Evaluation of SMAP-Enhanced Products Using Upscaled Soil Moisture Data Based on Random Forest Regression: A Case Study of the Qinghai–Tibet Plateau, China
, Fengmin Hu 2, Junjie Li 1, Yijia Xie 1, Wen Zhang 1, Changqing Huang 1 and Lingkui Meng 1,*Round 1
Reviewer 1 Report
This is an interesting, relatively well-written paper that uses evidence from a variety of sources. I have some questions with the manuscript, and it would be improved by a through editing for English-language usage.
1. The regional characteristics of the Qinghai Tibet Plateau are relatively unique. Qinghai-Tibet Plateau is chosen as study area, is the evaluation of SMAP soil water data on the Qinghai Tibet Plateau universal for other plateaus?
2. In situ to 30 × 30, then to 9km × 9km, with such a large span, how can you control the accuracy?
3.How can you ensure the accuracy of the evaluation when the number of sample points is only 13?
4. Soil water is influenced by various environmental factors. How can you choose these factors? It is recommended to assign weights to different factors.
5. The results accuracy of soil water in upscaling 9km is very low. What is the reason for this? What methods can be improved? Is it a problem with the upscaling algorithm or the SMAP data?
It would be improved by a through editing for English-language usage.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The study proposed a two-stage scheme to upscale in-situ measurements to validate the accuracy of SMAP enhanced soil moisture product (SPL3SMP_E). The research concluded that the random forest regression approach is the best among six schema, SMAP satisfies its scientific measurement requirements (RMSE < 0.04 m3/m3), and SPL3SMP_E underestimate the soil moisture (with a positive bias = 0.0205 m3/m3 on average).
Some specifics:
1. Section 2.2: The paragraph describes that there are 13 in-situ measurement sites. On page 4 (Line 176), a single in-situ measurement is used to represent the soil moisture at 30 m around that site. In other words, a site measurement is corresponding to one pixel on the bridge data resolution (30 m). As selected Lansat data are used over the time series, the total number of site measurements may be given here. This number should indicate how many training samples are used in training (2/3) and test (1/3).
2. Section 4.3: The R values between upscaled validation data and remote sensing soil moisture are low, most around 0.5. This may weaken the validation in some sense.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
An upscaled scheme based on random forest regression is proposed for the validation of coarse-resolution SMAP soil moisture products in this paper. The topic falls into the scope of the journal and interests the readers. Generally, the paper is well-organized and presented. I would suggest that the authors consider the following comments in their revision.
1. “Converting small-scale data into large-scale” cannot provide “the real SM value of satellite scale”, so I would suggest revising the word “real” to “reference”.
2. “applied thermal inertia”-> apparent thermal inertia
3. When mentioning SMAP validation work, the following references are suggested to be cited.
Pan, M.; Cai, X.; Chaney, N.W.; Entekhabi, D.; Wood, E.F. An Initial Assessment of Smap Soil Moisture Retrievals Using High-Resolution Model Simulations and in Situ Observations. Geophys. Res. Lett. 2016, 43,9662–9668. [CrossRef]
Ma, C.; Li, X.; Wei, L.; Wang, W. Multi-Scale Validation of SMAP Soil Moisture Products over Cold and Arid Regions in Northwestern China Using Distributed Ground Observation Data. Remote Sens. 2017, 9, 327. https://doi.org/10.3390/rs9040327
4. Since mentioned SMAP’s radar, you may mention that it stopped working after July 2015 in line 71.
5. You may point out the new knowledge gained from the present work compared to the previous validation work.
6. It is a little confusing to me that the in-situ measurements span from 2013-2019 while the SMAP products span from 2015-2021, so they share 2015-2019 common time, which is not so “longer time series”, you may need to rephrase this statement.
7. Figure 1(b) is not necessary, and the authors can show the SMAP grid in Figure 1(a).
8. The performances of other algorithms (rather than RF and GBRT) “are not satisfactory”, how do you reach this conclusion?
9. The font size of Figure 4 can be enlarged to make it clear.
10. Section 4.2, what is “simulated soil moisture”? do you mean upscaled?
11. It is hard to see Figure 7.
12. Please revise the x- and y-axis labels of Figure 8 to make them in accordance with the text. SMAP (NOT Smap), Upscaled (NOT upscaling).
13. As shown in Figure 8, most values (both upscaled SM and SMAP products) are smaller than 0.12 m3/m3, and you get an RMSE of 0.031 m3/m3, so how did you draw a conclusion of “meeting the requirements of SMAP mission”. Of course, the mission requirement metrics of ubRMSE and RMSE for all conditions (such as extremely dry conditions) are not reasonable. We may consider the relative error in such conditions as well.
14. Based on the above comments, I think it does not make sense to show Table 2 since it cannot convey an informed message to the readers.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
