Estimation of Topsoil Moisture on Bare Agricultural Soils at the Intra-Plot Spatial Scale Using a Statistical Algorithm and X- and C-Bands SAR Satellite Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript on estimating soil moisture at the patch scale is excellent. I have no further comments beyond a few minor revisions.

Lines 161-163. Can't tell which image is (a) and which is (b)? Sampling data is crucial. Please specify in detail: how many sites are involved? And how many sampling points are present at each specific sampling site?

 

In Section 2.2.1. Sampling Protocols, there are entries for Top soil moisture, Top soil moisture, and Soil roughness. Similarly, entries for Top soil moisture appear in Sections 2.2.2., Soil texture in Section 2.2.3., and Surface roughness in Section 2.2.4.

Could you clarify whether it would be preferable to differentiate the descriptions in the headings? Otherwise, the formal structure may lead to confusion.

What do (14) and (15) on lines 256-257 mean?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This paper focuses on the estimation of surface soil moisture in bare farmland at the intra-plot scale by using high-resolution X-band (TerraSAR-X) and C-band (Radarsat-2) SAR satellite data combined with random forest regression algorithm. The topic is closely related to the actual needs of precision agriculture and has important theoretical and application value.The research design is systematic. Through the comparative analysis of multi-scale (5-30m buffer zone and plot scale) and multi-sensor configuration (different polarization mode and incidence angle), the potential of SAR data in soil moisture inversion is comprehensively explored. The experimental data are detailed and the methodology is innovative. The research results can provide technical support for irrigation optimization and early detection of water stress.
1. In the calculation of the backscattering coefficient of TerraSAR-X satellite data, the determination basis and specific value of the calibration factor (CF) are not clear; the extraction process and source details of the gain parameters (A2i,j) of Radarsat-2 data are not clear, which affects readers 'judgment on the reliability of data preprocessing.
2. The specific methods of speckle noise removal in SAR images (such as multi-view processing, filtering algorithm, etc.) are not described, and only the noise effect is reduced by increasing the buffer band, and the quantitative evaluation of noise processing effect is lacking.
3. The conclusion that 30m buffer band is the best retrieval scale is attributed to the improvement of radiation resolution and the reduction of speckle noise, but the matching between this scale and the spatial variation scale of soil moisture is not analyzed deeply, and whether the accuracy will decrease when the buffer band is further increased (such as more than 30m) and the reasons are not discussed.
4. Although it was found that the variability of soil moisture within plots often exceeded the differences between plots, the dominant driving factors of variation (such as topographic slope, spatial distribution of soil texture, micro-topographic differences, etc.) and their interactions were not analyzed in depth, and only the anomalies of individual points (such as E5 point) were simply speculated, lacking systematic mechanism analysis.
5. There is no quantitative evaluation on the uncertainty of the whole process of the study, including in-situ sampling error, satellite data radiation correction error, model parameter uncertainty, etc., so it is impossible to clarify the contribution degree of each link error to the final inversion result, which affects the reliability and rigor of the study conclusion.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The article focuses on estimating surface water content using TerraSAR-X band and Radarsat C-band data. The authors implement a random forest method, taking into account surface variables (roughness, humidity, texture), radar configuration and radar data as input. Thanks to a robust database including numerous plots and dates, the authors obtain good estimates of water content with an RMSE of less than 4%. In addition, the authors show that working with averaged values within a 30 m radius leads to the best results. This allows for smoothing of speckle while respecting intra-plot variability.

My general comments concern

1)    The lack of discussion on the use of the statistical RF models in other years and other territories. Are the models transferable or is calibration necessary? This is a key element in the interest of the method and its generalisation.

2)    It would be really interesting to see the weight of all the variables in the accuracy of the model.

3) the quality of the Figures. The characters are too small

L58  I am not sure that TSM would be very usefull fo assess irrigation strategies. The TSM remains very superficial while irrigation strategies implies the knowledge of the SM in the root zone

L155 following variable are not “bio” but more “geo”. Consider geophysical.

L165 what is the depth of investigation

Surface soil texture  are the 16 samples mixed before making th e analysis

L178 I understand Harrowed and ploughed, but not prepared and worked. Please give more precision (type of tool for instance).

L180 consider spaced by one centimeter

Figure 4 difficult to read . lines too thins, red character two small, grey strips too pronounced

L247 less chaotic ??? Just a longer correlation length induced by the tillage tools. Note that the HRMS is larger in perpendicular  that can contribute to a chaotic pattern

L298 you give 3 buffers + PS while further mentioning 7 RF models

Figure 7 precise what are the parameters taken into account in the satellite configuarion

Figure 8 caption Circle rather than dot

Figure 9 not really usefull. Moreover I found discrepencies in the stat between Fig 9 and 8. The RMSE of the validation is always smaller than that of the calibration in Figure 9 which is not the case in Figure 8. For me the stat must be consistants between the Figures or I missed something

L523   the hypothesis of irrigation leaks is very speculative. There is only 4% of variation which can be explained by environmental factors easily

Fig 13 a soil moisture map showing the structure of the variability would be interestin, rather than focussing on the sampling transect

L543 there are no complementarity demonstrated in the paper. Just a coherence between radar configurations.

L547 speculative. The weight of roughness  and texture on TSM estimation was

L553-L555. Not really demonstrated. The fact that in some cases the intraplot variability might be larger than the between plot variability is not a surprise.

Author Response

Please see the attachment.

Author Response File: Author Response.docx