Caravan-CMIP6: Bias-corrected climate model projections for ten large-sample hydrometeorological datasets and over 23,000 global catchments

Ruzzante, Sacha Walde

doi:10.5194/essd-2026-350

Preprints

https://doi.org/10.5194/essd-2026-350

Preprints

11 Jun 2026

| 11 Jun 2026

Status: this preprint is currently under review for the journal ESSD.

Caravan-CMIP6: Bias-corrected climate model projections for ten large-sample hydrometeorological datasets and over 23,000 global catchments

Sacha Walde Ruzzante

Abstract. This data paper introduces Caravan-CMIP6, a dataset of climate change projections for large-sample hydrologic studies. Caravan-CMIP6 includes projections from an ensemble of 12 climate models from the sixth Coupled Model Intercomparison Project (CMIP6), for the historical experiment (1850–2014) and three Shared Socio-economic Pathways (SSPs). The dataset includes projections for all catchments within ten large-sample hydrometeorological datasets: Caravan, CAMELS, CAMELS-AUS-v2, CAMELS-BR, CAMELS-CH CAMELS-CL, CAMELS-COL, CAMELS-DE, CAMELS-GB-v2, CAMELS-IND. For each large-sample hydrometeorological dataset, I provide CMIP6 projections for all meteorological variables that can be readily extracted from climate models, including precipitation, temperature, evapotranspiration, humidity, radiation, pressure, and wind speed. I bias-correct the climate model output to match the observed climatology within each dataset. This dataset can facilitate the use of large-sample hydrologic datasets and models for climate change projection. All raw and bias-corrected data are available at https://doi.org/10.20383/103.01644.

Received: 07 May 2026 – Discussion started: 11 Jun 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Sacha Walde Ruzzante

Status: open (until 20 Aug 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on essd-2026-350', Anonymous Referee #1, 24 Jul 2026 reply
The manuscript presents Caravan-CMIP6, a bias-corrected CMIP6 climate projection dataset prepared for ten large-sample hydrological (CAMELS-style) datasets. The dataset is expected to be valuable for the hydrological community, particularly for climate change impact studies requiring consistent meteorological forcing across multiple catchments. The dataset itself appears useful, and both the raw and bias-corrected products are made available. However, I found that many methodological choices are insufficiently explained, several implementation decisions lack scientific justification, and the manuscript relies too heavily on previous publications for essential methodological details. As an ESSD paper, the manuscript should serve as the primary documentation of the dataset and therefore should be more self-contained. I recommend major revision before the manuscript can be considered for publication.
General comments
While active voice and first-person perspective are acceptable for a single-author paper in ESSD, the current manuscript relies very heavily on personal narrative phrasing throughout (e.g., "I provide...", "I bias-correct...", "I produce...", "I required..."). At times, this shifts the focus away from the dataset and methodology and onto the author's workflow. I recommend varying the sentence structure by focusing more on the dataset, processing steps, and analytical framework to improve the scientific flow and readability.

Throughout the manuscript, many methodological descriptions are replaced by references to previous studies (e.g., model selection, weighting strategy, bias correction, PET calculation). While citing previous work is appropriate, the manuscript should provide concise but sufficient descriptions of the methodology so that readers can understand how Caravan-CMIP6 was constructed without repeatedly consulting multiple publications.

A summary table describing the final Caravan-CMIP6 dataset would greatly improve readability. Such a table could summarize the included CAMELS datasets, number of catchments, reference observational datasets used for bias correction, variables provided, derived variables, temporal resolution, temporal coverage, and spatial representation.

Introduction
The statement describing Caravan-CMIP6 as covering "CAMELS-style datasets" should be clarified. It is not clear whether "CAMELS-style" refers to the catchment concept, temporal resolution, spatial representation, variable availability, or another characteristic.

Methodology
Dataset description
Table S1 summarizes the original CAMELS datasets, whereas some variables are excluded in the present work. A summary of the datasets and variables actually included in Caravan-CMIP6 would improve clarity.

The exclusion of snow-water equivalent appears to be primarily motivated by implementation difficulties associated with the bias-correction algorithm. Additional scientific justification should be provided together with discussion of the implications for users working in snow-dominated catchments.

The manuscript states that spatial median, minimum, maximum, and standard deviation cannot be provided because of the coarse CMIP6 resolution. However, it is unclear what information is provided instead.

The sentence "I required each model and variant to report nine variables" should be clarified. It is not clear what is meant by "report" and how the model selection was performed.

The statement "replacing the central 10 years within each 30-year window" is difficult to follow and should be explained more clearly.

The temporal characteristics of the dataset should be introduced explicitly, including temporal resolution, temporal coverage, treatment of different CMIP6 calendars, and any harmonization procedures.

Model selection
The discussion of model selection lacks coherence. The manuscript first discusses ensemble selection, then model modifications, and then abruptly moves to the "hot model" problem and model weighting. Consider separating these topics into distinct subsections.

The discussion of ECS/TCR and the "hot model" issue appears disconnected from the construction of the dataset. It is not immediately clear whether this section describes methodological decisions used to build Caravan-CMIP6 or recommendations for future users.

The model weights provided in Table 1 require additional explanation. Their meaning, derivation, interpretation, and intended application should be described.

Data processing
Section 2.2 is entitled "Raw CMIP6 model output extraction", although the section mainly describes basin-scale spatial aggregation rather than raw data extraction. A more appropriate section title would improve clarity.

Basin averaging inevitably modifies the native spatial representation of the climate models, particularly for small catchments and heterogeneous terrain. This limitation should be acknowledged and discussed.

Throughout Section 2.3, implementation details are often introduced before explaining why they are required. For example, basin elevation is introduced before explaining that it is needed to estimate surface pressure. Similar issues occur throughout the methodology. Reorganizing the workflow by first explaining the motivation and then describing the implementation would substantially improve readability.

Bias correction
The rationale for selecting Quantile Delta Mapping (QDM) requires further discussion. QDM is a well-established bias-correction method but is not without limitations. In particular, it does not preserve multivariate dependence and, without multivariate extensions, does not preserve spatial and temporal dependence structures. The manuscript should explain why QDM was selected over other available approaches and discuss its limitations more explicitly.

The statement that variables "are not always provided in the same form" or "are not in the optimal form for bias correction" is unclear. Please clarify what is meant by "form" and whether these transformations are required because of differences between CMIP6 outputs and the reference datasets or because of assumptions of the selected bias-correction method.

The temperature correction workflow is difficult to follow because the notation for bias-corrected variables is introduced before the bias-correction procedure itself is described.

Earlier sections describe strict model selection criteria, whereas Section 2.3 repeatedly states that some variables are "not always provided". This creates unnecessary confusion. Please clarify that this refers to different representations of equivalent variables rather than missing variables.

Several derived variables (dew point temperature, surface pressure, PET, etc.) are introduced without first explaining why they are required or how they relate to the objectives of Caravan-CMIP6.

The treatment of wind deserves further discussion. The manuscript acknowledges the difficulty of bias correcting the wind vector components but ultimately bias corrects only wind speed. Please clarify whether the released dataset contains only wind speed or also the vector components, and discuss the implications of not preserving wind direction.

The derivation of PET products requires stronger scientific justification. Several methodological choices appear to be motivated primarily by implementation convenience (e.g., assumptions regarding Priestley-Taylor parameters or replacing cloud-cover-based radiation calculations because cloud cover was not downloaded). These choices should be supported by scientific references or discussion demonstrating that the approximations are appropriate and their implications are acceptable.

The physical limits imposed after bias correction should be better justified. It would also be useful to report how frequently these limits are activated and whether they have any measurable impact on the resulting dataset.

The observational or reference datasets used for bias correction across the different CAMELS datasets should be summarized more clearly.

Results and Discussion
The discussion states that the bias correction is performed separately for each half-month, although this temporal framework has not been introduced previously. This creates confusion regarding the implementation of the bias-correction procedure.

The description of return levels is difficult to follow. The rationale for estimating return levels using annual block maxima and empirical ranking should be better justified rather than being presented mainly as a matter of simplicity.

The concepts of ratio and interval variables should be clearly defined before discussing the results.

Figures 2 and 3 evaluate only selected variables. Please explain why the remaining variables are not presented or provide justification for the selected subset.

Figures 4 and 5 present projections only for the 3°C warming level, although four warming levels are introduced earlier in the manuscript. Please explain why only one warming level is shown.

The current evaluation focuses primarily on whether QDM preserves the projected climate change signal by comparing the raw and bias-corrected CMIP6 outputs (Figures 2 and 3). While this is an important characteristic of QDM, it does not demonstrate how well the bias-corrected historical simulations reproduce the reference observations. Since Caravan-CMIP6 is intended to provide bias-corrected meteorological forcing, I recommend including a concise validation of the historical period against the observational datasets used for bias correction. Representative performance statistics (e.g., bias, RMSE, correlation, distributional agreement, or seasonal climatology) for selected variables and representative catchments or regions would provide users with greater confidence in the quality of the released dataset and complement the current evaluation.

Conclusions
The opening paragraph discusses projected streamflow changes, although streamflow itself has not been analysed in this study. The conclusion should better reflect the actual scope of the manuscript.

The justification for not adopting multivariate bias correction focuses primarily on computational cost. A stronger scientific justification should be provided, together with discussion of the methodological trade-offs.

The statement that Caravan-CMIP6 is biased toward Europe and North America has not been analysed or discussed elsewhere in the manuscript. This limitation should either be discussed earlier or omitted from the conclusion.

Overall, I believe the dataset has significant value for the hydrological community. However, the manuscript requires substantial revision to improve the clarity of the methodology, strengthen the scientific justification for several processing decisions, and provide a more complete evaluation of the released bias-corrected dataset. Addressing these issues would considerably improve the reproducibility and usability of Caravan-CMIP6 as a long-term community resource.

Reply
Citation: https://doi.org/10.5194/essd-2026-350-RC1

Sacha Walde Ruzzante

Data sets

Caravan-CMIP6: Bias-corrected climate model projections for ten large-sample hydrometeorological datasets and over 23,000 global catchments Sacha Ruzzante https://doi.org/10.20383/103.01644

Model code and software

Caravan CMIP6 GitHub Sacha Ruzzante https://github.com/sruzzante/Caravan-CMIP6

Interactive computing environment

Caravan CMIP6 Colab Example Sacha Ruzzante https://github.com/sruzzante/Caravan-CMIP6/blob/main/Caravan_CMIP6_Colab_example.ipynb

Sacha Walde Ruzzante

Viewed

Total article views: 251 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
185	57	9	251	9	9

HTML: 185
PDF: 57
XML: 9
Total: 251
BibTeX: 9
EndNote: 9

Views and downloads (calculated since 11 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	150	37	7	194
Jul 2026	35	20	2	57

Cumulative views and downloads (calculated since 11 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	150	37	7	194
Jul 2026	35	20	2	57

Viewed (geographical distribution)

Total article views: 237 (including HTML, PDF, and XML) Thereof 237 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 24 Jul 2026

Short summary

Caravan-CMIP6 provides climate change projections of weather for many rivers worldwide. Using Caravan-CMIP6, researchers will be able to use existing hydrologic models, which are calibrated on historical data, to predict future river flows under three climate change scenarios.


Total:	0
HTML:	0
PDF:	0
XML:	0