A Spatial–Temporal Bayesian Deep Image Prior Model for Moderate Resolution Imaging Spectroradiometer Temporal Mixture Analysis

Round 1

Reviewer 1 Report

Paper is interesting I have accepted the article with minor revision. Details comments attached in the PDF file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper can be published in the present form.

Author Response

Thanks for the reviewer's affirmation, we have made a more complete supplement to the manuscript. Hope the reviewer can find the newly submitted manuscript more academic.

Reviewer 3 Report

The paper proposes a method for spatial-temporal unmixing of time-series remote sensing images, which uses the U-Net architecture and TMA to obtain the enhanced abundance, and designs the posterior model under the framework of Bayesian, which better takes into account the heterogeneous noise. The method is innovative and valuable to be applied in practical applications, however, the article still has some problems, detailed as follows：

1. Even with the use of U-Net, it is difficult to obtain fine image abundance at the low resolution of MODIS data, so why not consider the use of spatial-temporal fusion to improve the resolution of the data?

2. Is the noise only Gaussian type noise considered, there will also be outliers in the time series data caused by pretzel noise, speckle noise caused by sensors and the atmosphere, etc., have these been taken into account?

3. There are only four features in the overall simulation data used, and the scenarios are relatively simple, while in reality the scenarios are more complex, can more complex scenarios be considered for simulation experiments?

4. Since the method is mainly for unmixing of time-series remote sensing data, the method chosen for comparison is the traditional unmixing method, and as far as I know, there are some methods that have been proposed for unmixing of time-series data, such as 'Spatio-temporal spectral unmixing of time-series images', so why not choose these methods for comparison?

5. In the experimental results of the real dataset, both the unmixed abundance results, as well as the extracted endmember curves, still have a large gap compared to the true value, and it is recommended to analyze the problems in the method based on the experimental results.

6. The authors seem to have omitted some important literature references, suggest authors to refer to the researches in image processing, like: A multiscale spectral features graph fusion method for hyperspectral band selection，IEEE TGRS.

Comments for author File: Comments.pdf

Written English should be improved carefully.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf