GBRT-Based Estimation of Terrestrial Latent Heat Flux in the Haihe River Basin from Satellite and Reanalysis Datasets

Round 1

Reviewer 1 Report

1. It is recommended to make the keywords more generic and less ambiguous.  The keywords , in general , are well understood terminologies or phrases in the specific areas of scientific studies.  Thus, the keywords must be introduced in such a way that they must be understood at a glance by a global scientific community.  The mdpi journal recommendation is to choose 3-10 keywords that are specific to the article yet reasonably common within the subject discipline. 

 2.  Machine learning (ML) methods has been selected as a keyword. Among the list of many machine learning methods, only GBRT, RF and ETR are selected for the current study. What make these algorithms worthy over the other ML algorithms that are not considered for this study. Literature review of this aspect seems lacking in this study. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors should explain why they need this ML based ET algorithm if the physics based solution has been published 50 years ago and physics based operational MODIS ET product has been in production for 10 years. I can see ML methods being used as backup algorithms for ET production outside of MODIS land suite but this has not been elaborated in the paper.   

Major suggestions:

1. Please make a terminology choice in the paper and talk about ET(component of water balance) and report it in water volume /area/time units. Alternatively, talk about Latent Heat (component of energy balance closely connected to ET) and report it in energy/time/area units.
2. If authors choose ET please report annual and seasonal ET in cumulative fashion e.g. 700 mm/m2/year/.
3. If authors choose Latent Heat, then they should exclude nighttime observation form modeling as those would hold back the RMSE .
4.  Since authors are using EC data please show how MOD16 ET product performs in your flux tower locations as well, it is not too hard o aggregate data in 8 day cumulative or average values.
5. In your discussion please elaborate more on performance differences across different EC tower sites. Looking at your second cross-validation experiment one may draw conclusion that models for mixed forest should be trained on agricultural sites.
6. Why is MOD16 and GBRT ET compared only on annual averages of instantaneous. Readers would be more interested in seeing temporal components.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed all of my earlier comments. I recommend for the acceptance of the manuscript.