the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A Copernicus-based evapotranspiration dataset at 100-m spatial resolution over the Mediterranean region
Abstract. Evapotranspiration (ET) is responsible for regulating the hydrological cycle with a relevant impact on air humidity and precipitation, particularly important in the context of acute drought events in recent years. With the intensification of rainfall deficits and extreme heat events, the Mediterranean region requires regular monitoring to enhance water resources management. Even though remote sensing provides spatially continuous information for estimating ET on large scales, existing global products with spatial resolution ≥ 0.5 km are insufficient to capture spatial detail at a local level. In the framework of the ESA 4DMed-Hydrology project, we generate an ET dataset at both high spatial and temporal resolutions by the Priestley-Taylor Two-Source Energy Balance model (TSEB-PT) driven by Copernicus satellite data. We build an automatic workflow to generate 100-m ET product by combining data from Sentinel-2 (S2) MSI and Sentinel-3 (S3) land surface temperature (LST) with ERA5 climate reanalysis derived within the period 2017–2021 over four Mediterranean basins in Italy, Spain, France, and Tunisia (Po, Ebro, Hérault, Medjerda). First, original S2 data are pre-processed before deriving 100-m inputs for the ET estimation. Next, biophysical variables, like leaf area index, and fractional vegetation cover are generated, and then they are temporally composted within a 10-day window according to Sentinel-3 acquisitions. Consequently, decadal S2 mosaics are used to derive the remaining TSEB inputs. In parallel, we sharpen 1-km S3 by exploiting dependency between coarse-resolution LST and 100-m S2 reflectances using the decision trees algorithm. Afterward, climate forcings are utilized for modeling energy fluxes, and next for daily ET retrieval. The daily ET composites demonstrate reasonable TSEB-PT estimates. Based on the validation results against eight Eddy Covariance (EC) towers between 2017–2021, the model predicts 100-m ET with an average root mean square error of 1.38 mm day-1 and Pearson coefficient equal to 0.60. Regardless of some constraints, mostly related to the high complexity of EC sites, TSEB-PT can effectively estimate 100-m ET, which opens up new opportunities for monitoring the hydrological cycle on a regional scale. The full dataset is freely available at https://doi.org/10.48784/b90a02d6-5d13-4acd-b11c-99a0d381ca9a, https://doi.org/10.48784/fb631817-189f-4b57-af6a-38cef217bad3, https://doi.org/10.48784/70cd192c-0d46-4811-ad1d-51a09734a2e9, and https://doi.org/10.48784/7abdbd94-ddfe-48df-ab09-341ad2f52e47 for Ebro, Hérault, Medjerda, and Po catchments, respectively (Bartkowiak et al., 2023a–d).
- Preprint
(1890 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2023-466', William Kustas, 27 Mar 2024
The paper by Bartkowiak et al on the application of remote sensing-based energy balance model (TSEB-PT) in the Mediterranean region is well written with a robust scientific approach and analysis. However, there are two areas that the authors need to address in order for the reader to have a better understanding of the uncertainty in both the model and measurements. It appears the authors have chosen several challenging sites (some in complex topography) to conduct their model application and validation. This requires them to discuss in greater detail the measurement uncertainty. For example, they should make mention of the kind of energy balance closure they obtain at the different sites and if they used some method to force closure. Merely providing a reference to the processing of the eddy covariance data isn’t sufficient for the reader to easily interpret these results. In addition, for sites with sloping/complex terrain do they know if a planar fit was incorporated in post-processing the eddy covariance measurements (e.g., Ross and Grant, 2015)? Do the sites with complex topography have worse energy balance closure than more flat terrain? If so, this could factor into larger scatter observed at those sites. Finally, TSEB was not originally developed to be applied in complex terrain, although ways to incorporate refinements to TSEB for complex terrain is a worthwhile endeavor and should be mentioned.
I would also like to draw their attention to other studies that have been able to find better results over forested sites, although still a tendency for larger scatter (Hadi et al., 2022). Others have accounted for the green fraction from remote sensing and PT alpha term from knowledge of land cover (Guzinski et al., 2013; Andreu et al., 2018). Of course one may consider adjusting the PT alpha term a kind of tuning, but I am sure the authors are aware that land cover information should be used wherever possible since knowledge of the land cover type factors into a number of the TSEB-PT model parameters. Finally, there are other studies applying the multiscale version of TSEB that have obtained good results over pine forests (Yang et al 2017;2020). Although the authors do point out that some of the sites are challenging, I think they should also reference work that suggests applications of TSEB over forested areas can achieve reasonable results, especially when the surface is not complex.
References:
Andreu, A., Kustas, W.P., Polo, M.J., Carrara, A., & González-Dugo, M.P. (2018). Modeling surface energy fluxes over a dehesa (oak savanna) ecosystem using a thermal based two-source energy balance model (TSEB) I. Remote Sensing, 10, doi:10.3390/rs10040567
Guzinski, R., Anderson, M.C., Kustas, W.P., Nieto, H., & Sandholt, I. (2013). Using a thermal-based twosource energy balance model with time-differencing to estimate surface energy fluxes with day-night MODIS observations. Hydrol. Earth Syst. Sci., 17, 2809-2825
Jaafar, H.H., Mourad, R.M., Kustas, W.P., & Anderson, M.C. (2022). A global implementation of single and dual-source surface energy balance models for estimating actual evapotranspiration at 30-m resolution using Google Earth Engine. Water Resour. Res., 58, doi:10.1029/2022WR032800.
Ross, A.N. and Grant, E.R. (2015) A new continuous planar fit method for calculating fluxes in complex, forested terrain Atmos. Sci. Let., 16, 445–452
Yang, Y., Anderson, M.C., Gao, F., Hain, C.R., Semmens, K.A., Kustas, W.P., Normeets, A., Wynne, R.H., Thomas, V.A., & Sun, G. (2017). Daily Landsat-scale evapotranspiration estimation over a managed pine plantation in North Carolina, USA using multi-satellite data fusion. Hydrol. Earth Syst. Sci., 21, 1017-1037
Yang, Y., Anderson, M., Gao, F., Hain, C., Noormets, A., Sun, G., Wynne, R., Thomas, V., & Sun, L. (2020). Investigating impacts of drought and disturbance on evapotranspiration over a forested landscape in North Carolina, USA using high spatiotemporal resolution remotely sensed data. Remote Sens. Environ., 238, 111018
Citation: https://doi.org/10.5194/essd-2023-466-RC1 -
RC2: 'Comment on essd-2023-466', Anonymous Referee #2, 10 Apr 2024
Review of "A Copernicus-based evapotranspiration dataset at 100-m spatial resolution over the Mediterranean region" by Bartkowiak et al.
This paper presents an original evapotranspiration (ET) data set at 100 m resolution over 4 basins of the Mediterranean region : Ebro basin in Spain, Po basin in Italy, Herault basin in France and Medjerda basin in Tunisia. The main originality of the data set is its high spatial resolution compared to that of existing ET products classically available at 1 km or coarser resolution. The new 100 m resolution ET data set is derived by automatizing existing codes based on TSEB (Two-Source Energy Balance) model and Sentinel-2 (S2) and Sentinel-3 (S3) remote sensing data. The satellite-derived ET estimates are evaluated with eddy covariance measurements collected at 8 sites with 7 located in Italy and 1 in France and with several land covers (grassland, evergreen broadleaf forest, evergreen needleleaf forest and vineyard). Although the presented data set may be of interest for many different applications over the basins studied, I think that the evaluation strategy must be improved to really demonstrate the better accuracy of the new data set.
I recommend major revisions taking into account the concerns listed below :
1) Title, abstrat and conclusion :
The extent of the data set is confusing and somehow over-sold as the actual data set does not cover the entire Mediterranean region but only four selected basins within the Mediterranean region. It is true that the algorithms developed by the authors should work over other parts of the Mediterraean region but the paper focuses on the dataset. The authors should be more specific in the title, abstract and conclusion.2) Evaluation of the 100 m resolution evapotranspiration dataset :
The evaluation of satellite-derived evapotranspiration estimates is generally sound. However in my opinion it suffers from two major weaknesses. As outlined in the abstract and introduction and other parts of the paper, the rationale for developing a new ET product at high spatial resolution is that common products available at coarser spatial resolution are not sufficient to characterize the very high heterogeneity of land surfaces. The validation strategy of their product should support this key point. This is all the more needed as the ET product relies on the downscaling of 1 km resolution S3 land surface temperature (LST) data from 1 km to 100 m resolution. The evaluation of 100 m satellite-derived ET must be consolidated by estimating the gain in accuracy provided by the use of 100 m resolution remote sensing data, instead of 1 km resolution remote sensing data. One way of achieving this would be to implement PT-TSEB at 1 km resolution at the validation sites and calculate performance metrics as is done for the 100 m resolution dataset. Another drawback of the validation exercice is that it is based on only 8 stations, with 7 located in the same (Po) basin. Readers need to be convinced that this data set is significantly better than other more classical data sets.
The rationale for developing a new ET product at high spatial resolution :
Line 9-10 : "existing global products with spatial resolution >=0.5 km are insufficient to capture spatial detail at a local level"
Line 239 : "the Mediterranean region characterized by complex topography and highly patched landcover, where 1-km ET maps might not fully represent spatial heterogeneities of the land surface"3) Introduction :
- Second paragraph of the introduction : when the authors review existing evapotranspiration models, they mention process-based (energy balance models) and data-driven (statistical models) approaches. The so-called contextual/semi-empirical approaches are missed I recommend completing this state of the art by adding a few references to contextual methods.
- Line 96 : "many data-driven approaches have been proposed, relying on empirical relationships between 1-km surface temperatures and high-resolution explanatory variables derived from Synthetic Aperture Radar (SAR) and Visible Shortwave Infrared (VSWIR) sensors (Li et al. 2019 ; Mao et al., 2021 ; Pu and Bonafoni, 2023)." As none of the above references include SAR data I suggest this one: Amazirh et al. 2019. Including Sentinel-1 radar data to improve the disaggregation of MODIS land surface temperature data. ISPRS journal of photogrammetry and remote sensing, 150, 11-26.4) Eddy covariance measurements :
I could find no information on how the authors derived daily ET estimates from 30-min eddy covariance measurements. Equation (1) explains the unit conversion from W/m2 to mm/day, but the aggregation of hourly eddy covariance measurements at the daily scale is not described at all (?).5) Spatio-temporal coverage of the dataset :
I am surprised by the relatively large and frequent gaps in the ET dataset due to cloud cover. I imagine that the S2 dataset composited over 10 days and the S3 dataset composited over 10 days separately have greater spatial coverage. I wonder if the relatively low spatio-temporal coverage of the ET dataset is associated with the temporal mismatch between S2 and S3 overpasses ?6) TSEB modeling
One of the difficulties in spatializing the TSEB over large areas is characterizing the aerodynamic resistance (linked to canopy height, leaf size, etc.) and the green component fg of the vegetation cover. Can you briefly present the range of values chosen for these key parameters for the main vegetation types in the basins studied?7) Correction of input meteorological data for topography effects :
Line 337 : "All extracted variables from the reanalysis dataset, except for wind speed, are corrected for terrain using the SRTM DEM product"
Line 440 : "The distribution of solar radiation, wind speed, and air temperature gradients are less influenced by a landscape complexity over mountain plateau than over steep slopes, and thus coarse resolution ERA5 might be more representative for … "
Line 474 : "The ET models are controlled by climate inputs derived from 31-km fields… "
The above statements seem contradictory. Can you please describe how solar radiation and air temperature are dowsncaled at 100 m resolution using the DEM ? Both variables have a very strong effect especially in areas of complex topography such as the basins studied ?8) Shadows effects
Line 450 : « The poor accuracy at forested sites might be related to their complex tree structures and multilayer composition which is not considered in Sen-ET ». Since the authors are evaluating their product at Puechabon site, I suggest referring to Penot et al. (2023). Estimating the water deficit index of a Mediterranean holm oak forest from Landsat optical/thermal data: a phenomenological correction for trees casting shadow effects. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing9) Discussion of sources of uncertainty in the ET dataset (Lines 465-473)
Another source of uncertainty that should be mentionned and discussed is the intrinsic limitation of downslcaing methods of LST data using reflectances as high resolution ancillary information.Edits :
- Once defined, acronyms must be used systematically (this problem appears in many places in the text). Also acronyms should be defined only once. - There is an acronym for Languedoc Roussillon but not for the study basin (Herault basin) ?
- Composted/composting (line 17, fin 2, 396, line 486, )
- Unit is missing for RMSEs at line 584Citation: https://doi.org/10.5194/essd-2023-466-RC2
Data sets
Daily evaporation product - Ebro basin P. Bartkowiak et al. https://doi.org/10.48784/b90a02d6-5d13-4acd-b11c-99a0d381ca9a
Daily evaporation product - Herault basin P. Bartkowiak et al. https://doi.org/10.48784/fb631817-189f-4b57-af6a-38cef217bad3
Daily evaporation product - Medjerda basin P. Bartkowiak et al. https://doi.org/10.48784/70cd192c-0d46-4811-ad1d-51a09734a2e9
Daily evaporation product - Po basin P. Bartkowiak et al. https://doi.org/10.48784/7abdbd94-ddfe-48df-ab09-341ad2f52e47
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
302 | 63 | 18 | 383 | 11 | 15 |
- HTML: 302
- PDF: 63
- XML: 18
- Total: 383
- BibTeX: 11
- EndNote: 15
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1