author-contribution statement should be “20CRv3”;
- author contributions should preferably follow CRediT roles.
Minor comments
- Standardize “Southeast Asia” and “South-East Asia.”
- Define “record,” “series,” “station,” and “observation.”
- Replace “M.a.s.l.” with “m above sea level.”
- Replace “responsive homogenisation method” with “relevant homogenisation method.”
- Correct encoding errors in station names.
- Define centered RMSD separately from ordinary RMSD.
- Include °C units for all RMSD values.
- Clarify the treatment of 2020 and 2021.
- State whether adjustment is applied to earlier or later segments.
- Distinguish observed, adjusted, and reconstructed values in all figures.
- Provide exact station counts for every regional analysis.
- Explain the choice of the eight RMSD stations.
- State whether these stations were selected before examining the outputs.
- Replace “Napoleon bump” with a formally defined and referenced term or a descriptive phrase.
- Use accessible colour palettes.
- Provide machine-readable station-level diagnostics.
- Add a consolidated limitations subsection.
- Check every table and figure citation.
- Remove duplicated wording and unnecessary repetition.
- Rewrite the Code and Data Availability section as complete sentences.
Citation: https://doi.org/10.5194/egusphere-2026-3936-RC2 -
CC1: 'Reply on RC2', Elin Lundstad, 19 Oct 2026
We sincerely thank the reviewer for the thorough, constructive, and detailed evaluation of our manuscript. We appreciate the reviewer’s recognition of the scientific importance of the HCLIM archive, the value of comparing three automated homogenisation approaches, and the potential contribution of HOM-HCLIM to the study of historical instrumental temperatures.
We acknowledge the reviewer’s central concern that the current analyses demonstrate intermethod agreement, data retention, modification magnitude, and consistency with 20CRv3 more directly than they establish the absolute accuracy of homogenisation. We agree that this distinction is important, particularly for early instrumental records, for which complete and independent reference observations are generally unavailable.
In response, we will revise the manuscript to distinguish more explicitly between intercomparison, consistency assessment, and validation of homogenisation accuracy. We will moderate claims concerning breakpoint detection, reconstruction quality, and method performance wherever these claims are not supported by an independent benchmark. We will also clarify the construction of HOM-HCLIM, improve the description of methodological choices and statistical analyses, and expand the discussion of uncertainty and limitations.
We will address the identified inconsistencies in the text, tables, figures, and references, and improve the documentation of the released dataset to make its contents and processing procedures more transparent and reproducible.
Where additional analyses are feasible, we will include them to strengthen the evaluation. Where a proposed analysis cannot be conducted reliably with the available data, we will explain the limitation explicitly and adjust the corresponding claims rather than imply that the issue has been resolved.
We are grateful for the reviewer’s careful assessment and believe that these comments will help us improve the scientific rigour, transparency, and practical value of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2026-3936-CC1 -
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
Dear Jose Guijarro,
Thank you very much for your careful and constructive comments. It is especially nice to hear from you, and we greatly appreciate that you took the time to read and comment on our manuscript. Your detailed comments on CLIMATOL are particularly valuable and will help us improve the methodological accuracy and clarity of the paper. We will carefully address all the points raised in the revision.
Best wishes,
Elin, on behalf of the authorsCitation: https://doi.org/10.5194/egusphere-2026-3936-AC1
-
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
-
AC1: 'Comment on egusphere-2026-3936', Elin Lundstad, 28 Aug 2026
Dear Jose Guijarro,
Thank you very much for your careful and constructive comments. It is especially nice to hear from you, and we greatly appreciate that you took the time to read and comment on our manuscript. Your detailed comments on CLIMATOL are particularly valuable and will help us improve the methodological accuracy and clarity of the paper. We will carefully address all the points raised in the revision.
Best wishes,
Elin, on behalf of the authorsCitation: https://doi.org/10.5194/egusphere-2026-3936-AC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 236 | 99 | 28 | 363 | 94 | 26 | 28 |
- HTML: 236
- PDF: 99
- XML: 28
- Total: 363
- Supplement: 94
- BibTeX: 26
- EndNote: 28
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments:
The authors have applied three homogenization methods to improve the quality of the HCLIM historical climate observations dataset. They apply several statistics to compare the results and conclude that all of them seem to successfully remove the non climatic biases in the series. Differences between the methods can be ascribed to the density of observatories in time and space, and to the default parametrization of the methods. Their conclusions reassures the validity of homogenization for a better assessment of climate variability and trend studies based on observational series.
Apart from this article, the climatological community will greatly benefit if the authors undertake the intended future work including daily observations, additional reanalyses and novel approaches for breakpoint detection and adjustment.
Specific comments, preceded by the section or line number:
73-74: “SNHT identifies potential inhomogeneities through comparison of accumulated anomalies.” This is more a description of the Craddock method. SNHT works by standardizing the data and calculating a test statistic that identifies the peak point in the series where the before and after means differ the most.
2.2.1 Exclusions of records: Having read other papers comparing the performance of several homogenization methods, some reasons for the rejection of records seem quite inconsistent to me. This is the case, for example, with the exclusion of records from nHom_CLIMATOL because of outliers of temperature, since CLIMATOL automatically deletes outliers greater than a prescribed threshold and goes on with the homogenization process, so there is no reason to exclude series because this software detects some outliers in them. (But if by “records” you mean individual data items rather than series, then please disregard this comment.) Also strange is the exclusion because of low correlations (Table 2), since correlation has no role in the selection of neighbors in CLIMATOL, or “Too long distance (defined by the program)” and “Big gaps”, when no limits are set by the program to these parameters. It is clear that too long distance to neighbors may compromise the reliability of the homogenization, but this will be true also for BART, PHA and any other relative homogenization methods. As to BART and PHA, readers would probably like to know which are those “Other reasons” in Table 2.
205-207: CLIMATOL provides a function for the automatic conversion of SEF files into its own input formats. Apart from that, the sentence “CLIMATOL requires CSV input files, with all series combined into a single spreadsheet and accompanied by an .esd metadata file.” is wrong, since both the data and metadata files are plain text.
224: “frequency-based Standard Normal Homogeneity Test”. Frequency-based? Please correct or explain what you mean with this sentence.
428-429: “No statistically significant differences were detected among hom_CLIMATOL, hom_BART, and hom_PHA.” This sentence is repeated in line 427.
Fig. 11, caption: “While the hom_BART dataset generally has smaller RMSD values compared”. Should not “smaller” be changed by “higher”?
698: Following CRAN, the canonical form to cite the CLIMATOL package is https://CRAN.R-project.org/package=climatol
787: “The Twentieth Century625”. “625” should be deleted.
817: “StateM eteorologicalAgency(AEM ET )” should be “State Meteorological Agency (AEMET)”.
Table S1. Corrections to CLIMATOL column:
Metadata usage: Optional.
Station distance constraints: None in the code. (But the user should avoid using reference stations so far away as to compromise a fair correlation with the tested station.)
Internal option for reference stations: Geographical.
Minimum number of series: 3 (2 if one is assumed to be homogeneous, such as from reanalysis).
Step 4: Missing data estimation.
Correlation threshold (Step 1, …): None (in any step).
(Correlation is only calculated for the initial cluster analysis, which is offered to the user to allow judgment on whether the area can be considered free of clear climatic boundaries. No further use of the clustering is applied to the selection of reference stations.)
Correction method: Missing data estimation.