1. VALIDATION OF FINE RESOLUTION LAND-SURFACE ENERGY FLUXES DERIVED WITH COMBINED SENTINEL-2 AND SENTINEL-3 OBSERVATIONS IGARSS 2018 –
Radoslaw Guzinski, Internal Research Fellow, ESA / DHI-GRAS
H. Nieto, T. El-Madany, M. Migliavacca, A. Carrara

2. Outline What and why of evapotranspiration (ET) How to estimate ET
Results obtained with Sentinel observations
Conclusions

3. What and why of Evapotranspiration

4. Evapotranspiration

5. Transpiration through leaf stomata
Control gas exchange between leaf and atmosphere
Opened when sufficient light and water available
Maximize photosynthesis
Leaf temperature lowered
Closed when resources are limited
Conserve water
Leaf temperature increased
Credit:

6. Water use in agriculture
FAO AQUASTAT website. Food and Agriculture Organization of the United Nations (FAO). Website accessed on 2018/04/23
World population increase - 4.4
Water withdrawal increase - 7.3
Percentage of withdrawals for agriculture – 60 – 70%

7. Water use in agriculture
FAO AQUASTAT website. Food and Agriculture Organization of the United Nations (FAO). Website accessed on 2018/04/23
Spain – 60%
Italy – 44 %
Greece – 89%

8. HOW to estimate ET

9. Evapotranspiration – energy perspective
Rn – incoming solar and thermal radiation minus the reflected components
G – ratio of Rn reaching the soil
H – driven by temperature difference between surface/vegetation and air and controlled by wind speed and vegetation characteristics
λE – driven by moisture difference between surface/vegetation and air and controlled by wind speed and vegetation characteristics
Rn ≈ G + H + λE
Source:

10. Modelling of evapotranspiration
Two Source Energy Balance (TSEB) modelling scheme
Introduced by Norman et al. (1995)
Energy balance:
Two source:
github.com/hectornieto/pyTSEB
Source: Mecikalski et al., 1999

11. Data requirements Optical remote sensing data Leaf Area Index (LAI)
Fractional vegetation cover
Fraction of vegetation that is green
Albedo
Thermal remote sensing data
Land surface temperature (LST)
Meteorological data
Air temperature
Incoming solar radiation
Wind speed
Ancillary data
Vegetation height
Leaf transmittance

12. Data Sources Sentinel-2 – 20 m Atmospheric correction with Sen2Cor
LAI and fractional cover from SNAP
Albedo based on L7 algorithm
Sentinel-3 – 1000 m
Land Surface Temperature (LST) - doi.org/ /j.rse
ECMWF ERA-5 reanalysis meteorological data
Corine 2012 land cover map

13. Spatial sharpening Optical20 Thermal1000 Aggregate Optical1000
Establish relation
Apply relation
Thermal20
Implemented method based on Gao et al. (2012): A Data Mining Approach for Sharpening Thermal Satellite Imagery over Land
Homogenous sample selection
Regression tree + leaf linear regression
Local and global regressions
Residual analysis and correction
Bagging regressor
github.com/radosuav/pyDMS

14. Results obtained with Sentinel OBSERVATIONS

15. Majadas - site Period: December 2016 – November 2017
Location: Central Spain
Mediterranean savannah (dehesa)

16. Majadas - Results RMSE: 62 W/m2 RMSE: 60 W/m2 Relative error: 46 %
Statistics for LE, shown in blue
On the left using original resolution S3 LST
On the right using sharpened (20 m) S3 LST
RMSE: 62 W/m2
Relative error: 46 %
Correlation: 0.64
RMSE: 60 W/m2
Relative error: 44 %
Correlation: 0.69

17. True colour composite (20m - S2)
Majadas - comparison
True colour composite (20m - S2)
Maps of the area surrounding the flux tower, indicated as black dot. Left panel shows Sentinel-2 true colour composite on , while middle and right panels shows respectively fine scale and coarse scale instantaneous latent heat flux estimates obtained during Sentinel-3 overpass on
Fine resolution ET
(20 m - S2 & S3)
Coarse resolution ET
(1 km - S3)

18. Accuracy vs. revisit frequency
Statistics for LE, shown as blue dots
Using ERA5 EDA EM meteorological data and S3 images falling within days of S2 image.
RMSE: 95 W/m2
Relative error: 70 %
Correlation: 0.36
No. Observations: 86
RMSE: 71 W/m2
Relative error: 50 %
Correlation: 0.46
No. Observations: 57
RMSE: 64 W/m2
Relative error: 41 %
Correlation: 0.60
No. Observations: 37

19. Accuracy vs. revisit frequency
Statistics for LE, shown as blue dots
Using ERA5 EDA EM meteorological data and S3 with VZA <= 55 and +- days difference between S2 and S3 images as indicated.
RMSE: 68 W/m2
Relative error: 47 %
Correlation: 0.48
No. Observations: 53
RMSE: 66 W/m2
Relative error: 45 %
Correlation: 0.55
No. Observations: 43
RMSE: 53 W/m2
Relative error: 41 %
Correlation: 0.63
No. Observations: 32

20. Previous study: agricultural site

21. Previous study: agricultural site - results
Statistics for LE, shown as blue dots
RMSE: 51 W/m2
Relative error: 18 %
Correlation: 0.75
RMSE: 40 W/m2
Relative error: 14 %
Correlation: 0.92
RMSE: 42 W/m2
Relative error: 15 %
Correlation: 0.87

22. Previous study: agricultural site – spatial comparison
Landsat LST Sharpened MODIS LST

23. Sentinels for Evapotranspiration (SEN-ET)
Development of an efficient, open-source software implementation of a selected evapotranspiration methodology which exploits the synergies between the data collected by sensors on board of Sentinel-2 and Sentinel-3 satellites.
Literature review- report available online
Model selection:
Two Source Energy Balance (TSEB)
Once Source Contextual (Metric)
Two Source Contextual (ESVEP)
Two machine-learning sharpening methods
Validation site selection and data collection:
Agriculture, grassland, forests, savannah, wetlands
esa-sen4et.org

24. Conclusion

25. Conclusion
20 m ET can be estimated by sharpening S3 LST with S2 reflectance
Comparable, but not always same using high-res LST
Possibly reduced dynamic range
Method being refined and validated in SEN-ET
ET estimated every 5-6 days when S3A and S3B are taken into account
Frequency can be increased by reducing accuracy

26. Thank you