Thank you for your comments!

According to your advice, we will redraw some of the figures by increasing the font size and possibly change the colour map.

Zhen Hao

Thanks!

We hope that this work will promote hydrological research in China, especially the code we are about to open source for generating the data for any watershed.

Specific response to the review:

Source code: the source code is now under development in this GitHub repository: 

https://github.com/haozhen315/catchment-attributes-and-meteorology-for-large-sample-study-in-contiguous-china. 

Part of the codes has been made public. We are currently sorting out others. The existing codes may also be updated. The goal is to make the proposed code easy to use and accurate.

Runoff data: The advice to calculate the length of the complete (continuous) streamflow records is great. We will definitely make up for this, which will be added to the JSON file of the Yellow River basins as a new attribute. Regarding the normalization of runoff data, this is mainly for confidentiality reasons. It isn't easy to release the original runoff data in China due to management reasons. The proposed work is the best effort we can make regarding the Yellow River Basin, which can support the development of machine learning models and hopefully much other research. As for the inability to calculate the annual runoff, we may consider adding annual runoff as new attributes to the JSON file.

Specific Comments: Thanks. We will rewrite those sentences, correct the reference style, and redraw some of the figures. The purpose of correlation analysis is to provide an overall understanding of the proposed dataset. Assumptions behind using Pearson's correlation coefficient may not apply to all static attributes. We may change the method of analysis or add an explanation to the current result.

New comments will be submitted once there is further progress.

To make the response more structured, we have organized the response.

Point-by-point reply to Referee 1

Date: 2021-08-16

Manuscript Number: ESSD-2021-71

Title of Article: Catchment attributes and meteorology for large sample study in contiguous China

Name of the Author: Zhen Hao

Email Address of the Author: zhen.hao18@alumni.imperial.ac.uk

RC: However, at this time, the source code is not yet available. If possible, however, the source code should already be published along with the data, which would add significant value to the paper.

AC: The code is now available at: https://github.com/haozhen315/CCAM-China-Catchment-Attributes-and-Meteorology-dataset

To further improve the usability and influence of the code, we have reformulated the code such that the user can generate a basin's characteristics with just a "one-click" when the required source data are prepared, which will significantly improve the accessibility of the catchment attributes.

In the past few days, we have made extensive efforts to reorganize our code. Combined with the data set that has been released, we are aiming to achieve two goals: 

(1) Researchers can quickly obtain catchment attributes and meteorological time series of the local catchment from our data set. 

(2) If the local catchment has a custom boundary, using our code can calculate the catchment attributes and meteorological time series quickly based on the given boundary. Our code currently supports one-click generation of all static attributes, as long as the required source data has been prepared according to the instruction, and the generation of the meteorological time series can also be quite effortless.

RC: Many of the normalized discharge time series seem to have gaps, which reduces their usability for further analysis. It would be useful for the reader if the mean length of complete discharge time series could be quantified, or the gaps could be addressed in general

AC: The original streamflow observations are not continuous. The average record length is 11.3 years. In the revised version of the dataset, we separately provide continuous streamflow observations with an average record length of 8.3 years.

RC: Also, the reason for the normalization should be addressed; as from the normalized time series, e.g. no mean annual discharge can be derived; and this parameter is also not given in the metadata of the catchments.

AC: Due to the strict redistribution policy of streamflow data, we are afraid not to be able to release the original streamflow data and the mean discharge. We must ensure that the source data is not released, but we want the released data to be useful, so we present the current solution. The current data can be used in such a situation: when it is desirable to verify the generalization ability of a machine learning model on a global scale, HydroMLYR (new name) can support the verification of the performance in the Yellow River Basin.

RC: line 28: Either name the main processes of the hydrological cycle, or focus on specific processes in terrestrial catchments. Careful with the right terms: rainfall instead of raindrop

AC: This sentence has been reformulated as “Rainfall, interception, evaporation and evapotranspiration, groundwater flow, subsurface flow and surface runoff are the main components of the terrestrial hydrological cycle.”

RC: line 33: "[..] it is possible for the hydrological model to learn [..]": this is only true for a spacial type of models (machine-learning models)

AC: This sentence has been reformulated as “However, by examining a large sample of catchments, it is possible for a data-driven model to learn the similarities and differences of hydrological behaviours across catchments (Kratzert, Klotz et al. 2019).”

RC: line 37: citation style: instead of "(Kratzert, Klotz et al. 2019) shows" it should be "Kratzert, Klotz et al. (2019) shows". also applies for similar citations later in the text

AC: "(Kratzert, Klotz et al. 2019) shows" should be "Kratzert, Klotz et al. (2019) shows". Other similar issues have also been fixed.

RC: Figures 3,4,5: as it is already written in a community comment, the text in the figures is hardly readable.

AC: We have fixed it by redrawing these figures.

RC: Appendix B: Assumption for the use of Pearsons Correlation Coefficient is normal distribution; and linear relationship is assumed. is this really always the case here?

AC: Although the Pearson's correlation can only provide a complete description of the association when both the two variables are standard, we think the most doubtful part of using it is that it assumes a linear relationship (a change in one variable will cause a proportional change to the other). Because there are too many variables, we can't plot the scatterplot one by one to check whether the relationships are linear. We think Spearman or Kendall's Tau may be more suitable in this case due to its wider application range. Even if the relationship between the two variables is linear, Spearman or Kendall can also return a very close result to Pearson. However, if the relationship between two variables is only monotonic, Pearson will have information loss.

Best,

Zhen Hao

1. Zenodo data has been updated;

2. As suggested by RC1, we have added the length of continuous (>95% complete) streamflow record (year) to the "attibutes.json" for each YR basin in the dataset; ~40 basins are having over ten years record, and the mean length of the continuous record is ~7 years.

To make the response more structured, we have organized the response.

Point-by-point reply to Referee 2

Date: 2021-08-16

Manuscript Number: ESSD-2021-71

Title of Article: CCAM: China Catchment Attributes and Meteorology dataset

Name of the Author: Zhen Hao

Email Address of the Author: zhen.hao18@alumni.imperial.ac.uk

RC: In the abstract, the authors state their “dataset provides numerous opportunities for comparative hydrological research, such as examining the difference in hydrological behaviours across different catchments and building general rainfall-runoff modelling frameworks for many catchments instead of limited to a few”. My concern is that the scope of this dataset (in its current form) might be more limited, because of the following restrictions imposed to the streamflow data: […]

AC: Due to the strict redistribution policy of streamflow data, we are afraid not to be able to release the original streamflow data and the mean streamflow. We must ensure that the source data is not released, but we want the released data to be useful, so we present the current solution. The current data can be used in such a situation: when it is desirable to verify the generalization ability of a machine learning model on a global scale, HydroMLYR (new name) can support the verification of the performance in the Yellow River Basin. The abstract has also been rewritten to avoid those claims.

RC: I understand that releasing the true streamflow time series is challenging, but some decisions made by the authors are puzzling. For instance, “for confidentiality, the names of these basins have not been announced”, but shapefiles are provided and give (presumably) the exact location of the catchments. Likewise, the mean streamflow can be inferred quite readily from catchment descriptors, which makes me feel that the normalisation of the timeseries is unnecessary.

AC: In fact, the location of hydrological observation stations can be observed through remote sensing satellite images and then combined with HydroSHEDS's River network to determine their location. Then the boundaries of the basin can be determined based on the publicly available DEM, but we cannot release the names of these hydrological observation stations; these are sensitive information. In addition, indicators that can be used to calculate the mean discharge have not been released.

RC: Hence, I recommend that the authors do not use the name CAMELS, as all the CAMELS datasets provide daily streamflow timeseries for their hundreds of catchments, which many in the community see as their most important characteristic. There are many alternative naming options, including exotic animal names (as illustrated by the recent LamaH dataset - https://essd.copernicus.org/preprints/essd-2021-72/).

AC: The paper title sounds like the title of a CAMELS dataset, and the use of "Normal-Camels-YR" might be misleading. We suggest using CCAM to stand for "China Catchment Attributes and Meteorology dataset" and HydroMLYR to stand for "Hydrology dataset for Machine Learning of the Yellow River Basin." The new names may avoid readers' wrong expectation of the data set and more clearly indicates the purpose.

RC: Data availability/reproducibility: The abstract mentions that “complement code for generating the dataset will be open-sourced such that the user can generate meteorological series and catchment attributes for any watershed within contiguous China”, yet the conclusion makes it clear that the forcing dataset SURF_CLI_CHN_MUL_DAY is only freely available for Chinese researchers (L444). This is a non-negligible constraint. Furthermore, I don’t see a paper documenting the SURF_CLI_CHN_MUL_DAY dataset, and the link provided (http://data.cma.cn/data/cdcdetail/dataCode/SURF_CLI_CHN_MUL_DAY.html) leads to page in Chinese.

AC: We are sorry that we made a false statement about the facts at the beginning. We found on the registration page that foreign researchers can also register, but the interface is still in Chinese, which is out of our control. The SURF_CLI_CHN_MUL_DAY data was issued by the National Meteorological Information Center of the China Meteorological Administration (NMIC/CMA). The data is quality controlled, and it is widely used in research in China. However, it does not have a related paper.

Best,

Zhen Hao