Synergy of Sentinel-1 and Sentinel-2 Time Series for Cloud-Free Vegetation Water Content Mapping with Multi-Output Gaussian Processes

Round 1

Reviewer 1 Report

The manuscript aims to generate time series crop VMC from both S1 and S2. A multi-output GP was employed for joining both S1 and S2. IN general, the manuscript is clear, and its topic is of interest to the journal. My major concern is : Details about the two satellites, especially the temporal variability of satellite acquisitions used in the study were not clearly present. More importantly, the times of available satellites between S1 and S2 were different, how to make the VCM derived from S2 correspond to RVI of S1 is not clear. Details about samples selection and size used in the model calibration and validation are not clear as well.

Author Response

Dear reviewer thanks for your valuable suggestions. Your concerns have been addressed aiming to improve the quality of the manuscript. Please, find attached the response letter. 

Sincerely,

Gabriel Caballero & co-authors. 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

The research was good designed and well written. The methods were descried in detail and the results were also properly expressed. I believe it could be published.

Author Response

Dear reviewer, thanks for your valuable comments. We appreciate your time devoted to reading and analysing our manuscript. 

Sincerely,

Gabriel Caballero & co-authors. 

 

Reviewer 3 Report

To authors:

1. Introduction section

 The introduction provides sufficient background on the synergy of Sentinel-1(S1) and Sentinel-2 (S2) time series for cloud-free vegetation water content mapping with multi-output Gaussian processes. It includes literature review about the importance of water availability for proper crop growth and how vegetation water content (VWC) can be an effective indicator of water status for crops. It also highlights the limitations of optical remote sensing and how radar-based remote sensing through S1 can mitigate these limitations. The introduction also discusses the benefits of using S2 for crop trait mapping and the challenges of processing S1 and S2 data together. It further highlights how the Google Earth Engine platform offers a solution to these challenges by providing cloud-based processing of petabytes of S1 and S2 data. The paragraphs also discuss Gaussian processes, a nonparametric Bayesian machine learning regression algorithm, and how it can be used to create continuous, cloud-free S1 and S2 time series data streams. However, it could be beneficial to include more specific examples of the success of specific techniques used, such as spatial speckle filtering, Savitzky-Golay smoother, and active learning technique.

2. Research design

The research design follows a workflow consisting of three processing blocks (Figure 2) and six main steps (section 2.7). The three processing blocks involve acquiring and pre-processing Sentinel-1 (S1) and Sentinel-2 (S2) imagery, selecting the best multiple-output Gaussian processes (MOGP) model, and retrieving vegetation water content (VWC) using MOGP on S1 and S2 time series datasets. The six main steps include building a VWC time series by training a Gaussian processes (GP) model with in-situ data, assembling S1 and S2 datasets, configuring MOGP kernels, training MOGP models, mapping VWC retrieved over multiple seasons, and reconstructing artificially removed S2 GP VWC data gaps over winter wheat cropland. To improve the clarity of the research procedure, it would be beneficial to integrate the ordinal number of processing steps into the workflow to enhance the clarity of the research procedure.

3. Data sources and methods

 The datasets used are extensively described in sections 2.3 and 2.4, covering the preprocessing of S2 and S1 time series, respectively. However, an error was identified in item 2.4, where a typographical mistake led to the inclusion of the same number of images from the BVCR campaign (rows 216 and 217) corresponding to the same year. While Section 2.3 is informative and well-written, it could be improved by providing a brief explanation of some technical terms for the benefit of a wider audience. For example, defining terms such as VWC and including its calculation formula would help readers who may not be familiar with this concept. Additionally, introducing active learning techniques would allow readers to understand better the methodology used in the study. Section 2.4 provides a detailed description of the S1 time series preprocessing used to monitor a region of interest (ROI). However, for a more comprehensive understanding of the preprocessing steps, it would be beneficial to include brief explanations of spatial speckle filtering and additional Savitzky-Golay smoothing techniques. Furthermore, it would be helpful to explain the reasoning behind the selection of specific filter sizes and smoothing parameters. For example, the authors could clarify why the 7x7 refined Lee filter and Savitzky-Golay smoother with a window length of 9 and polynomial order of 2 were chosen.

 

4. Results  

The result section presents a detailed description of the temporal profiles of the S1 SAR radar vegetation index and S2 Gaussian process vegetation water component. It involves training MOGP kernels for vegetation water component time series modeling and Spatiotemporal mapping of reconstructed vegetation water components based on S1 and S2 synergy. A comprehensive discussion of the similarities between the S1 and S2 datasets in both the time and frequency domains, MOGP modeling and assessment, and S1 and S2-based spatiotemporal mapping of vegetation water content is included. Moreover, this section highlights the advantages and opportunities for improvement of the fusing approach. Although the output maps have a spatial resolution of 10 meters, the temporal resolution of the entire image datasets over the wheat cultivation land was not presented. Therefore, it would be beneficial to include information about the temporal resolution of the output images in the result section.

Author Response

Dear reviewer thanks for your valuable suggestions. Your concerns have been addressed aiming to improve the quality of the manuscript.  Please, find attached the response letter. 

Sincerely,

Gabriel Caballero & co-authors. 

Author Response File: Author Response.pdf