the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Dec 2024
A Globally Seamless Terrestrial Evapotranspiration Dataset Retrieved by a Nonparametric Approach with Remote Sensing and Reanalysis Datasets
Abstract. Evapotranspiration (ET) serves as a key indicator of the water change between the Earth’s surface and atmosphere, significantly influencing the hydrology cycle, surface energy cycle, and carbon cycle. Existing remote sensing models for estimating ET usually necessitate the parameterization of resistance parameters. In this study, we proposed the Remote Sensed Non-Parametric (RSNP) model, which leverages the nonparametric (NP) and Surface Flux Equilibrium-nonparametric (SFE-NP) approaches, and adapted remote sensing and reanalysis datasets of meteorological and surface parameters as model inputs. We estimate global monthly ET from 2001 to 2019 in the spatial resolution of 0.1° with RSNP model. Validation against FLUXNET sites globally yield RMSE of 23 mm/month (278 mm/yr), while regional-scale validation against water-balance ET results in a Root Mean Square Error (RMSE) of 113 mm/yr. In addition, the produced ET dataset have great accuracy in forest underlying and obtains spatial details of land surface ET. Furthermore, compared with ETMonitor, PEW and PML_V2, our dataset offers a continuous and seamless ET dataset suitable for global research. This study contributes to the advancement of global ET estimation and informs future water balance studies. The dataset presented in this article has been published in National Tibetan Plateau Data Center at https://doi.org/10.11888/Terre.tpdc.301343 (Pan, 2024).
- Preprint
(1616 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 01 Mar 2025)
-
RC1: 'Comment on essd-2024-495', Anonymous Referee #1, 28 Jan 2025
reply
General Comment
This paper describes a “globally seamless ET dataset…with remote sensing and reanalysis data”. The data is openly available at the National Tibetan Plateau Data Center for the period 2001-2019.
I believe that the paper requires a severe revision of its content, as it lacks most of the details behind the methodology adopted, details on the characteristics of the final products are difficult to find (is it monthly or daily? Is the model applied directly on monthly data or aggregated afterward?), and several key details on the validation are difficult to follow. In addition, numerous typos and unclear sentences can be found throughout the text (see examples in the specific comments below).
However, the most notable drawback of the dataset resides in its inception. The study claims that this is a “remote sensing based” dataset that “overcome the need for pre-defined parameters”. Regarding the first point, I found really difficult to see this as a remote sensing product, as the vast majority of the inputs came from ERA5-land. The remote sensing contribution is limited to emissivity and albedo only. This would not have been a major issue (beside the need to reword some of the model descriptions), but it highlights the second major problem of this dataset. The method uses skin LST from ERA5-land. These data are not observed from satellite, but they are modelled within the land surface component of the reanalysis system. It means that skin temperature depends on the parameterization used in ERA5-Land, the same parameterization that you are claiming to avoid. Following this consideration, the relationships used (1-1 and 1-2) acts only as a simplified version of the PM approach, where the skin temperature is derived from the more complex (and heavily parameterized) H-TESSEL.
Overall, the dataset may still have some useful applications related to multi-model assessment, but three key points need to be address: 1) a much better contextualization of the modelling framework and scope in view of the above-mentioned issue; 2) a much better description of the methodology, including differences from already existing approaches, and 3) an improved (especially in consistency) evaluation of the dataset against other similar products (e.g., ET from ERA5-land itself).
Specific comments
Title: A seamless global…
L17. Water exchange
L18-19. Hydrological, surface energy, and carbon cycles.
L20. Resistances. This is difficult to follow out of context into an abstract. Please reword.
L27. Explain acronyms.
L40. To conduct.
L41. Conducting. Repetition
L44. Sequentially?
L50-53. Add references to these datasets.
L54. Use consistent units for pixel size.
L58. Datasets available, they often…
L62. Metrology?
L62-63. In this sentence, it is not clear which problem (or problems) this dataset is trying to solve.
L66. A lot of repetitions (non-parametric) and unclarified terms (what is the role of Hamilton’s principle here).
L66-69. This sentence is unclear, please reword.
L76. I suggest introducing the methodology first, as explaining the data used, without introducing for what they are used for, make difficult to follow.
L78. As the inputs of
L79. To estimate ET…. Daily? Monthly? Not clear.
L80. At a spatial resolution
L81. Longwave radiation.
L87. Resempled… how? Especially land use, which is categorical.
Table 1. This table is not referenced in the text, as far as I can tell.
Table 1. Please clarify which input is from remote sensing and which from reanalysis. Also, please separate the model inputs from other data used for validation and analysis. The “data usage” column may not be seen by a reader, especially when things are mixed (2 retrievals, then validation, then retrieval again, …).
Table 1. Aridity index.
L93. Was the closure forced on the data. Which method?
L100. Some more details on this dataset are needed. A section (2.3) of just few lines is not acceptable.
Section 2.4. Same as before, some more details are needed. Modelling approach, main inputs, similarity in either inputs or methods, etc.
L116. Nearest neighbour method.
L116. “the differences…” at which time scale? Daily? Monthly? Again, not clear.
L122. Based on the Hamilton of microstate system,… Please reword, not clear the role of this principle on your method.
Table 2. This table is not referenced in the text.
L130. Some more details on the formulations are need. The reader needs to understand the basis of this approach without the need to go reading another full paper. For instance, the first term (Rn-G) is related to the available energy, and it is in common in all ET approach, but what about the other terms? What (Ts4-Ta4) represents? And the logarithmic term with Gs?
Eqs. (4) and (5) are not really needed, as they are basic physics. Please expand, instead, on the peculiarity of your method compared to other approaches. How do you “avoid parameters”?
L156. Why is the resampling needed? Just for a different projection? Not clear.
L157. How where water bodies, etc. excluded?
L158. Here there is the first reference to monthly scale, but it should be made clearer and it should be reported much earlier. Also, is the approach designed for monthly scale? Does eqs. 1-1 and 1-2 valid at monthly temporal scale?
L162. Aridity index.
L163. Aridity. Please fix throughout the text.
L163-165. Why 0.65 is used? Please add some reference to support this choice.
Fig. 3. The goal of the upper part of the figure (steps 1 to 5) is not clear. Is this just for gap filling? Very little is said about that in the main text.
L168. This sentence is unclear. How was this evaluated?
L173-176. This part of the pre-processing is very confusing. It needs rewording and expanding.
Section 3.3 I found this section mostly unnecessary.
L182-183. This sentence is not clear.
L186. This reference is not needed. These are standard metrics used in validation, not specifically introduced in that research.
L206. Absolute value
L227-229. This appears to be just 1 point. What is the point to compare this case with all the others? This analysis is very weak.
Fig. 8. Differences among models seems mostly systematic, so what is the point of showing multiple years? Wouldn’t be better to show the average year? Regional results would also be useful. Global average data are somewhat difficult to analyse.
L260. What is the difference? Please quantify. This is true for the entire results section, where often qualitative statements such as “is higher” is not accompanied by quantification.
L262-265. From the map in Fig. 9, it is not clear this difference in your dataset compared to the other. Also, here and later, there is a lot on emphasis on ET over the desert (missing values, values different than 0, etc.). Is this really that important? Are you expecting notable difference in water budget over these regions.
L274. Consistent behaviour with latitude.
L284. Seam?
L295-297. This result, and Fig. 10, raises the question: did you use the same dates for the analyses in Figs. 8 and 9 for all datasets? Average values should be computed on the same samples, so if you used a different number of dates for each dataset (based on availability or coverage) the results will be biased just for that and not for the differences in methodology.
As an example, if one dataset tends to have gaps during cloudy days, its average ET will be higher just because those cloudy days are not included. Please ensure consistency in the results reported.
L298-299. This statement is confusing. Mu refers to 24% of land surface. Where the 81% comes from? What is a middle-high latitude?
L302. Most of the missing values in the other datasets seems related to desert. How much the water balance can be compromised there when ET is mostly 0 anyway? The missing data is an important point, but it is more relevant in regions when ET is different than 0 when the data are missing. I will focus on these conditions to highlight your point.
L305. Shifts.
Fig. 11. Monthly availability… How is this monthly? Not clear.
L317. This dataset is seamless because is not a remote sensing-based product. If you try to use skin LST from satellite, then you would have a RS dataset but with some gaps.
L320. This statement is very confusing to me, first because skin LST from ERA5-land relies on these resistances, and second because the methodology does not explain how the method get rid of the resistances.
L327. Our dataset…
L328-332. This sentence is very confusing. Please reword.
L333. ERA5-land has already an ET product. You should include in your analysis a comparison with that product, as it is based on mostly the same forcings and is produced together with the skin LST used in this study. What is the added value of your methodology compared with what is already there?
L337. Residual surface energy balance is neglected here, which is weird as it is the more “direct” method from remote sensing to assess ET.
L337. “…may have similar systematic uncertainty”. Is your method so different from PM? A lot of the same factors plays a role in this method and in PM. In the text, you also confirm a lot of similarities between your dataset and the others. A better description of the methodology can help understanding the key differences and why it should not be affected by the same systematic differences as the other methods.
Citation: https://doi.org/10.5194/essd-2024-495-RC1
Data sets
Global seamless terrestrial evapotranspiration dataset (2001-2019) Xin Pan, Suyi Liu, and Jie Yuan https://doi.org/10.11888/Terre.tpdc.301343
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 307 | 78 | 8 | 393 | 9 | 14 |
- HTML: 307
- PDF: 78
- XML: 8
- Total: 393
- BibTeX: 9
- EndNote: 14
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1