the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Jul 2026
CAMELS-PE: Hydrometeorological time series and catchment attributes for 136 catchments in Peru
Abstract. Large-sample hydrological datasets are essential for advancing hydrological understanding and modelling across diverse environments, yet they remain scarce in South America, particularly in tropical Andean regions with strong climatic and physiographic gradients. Here, we present CAMELS-PE v1.0.1, a large-sample hydrological dataset for Peru that provides daily hydrometeorological time series and catchment attributes for 136 catchments. The dataset includes observed and simulated streamflow, meteorological forcing variables, geospatial layers, and attributes describing topography, climate, hydrological behaviour, land cover, geology, soils, and human intervention. All variables were generated under a consistent workflow involving temporal harmonisation, catchment-scale aggregation, and standardised formatting, with dedicated screening applied to observed streamflow records. The resulting dataset was evaluated through consistency checks across metadata and catchment attributes, together with plausibility analyses of regional hydroclimatic patterns. By capturing Peru’s pronounced environmental contrasts, CAMELS-PE expands the representation of tropical Andean and Amazonian headwater catchments within the CAMELS framework and provides an open benchmark dataset for hydrological modelling, regionalisation, climate–streamflow analysis, prediction in ungauged basins, and machine-learning applications. CAMELS-PE is publicly available through Zenodo at https://doi.org/10.5281/zenodo.21195425 (Llauca et al., 2026) and is supported by the RCamelsPE R package.
- Preprint
(4050 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (extended)
- RC1: 'Comment on essd-2026-386', Iiro Seppä, 14 Jul 2026 reply
-
RC2: 'Comment on essd-2026-386', Keirnan Fowler, 21 Sep 2026
reply
CAMELS-Peru is a welcome addition to the CAMELS family of datasets, providing valuable coverage of under-represented (particularly near-equatorial) environs. The authors have done an admirable job in gathering a wide range of relevant data and providing descriptions in this manuscript, which is supplemented by a data description in the repository. They have also emphasised the limitations of the dataset quite well, which is important as the available records are sometimes short and/or gappy, and the authors have turned to modelled to supplement information.
I suggest that the authors consider the following suggestions prior to the manuscript being considered further:
1. Units of streamflow. I suggest to provide the streamflow in volumetric units (e.g. m³/sec or ML/day) in addition to mm/d. Volumetric units correspond more closely to the original measurement, and are becoming standard in datasets globally, for the following reason. We are reaching the point now where a given catchment might be in multiple datasets (globally this could be, for example, in Robin or GRDC in addition to a local dataset; also we have datasets covering the same geography like Hysets vs CAMELS-US) and depending on the context, these other datasets may have their own methods for defining catchment area. This means we end up with different streamflow values (expressed in mm/d), but this is ok so long as we have the original streamflow in a volumetric unit, which is free of assumptions about catchment area (and presumably is consistent across all datasets that apply to a given catchment). In the case of the Peru dataset, which may be the source of information incorporated into other datasets, it's important that it provides this information.
2. PET values - contextualise and double check. The PET values seem not to have enough variation. The highest value in the dataset for mean annual PET is ~1600 mm/yr and the 75%ile for Pacific catchments is ~1400 mm/yr according to Figure 2b. I would have expected much higher values for arid catchments near the equator. For example, comparing with the Australia dataset, max and 75%ile values are much higher at 2100 and 1850 mm/yr, respectively. Is it possible to check these numbers? Related issues include Fig 2d, where the results seem quite unusual:
- there is only an exceedingly weak relationship between aridity and runoff ratio; and
- the most extreme catchment (top left of 2d) has a P/PET of less than 5% yet three quarters of the rainfall becomes streamflow, this might be worth discussing, is this streamflow more due to deep seepage from distant, wetter regions?3. Day definition. Good to see the information (line 164) about day definition in meteorological data. Please include the day definition for streamflow in the preceding section, or create a special section to cover this especially. If there are any offsets in day definition between P and Q, it is important that this be known to the modelling community (acknowledging that this may be outside of the control of the authors).
4. Hydrological signatures simulated, not observed. It is rather difficult for readers to assess their usefulness, given they are calculated from a model and the values arising are never compared with reality (although model KGEs are given in an appendix). Would it be possible to calculate the same signatures on observed flow from selected stations (those with long, complete records) and report on the accuracy of the modelled signatures? If the accuracy is not good and alternatives are being sought, one possibility would be to calculate signatures one year at a time and report results as some summary statistics of the distribution (e.g. the mean and standard deviation); thus the shorter records should have a greater spread due to the smaller sample size.
5. Versioning. Line 9: calling it v1.0.1 is a bit confusing, and I don't think it's explained anywhere in the manuscript. I think it would be better to call it v1 and then later explain that there exists different sub versions and explain the difference between 1.0.0 and 1.0.1. In practice, there won't be any issue as users just default to the latest version on Zenodo (provided you use the Zenodo link that points by default to the latest version and not to a specific version--please confirm).
6. Suggested additions:
6a: Budyko plot or similar. Related to the point about PET above, would it be possible to produce a Budyko-style plot, partly as a way of checking the PET? A motivation for the previous comment is that if the PET is unreasonably low then some catchments may lie off the Budyko-allowed space because the PET is not sufficient to close the gap between P and Q. If so, RR models will naturally struggle and it might be worthwhile considering to calculate another PET formulation and include it (perhaps one that can be calculated using the data already in the dataset?).
6b. Full sized map with catchment boundaries. Would it be possible for the authors to provide a full-sized map of Peru, shaded with some aspect of climate (mean annual precipitation or aridity) and with the catchment boundaries plotted along with the gauges as dots? This helps readers who are unfamiliar with Peru to orient themselves, and also to get a sense of how big the largest catchments are (noting that the smallest polygons will be barely visible). For an example, see Fig. 1 of https://doi.org/10.5194/essd-13-3847-2021
Minor points:
Line 89: The statement "timeseries and attributes were calculated over the full upstream drainage area, using national products where available andglobal products for attributes such as topography, geology, soils, and land cover"--this statement reads as if it is applicable only to transboundary systems. Is this the case? If so, please distinguish approaches between transboundary and non-transboundary rivers here (I realise you will clarify this later, but it is better if this statement can stand alone).
Line 99 / Figure 1c. This is great. To ease adoption, users will be interested in whether this missing data periods are haphazard or follow clear patterns (for example, the period of highest flow is commonly missing). Can you comment on this? If it is discussed later, can you provide a cross reference to that section at line 102?
Line 191: "All series were formatted on a common 1981–2025 daily calendar, with missing values retained outside the native temporal coverage of each product." I don't understand, please clarify, is this statement about start/end date or about gaps?
Table 2. Re: the final column, it is not possible to see how many rows each entry applies to. When this manuscript is being typeset, I suggest the authors prompt the typesetters to ensure this is visible. Same for other similar tables.
Table 2 / Line 255. Please provide more detail about how slope was calculated. Usually this would be from a gridded dataset of slope which is then averaged across space. However, given the low values here (20 m per km is really not high!), I suspect it is calculated by taking the maximum and minimum elevations in a catchment and dividing it by a measure of catchment length, is this correct? Would the authors consider a gridded dataset instead?
Line 236. Did catchment outlets need to be moved to be on a flow line in the DEM-based stream network? Please provide a description of how this was done (usually quite subjective!) and any uncertainties that arise. What was the maximum distance moved?
Table 8. It is difficult for readers/users to understand what proportion of water resources is used by humans. Would it be possible for the authors to present the volumes as a % of the mean annual streamflow in the catchment? I understand that the authors are keen to avoid the impression that these are "direct measurements of abstraction, consumption, or hydrological alteration" (L549) but it still needs some sort of standardisation by the size/typical flow in a catchment.
Citation: https://doi.org/10.5194/essd-2026-386-RC2
Data sets
CAMELS-PE: Catchment Attributes and Meteorology for Large-sample Studies in Peru H. Llauca et al. https://doi.org/10.5281/zenodo.21195425
Model code and software
RCamelsPE R package H. Llauca https://github.com/hllauca/RCamelsPE
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 520 | 222 | 99 | 841 | 121 | 90 |
- HTML: 520
- PDF: 222
- XML: 99
- Total: 841
- BibTeX: 121
- EndNote: 90
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Dear authors of CAMELS-PE
Thank you for creating a new CAMELS dataset! It takes an enormous amount of work and expertise, and that is greatly appreciated. CAMELS datasets have proven very useful for hydrology, and increasing the diversity of catchments in them is always welcome, in my opinion.
The article is well written, was easy to read and the figures were high quality and visually clear. The dataset is well organized and documented, and seemed to be approximately on par with the general level of quality of many CAMELS datasets.
I hope that my questions, comments and suggestions will help further enhance the already good quality article and the dataset. Please find them in the attached pdf.
Best Regards,
Iiro Seppä