Long-Term Characteristics of Surface Soil Moisture over the Tibetan Plateau and Its Response to Climate Change
Round 1
Reviewer 1 Report
The manuscript focuses on the temporal and spatial changes of soil water on the Qinghai-Tibet Plateau and its driving factors, which are of great scientific value for global climate change, water cycle, and energy balance. The manuscript method is rigorous, the results are rigorous, and the discussion is clear, but there are deficiencies in explaining the source of the results. For example: How are the results in Figure 7 calculated? For the convenience of the reader to read and understand, a description is required. It is recommended that the manuscript be published after minor revisions..
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
Dear Authors,
The paper is devoted to analysis via inter-comparison of multi-source soil moisture products over the TP region with the two sets of in-situ observations; to study the longterm characteristic of soil moisture; and the responses of soil moisture over the TP region to climatic variables are estimated with a machine learning method.
In my opinion the manuscript needs some improvements I recommend to add and enhance the following sections: introduction, discussion and conclusions. The sections data, methods and results are written well and presented clearly for reading.
1. Coming to Introduction [Lines 59-67]
You wrote about global soil moisture products you are interested in the study. However I suggest to explain the role of the reference data in algorithms simulating SSM prudcts ECV COMBINED, ERA-5 Land, MERRA2 etc.
Please see Copernicus LMS SSM based on Sentinel-1 SAR satellite data available at the given source
https://land.copernicus.eu/global/products/ssm and find out on the reference observations for ssm simulations. Unfortunately Copernicus SSM spatial coverage concerns Europe only, while it is stated clearly in validation report that the copernicus ssm models are worked out on in-situ observations from dry, semi-dry regions. Thus each of ssm product you test has some limitations, that you need absolutetly mention in introduction. I hope you understand my point of view on quality of the satellite-based ssm products you evaluate for long-term characteristics.
2. Coming to Discussion (and results you received)
You applied RF regression model (figure no 9.). As far as I know for such kind of the data deep learning algorithm is not clearly effective for estimating performance. Based on experience on machine learning XGBoost algorithm, I suggest to mention in the discussion the possibilites for improving the results you obtained (R2 still low, RMSE is critical). Please refer to very interesting Discussion on the quality of XGBoost algorithm presented in
[Discussion, page 13 of 18], You can find out what it is written:
"XGBoost and deep learning models are both machine learning algorithms, but they differ in several ways: (a) Model architecture: XGBoost uses an ensemble of decision trees as base learners, whereas deep learning models use artificial neural networks that are composed of multiple layers.
(b) Input data format: XGBoost is well-suited for structured data that are arranged in rows and columns, while deep learning models can handle both structured and unstructured data, such as images and text. (c) Computational requirements: deep learning models typically require more computational resources than XGBoost, such as GPUs or TPUs, and can take longer to train. (d) Interpretability: XGBoost provides more interpretability
than deep learning models, as it can output feature importance scores and decision rules. In contrast, deep learning models are often considered “black box” models, as it can be difficult to interpret them and understand how they arrive at a particular prediction. (e) Performance on small datasets: XGBoost can perform well on small datasets, whereas deep learning models typically require large amounts of data to perform well"
Please cite the reference for adding sentence on opportunity of applicability xgboost regarding the above explanations. It might be very helpful in your future studies on long-term characteristics applying other ML techniques.
3. Coming to conclusions:
You should clearly state the limitations of satellite-based products. I am not for statement you wrote that ERA5-Land is worse/better performance product. The difference between in-situ and satellite-based soil moisture products is not the results of the quality of the product, rather the result on uncertainties and limitations of each of ssm product you studied. Please mind inter-comparison should not provide the answer for question (good/bad product because the agreement with your in-situ observations is low/average/good). I suggest you should rethink on the sentence and your statement which is absolutetly mispelled [Lines 480-481]
"ERA5-Land has the worst performance compared to all other products".
It is not true ERA5-Land has the worst performance. It is completely different product to others (basically it is reanalysis) and has some advantages/limitation and uncertainties you did not even mention.
I would suggest change the statement like:
"While performances of ssm products are at similar level, the Era5-land product seems to be slightly different due to specific characterizations of the products that might be not proper for evaluating ssm over tibetan plateu region". It is much safer for you to publish.
Thank you
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
In this study, soil moisture data from satellite and reanalysis products is validated over the TP region against two sets of in-situ observations. Furthermore, an analysis of long-term soil moisture characteristics has been performed, along with estimating the responses of soil moisture to climatic variables using a machine learning approach. The study is well-written overall, but there are several fundamental issues that need to be addressed, as outlined below:
1. In Figure 7, the study indicates that soil moisture sensitivity to longwave radiation is least. Can the author physical explanation for the same? In contrast, a recent study by Nayak et al. (2022) demonstrates higher soil moisture sensitivity to longwave radiation. This contrast requires a clear explanation.
[Nayak, H.P. Nayak, S. Maity, S. Patra, N. Singh, K.S. Dutta, S. Sensitivity of Land Surface Processes and Its Variation during Contrasting Seasons over India. Atmosphere 2022, 13, 1382. https://doi.org/10.3390/atmos13091382]
2. The validation observations used are not evenly distributed and are limited to a smaller region over TP. Therefore, the analysis may not be applicable to the entire TP region. Additionally, it appears that SSM/I soil moisture values are drier compared to other sources.
3. The study suggests that the ERA-5 trained model performed the best with the lowest RMSE (0.38) and the highest R2 (0.74). However, ERA-5 soil moisture performed the worst among all the products. This outcome could potentially lead to misleading information within the community. An explanation from the author regarding this apparent inconsistency is needed.
4. In lines 332-333, the statement "SSM/I outperforms GLDAS Noah using ZH2021 as the reference" is inaccurate. In reality, SSM/I only marginally performs better than GLDAS Noah.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Dear Authors,
Thank you for all your replies and comments. I figured out you improved highly the manuscript and answered all quetions very carefully. Thank you for all your explanations. I recommend accept in present form.
Best wishes
Reviewer 3 Report
The authors have addressed all the comments. The responses are satisfactory.