An all-sky 1&thinsp;km daily land surface air temperature product over mainland China for 2003–2019 from  MODIS and ancillary data

Chen, Yan; Liang, Shunlin; Ma, Han; Li, Bing; He, Tao; Wang, Qian

doi:https://doi.org/10.5194/essd-13-4241-2021

Articles | Volume 13, issue 8

https://doi.org/10.5194/essd-13-4241-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/essd-13-4241-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 13, issue 8

Data description paper

|

30 Aug 2021

Data description paper |

| 30 Aug 2021

An all-sky 1 km daily land surface air temperature product over mainland China for 2003–2019 from MODIS and ancillary data

Yan Chen, Shunlin Liang, Han Ma, Bing Li, Tao He, and Qian Wang

Download

Final revised paper (published on 30 Aug 2021)
Preprint (discussion started on 12 Mar 2021)

Interactive discussion

Status: closed

CC1:
'Comment on essd-2021-31', Joshua Fisher, 12 Mar 2021

Check out:
• Famiglietti, C.A., Fisher, J.B., Halverson, G., Borbas, E.E., 2018. Global validation of MODIS near-surface air and dew point temperatures. Geophysical Research Letters 45: doi.org/10.1029/2018GL077813.

Citation: https://doi.org/10.5194/essd-2021-31-CC1
- AC1: 'Reply on CC1', Yan Chen, 24 Mar 2021
  
  We would like to thank Dr. Fisher for the constructive comment on our manuscript. We have read the article you provided us. In the study conducted by Famiglietti et al. (2018), the near surface air and dew point temperatures were obtained by hypsometrically interpolating the MOD07 Level-2 atmospheric profile product to surface pressure level, and the results were in good agreement with surface measurements from 104 global stations. This is a classical and valuable study, which provides a feasible method for estimating surface air temperature using remotely sensed atmospheric profile products. Estimation of surface air temperature using MODIS atmospheric profiling products is also mentioned on page 3, lines 66-69 in our manuscript:
  In principle, the atmospheric profile products from satellite observations include temperature profile of the entire atmosphere, but usually require additional processes to obtain Ta. The Moderate Resolution Imaging Spectroradiometer (MODIS) atmospheric profile product has been used for this purpose (Bisht and Bras, 2010; Borbas and Menzel, 2017; Zhu et al., 2017).
  However, in general, there are some limitations to the method of estimating surface air temperature using atmospheric profile products. Since MODIS atmospheric profile product have available values only under clear-sky conditions, the surface air temperature obtained by interpolation is usually also under clear-sky conditions. Moreover, the relatively coarse spatial resolution (5 km) of MODIS atmospheric profile product limits the spatial resolution of the generated surface air temperature.
  Thank you very much for letting us know and learn about this study. We decided to quote it into the above part of the manuscript, so that those who are interested in this method can learn about this study.
  
  Citation: https://doi.org/10.5194/essd-2021-31-AC1
RC1:
'Comment on essd-2021-31', Anonymous Referee #1, 02 Apr 2021

It is of interest to users to have high spatial resolution Ta data by combining different data sources together using new numerical methods. The results are described in a clear fashion. I have following comments/questions:

1. I'm wondering whether the data from 2003-2016 or 2003-2019 is used and produced. There seems to be inconsistency in the paper regarding the temporal period of the study.

2. For vadiation of the study, how is the performance of the dataset/model if validation is carried out using a time period different from training period? For example, training is done using data from 2003 to 2016 and validation is done using data from 2017-2019? This is to see whether the training coeffients or RF models can be used after Terra/Aqua fail in the future.

3. I suggest to redo Figure 1 showing the number of data pairs and land types at these stations. You could use the color or the size of the symbol to provide such information.

4. Could you show the accuracy of the results as a joint function of surface types and surface temperature?

5. If the FI factors are small for surface radiation measurements, why not remove them from your model?

6. There are places in the paper using "temporary gap filling model", but it should be "temporal" instead of "temporary".

7. are the station Ta measurements used in the prediction of Ta?

Thank you.

Citation: https://doi.org/10.5194/essd-2021-31-RC1
- AC2:
  'Reply on RC1', Yan Chen, 26 Apr 2021
  
  We are very thankful to Referee #1 for the valuable and constructive comments on our manuscript, which improved the manuscript. We have revised the manuscript accordingly. A point-by-point response to all comments is provided in the supplement file.
  
  Citation: https://doi.org/10.5194/essd-2021-31-AC2
  - RC2: 'Reply on AC2', Anonymous Referee #1, 26 Apr 2021
    
    The authors have fully addressed the comments I had on the previous version of the manuscript. Thank you.
    
    Citation: https://doi.org/10.5194/essd-2021-31-RC2
RC3:
'Comment on essd-2021-31', Anonymous Referee #2, 15 Jun 2021
MODIS LST was used a proxy for Tair and had been successfully applied to estimate Tair for various regions of the world due to the ability to describe the surface-atmosphere energy exchange process. This paper uses the MODIS LST and ancillary data to create all-sky daily surface air temperature across China, and compares its results with three datasets with lower resolution. In gerneral, the ariticle is well structured and the method is clear, but there are several issues which should be solved.

My biggest concern is the statistic results may not be credible over southwest China. In southwest China, it rains in the most time of a year. The sunshine is rare that it is said “sunny weather seldom lasts for more than three days”. In other words, the percentage of missing MODIS LST data is over 60% due to the presence of clouds. To control the uncertainty introduced by LST gap-filling, a temporal window of 2 days in this study was used to fill the gaps. Please provide detailed availability of LST for cloudy-sky model â using this simple gap-filling method.

In section 3.2, GLDAS assimilated Ta was used in three models as input features. In the feature importance of those three models, assimilated Ta ranked first for two cloudy models and was second to Terra nighttime LST for clear-sky models. The second biggest concern for this study is that it seems like GLDAS assimilated Ta determines the RMSE and R². From the fourth paragraph in introduction part, no author used assimilated Ta as the predictor. Instead, shen [1] only used the soil mositure content, albedo and soil evaporation from GLDAS as predictors. If the ground-based Ta ingested by GLDAS was introduced as the preditor, whether it is a circular reasoning that reach better results? I would suggest removing the assimilated Ta as the predictor for three models.

The smallest concern is the spatioâtemporal model validation strategy in this study which just relys on random cross validation.

However, ignoring spatial and time dependence in model cross-validation can create false confidence in model predictions and hide model overfitting, and this problem that has been well documented in recent works [2, 3]. Please give explanations why this study still used an overlyâoptimistic approach (random cross validation) to assess the prediction error in both space and time.

Reference:

Shen, H., Jiang, Y., Li, T., Cheng, Q., Zeng, C., & Zhang, L. (2020). Deep learning-based air temperature mapping by fusing remote sensing, station, simulation and socioeconomic data. Remote Sensing of Environment, 240, 1–35. https://doi.org/10.1016/j.rse.2020.111692

Meyer, H., Reudenbach, C., Hengl, T., Katurji, M., & Nauss, T. (2018). Improving performance of spatio-temporal machine learning models using forward feature selection and target-oriented validation. , , 1–9. https://doi.org/10.1016/j.envsoft.2017.12.001

Ploton, P., Mortier, F., Réjou-Méchain, M., Barbier, N., Picard, N., Rossi, V., Dormann, C., Cornu, G., Viennois, G., Bayol, N., Lyapustin, A., Gourlet-Fleury, S., & Pélissier, R. (2020). Spatial validation reveals poor predictive performance of large-scale ecological mapping models. , (1), 4540. https://doi.org/10.1038/s41467-020-18321-y
Citation: https://doi.org/10.5194/essd-2021-31-RC3
- AC3: 'Reply on RC3', Yan Chen, 11 Jul 2021
  
  We are very thankful to Referee #2 for making the thoughtful and outstanding comments, which contributed to our manuscript. We have revised the manuscript accordingly. A point-by-point response to all comments is provided in the supplement file.
  
  Citation: https://doi.org/10.5194/essd-2021-31-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Yan Chen on behalf of the Authors (11 Jul 2021) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (review by editor) (13 Jul 2021) by David Carlson

AR by Yan Chen on behalf of the Authors (22 Jul 2021) Author's response Author's tracked changes Manuscript

ED: Publish as is (27 Jul 2021) by David Carlson

AR by Yan Chen on behalf of the Authors (01 Aug 2021)

Short summary

This study used remotely sensed and assimilated data to estimate all-sky land surface air temperature (T_a) using a machine learning method, and developed an all-sky 1 km daily mean land T_a product for 2003–2019 over mainland China. Validation results demonstrated that this dataset has achieved satisfactory accuracy and high spatial resolution simultaneously, which fills the current dataset gap in this field and plays an important role in studies of climate change and the hydrological cycle.