Last modified on Oct 09, 2023 16:13

Contributors: Aymn Elhaddad (GRO INTELLIGENCE), Stephan Meyer zum Alten Borgloh (GRO INTELLIGENCE), Allard de Wit (WAGENINGEN ENVIRONMENTAL RESEARCH)

Issued by:  GRO

Issued Date: April 2019

Ref: D422Lot1.WEnR.2.3.2

Official reference number service contract: 2017/C3S_422_Lot1_WEnR/SC2

History of Modifications

Acronyms

Scope of the document

The objective of this document is to provide a detailed description and justification of the algorithm proposed for version 1.0 of the MODGroETa algorithm. The main model objective is to provide estimates of vegetation actual and potential evapotranspiration by combining meteorological products from CDS ERA5 with Copernicus Global Land Service CGLS-based inputs consisting of land cover type, surface albedo (ALB), vegetation fractional cover (FC) and leaf area index (LAI). Actual and potential vegetation evapotranspiration data is provided with dekadal temporal resolution and
0.1 degree spatial resolution for global coverage in the period from 2000 to 2018.

Executive summary

This dataset contains MODGroETa model output which is actual and potential vegetation and soil evapotranspiration (ET) in millimetres with dekadal (i.e. 10 day) temporal resolution and 0.1-degree spatial resolution. The ET data available in this product is from 2000 to current with a lag time of two weeks. ET estimation is derived from daily ERA-Interim meteorological reanalysis data along with CGLS dekadal remotely sensed data products. In the near future, it will switch to the AgERA5 meteorological data when it becomes available through the CDS. The theoretical base of the modelling approach is the Penman Monteith equation. The MODGroETa model algorithm runs on a daily time scale that is then aggregated to dekadal, daily ET is the sum of ET from daytime and night. Vertically, ET is the sum of water vapour fluxes from soil evaporation, wet canopy evaporation and plant transpiration at dry canopy surface.

Figure 1 shows a sample output of the model representing the global dekadal actual evapotranspiration at 0.1-degree spatial resolution and aggregated over the period 08/20/2004 - 08/31/2004.

Figure 1: global dekadal aggregated (for the period from 8/20/2004 to 8/31/2004) actual evapotranspiration at 0.1 degree lat/lon grid.

Product description

MODGroETa v1

Introduction

Remote sensing has been used as a feasible mean to estimate regional evapotranspiration (ET) due to its spectral, spatial and temporal characteristics. Remotely sensed data obtained from satellites provides continuous near-real time data that is useful in the monitoring of land surface biophysical variables that impacts ET.

Implementation of the MODGROETa code satisfies the need to generate near real time (NRT) global evapotranspiration. The MODGroETa is a remote sensing-based evapotranspiration model developed by Gro-Intelligence using the same theoretical basis of MODIS Global Evapotranspiration (MOD16) that was developed by the National Aeronautics and Space Administration (NASA). The MOD16 evapotranspiration project was initiated by NASA's Earth Observing System (EOS) and implemented by the Numerical Terradynamic Simulation Group (NTSG) from the University of Montana to estimate global terrestrial evapotranspiration from land surfaces by using satellite remote sensing data. The MOD16 model uses a combination of meteorological and remote sensing ground surface data and is based on the Penman-Monteith equation. The MODGroETa model (Gro's version of MOD16) uses two sets of input data ,the first set is daily meteorological reanalysis data such as air temperature and solar radiation obtained from European Centre for Medium- Range Weather Forecasts (ECMWF) and the second set is Copernicus Global Land Service (CGLS) remotely sensed data products such as vegetation fractional cover (FC), albedo (ALB), and leaf area index (LAI).

Indicator definitions

This dataset contains MODGroETa model output which is actual and potential vegetation and soil evapotranspiration (ET) in millimetres with dekadal temporal resolution and 0.1-degree spatial resolution. The ET data available in this product is from 2000 to current with a lag time of two weeks. ET estimation is derived from daily ERA-Interim meteorological reanalysis data along with CGLS dekadal remotely sensed data products. In the near future it will switch to the AgERA5 meteorological data when it becomes available through the CDS. The theoretical base of the MODGroETa modeling approach is the same theoretical basis of MODIS Global Evapotranspiration (MOD16) that was developed by the National Aeronautics and Space Administration (NASA). The MOD16 evapotranspiration project was initiated by NASA's Earth Observing System (EOS) and implemented by the Numerical Terradynamic Simulation Group (NTSG) from the University of Montana to estimate global terrestrial evapotranspiration from land surfaces by using satellite remote sensing data. The MOD16 model uses a combination of meteorological and remote sensing ground surface data and is based on the Penman-Monteith equations. The MODGroETa model algorithm runs on a daily time scale that is then aggregated to dekadal, daily ET is the sum of ET from daytime and night. Vertically, ET is the sum of water vapour fluxes from soil evaporation, wet canopy evaporation and plant transpiration at dry canopy surface.

Input data used

Model Input Data:

The model inputs are daily and weekly data. The daily data is mainly climatological data obtained from the ECMWF reanalysis (ERA5 or ERA-int) and consists of five variables: daily minimum and maximum air temperature, daily mean air temperature, daily dew point temperature and daily shortwave incoming radiation. The second set of inputs (which is weekly) is obtained from (CGLS), the data covers land surface characteristics such as leaf area index (LAI), vegetation fractional cover (FC) and surface albedo (ALB).

Land Surface Characteristics

Leaf area index, fraction of vegetation coverage and albedo are dekadal composite products. The MODGroETa initially uses the near real time product (RT0) to produce near real time ET products, and those products get updated once the final version (RT6) of inputs is available. All QC layers associated with the obtained land surface characteristics variables are inherited and added to the model output.

ECMWF daily meteorological data:

The MODGroETa algorithm computes ET at a daily time step. This is made possible by the daily meteorological data, including mean, maximum and minimum air temperature, solar radiation and dew point temperature, provided by ECMWF ERA5 . All data is aggregated to daily time step.

Algorithms used

Model algorithm approach:

The classical approach for developing a robust algorithm to estimate evapotranspiration required explicit characterization of many physical parameters, some of which are difficult to determine globally. Thermal remotely sensed data used to be an integral part in most of the previously developed regional ET models. However, obtaining a cloud free global land surface temperature data has always been a challenge. To overcome this challenge on a constant basis, Cleugh et al. (2007) proposed the usage of the Penman-Monteith approach (Monteith (1965)) to estimate global ET.

Figure 2 shows the flowchart (Mu et al. 2007; ATBD) that presents the logic sustaining actual ET computation. It consists of input data ingested from meteorological and remote sensing observations, calculation of intermediary soil and vegetation processes, and the final outputs. The model currently accommodates inputs from different collections including daily meteorological data from ECMWF reanalysis (ERA5 or ERA-int) and land use land cover characteristics from (CGLS).

Figure 2: Flowchart of the improved MOD16 ET algorithm (Mu et al. 2007; ATBD).

Model Algorithm:

The estimation of ET is based on the conservation of either energy or mass, or both. Hence, the computation of ET combines two main processes: first, the estimation of stomatal conductance that initiates transpiration from plant surfaces; and second, the estimation of ground surface (soil) evaporation. The governing energy partitioning equation at the ground surface is expressed as:

𝐴ʹ=𝑅𝑛𝑒𝑡−Δ𝑆−𝐺=𝐻+𝜆𝐸

Where Rnet is net radiation (main energy source) and H, 𝜆E and Aʹ are the fluxes of sensible heat, latent heat and available energy; G is the soil heat flux; ΔS is the heat storage flux. 𝜆 is the latent heat of vaporization.

The MODGroETa algorithm runs on a daily time scale with global coverage. However, the model outputs a dekadal product. Outputs are actual and potential Evapotranspiration, with daily ET being the sum of ET from daytime and night. Vertically, ET is the sum of water vapour fluxes from soil evaporation, wet canopy evaporation and plant transpiration at dry canopy surface. The total daily actual ET (λE) and potential ET (λE_POT unlimited water supply condition) are calculated as follows:

λE=λEwet_c +λEtrans+λESOIL

λE_POT=λEwet_c +λEPOT_trans +λEwet_SOIL +λESOIL_POT

Output data and post processing

Model Output description:

The model output consists of two NetCDF files, each with five layers. The first dataset is the actual ET with the following filename format:

ET_C3S-glob-agric_GLS_CDS_EO_dek_20040820-20040831.nc

• ET_C3S-glob-agric_GLS_CDS_EO Product short name “ET” is actual
• dek_20040820-20040831 start and the end of the dekadal period of the
• .nc - Data format NetCDF.

.Product layers information:

Table 1: Description of netCDF ET file attributes

The second dataset is the potential ET with name format:

PET_C3S-glob-agric_GLS_CDS_EO_dek_20040830-20040831.nc (at 0.1 degree lat/lon grid)

• PET_C3S-glob-agric_GLS_CDS_EO Product short name “PET” is potential
• dek_20040820-20040831 start and the end of the dekadal period of the
• .nc - Data format NetCDF.

Product layers information:

Table 2: Description of netCDF PET file attributes

Product validation

Validation overview

Model Validation:

The model was tested over one full year (2015). The ground latent energy (LE) data used for model validation was obtained from 29 Eddy Covariance towers (EC) managed by different universities and delivered through AmeriFlux network. A minimum number of 40 valid 30min readings for each day of latent heat (LE) and air temperature at each EC tower was the condition to use the tower daily aggregated data for model validation. Figure 3 shows a schematic of the validation process and Figure 4 shows the EC towers locations.

Figure 3: Copernicus ET model validation process.

Reference data, Comparison methodology

Reference data and Comparison methodology:

The model's performance has been assessed with ET measured at 29 eddy covariance flux towers. Ground tower ET observations were compared to model ET at equivalent locations. Average model estimated ET was calculated over three footprint sizes 1km2, 3km2 and 10km2 (shown in Figure 5) surrounding each of the 29 towers using ECMWF and CGLS data. The model performed very well over homogeneous and uniform land cover such as cropland and wetland.

Figure 4: Locations of the Eddy Covariance tower used for model validation.

Figure 5: Different model validation resolutions 1,1 , 3,3 and 10,10 km.

Validation results

Results:

The correlation between the model ET estimates and the corresponding Eddy Covariance ET varied in the testing period. The correlation was strong over cropland with the Root Mean square error (RMS) from 1 to 1.7 mm/day, while the RMS was higher in other types of land cover. Table 1 shows the EC tower names and land cover type. As it can be seen in the table, in some cases the land cover reported in the EC tower fact sheet did not match the land cover in the land use layer used by the model, which might cause errors in ET estimates. This discrepancy in land cover type is unavoidable due to the accuracy level of the land use layer. For tower Twt, located in cropland, the average daily ET value for year 2015 estimated by the model at 1,3, and 10 km resolution was 4.7, 4.83, and 4.32 mm/day those values corresponding to ground tower ET values of 5.25, 5.19, and 4.9 mm/day respectively. The same comparison of average annual value of daily ET values of EC tower against model estimates was done for EC tower Tw3, located in a cropland. The average daily ET value for year 2015 around tower Tw3 estimated by the model at 1, 3, and 10 km resolution was 3.85, 4.45, and 5.11 mm/day those values corresponds to ground tower ET values of 3.95,3.89 and
3.93 mm/day respectively. The missing values in the table are due to the majority of pixels being water for that specific resolution. Model ET values were not available in some runs because of pixel QA/QC issues. Correlations between daily model ET estimates with its three resolution (1 km,3 km and 10 km) and the EC ET estimates are shown in Figures 6,7 and 8 in two formats. The first format is a plot of daily ET values from model and ground data against time for the year of 2015 and the second format is a plot of model daily estimated ERT against the corresponding ground calculated ET . Figure 9 depicts the comparison of annual actual ET observations from the EC tower sites and

the MODGroETa ET estimates averaged over the MODIS 3×3 km cut-out. These plots were created using tower-specific meteorology and the global ECMWF meteorology data.

Table 3: EC tower names, location and land cover types reported by EC and the land cover used in the model.

Figure 6: Root Mean square error (RMS) and correlations between MODGroETa model estimated ET and ground ET estimated using Twt EC tower with cropland land cover at three model resolution (1 ,3 and 10 km).The solid red lines represent that the ratio of ET estimates to ET measurements is 1.0 and the solid black lines are the regression of the ET estimates to measurements.

Figure 7: Root Mean square error (RMS) and correlations between MODGroETa model estimated ET and ground ET estimated using Tw3 EC tower with cropland land at three model resolution (1 ,3 and 10 km) cover. The solid red lines represent that the ratio of ET estimates to ET measurements is 1.0 and the solid black lines are the regression of the ET estimates to measurements.

Figure 8: Root Mean square error (RMS) and correlations between MODGroETa model estimated ET and ground ET estimated using Myb EC tower with wetland land cover at 3km model resolution, red line is 1:1 line.

Figure 9: Comparison of annual actual ET observations from the EC tower sites and the MODGroETa ET estimates averaged over the MODIS 3×3 km cut-out. These data were created using tower-specific meteorology and the global ECMWF meteorology.

Conclusions

Discussion: Model performance is largely impacted by two factors. The first, and the most dominant factor is the land classification layer, which is the key factor in determining all model coefficients that drive surface and aerodynamic resistance. The second most important set of factors for the model are LAI and FC. Resampling the LUT layer from 300 meters to 10km merges different land types and creates a new pixel that holds the code for most of the land cover type. The resampling of LAI and FC has less impact than LUT resampling on ET estimation accuracy since it is a mathematical averaging rather than applying a majority value as the case with LUT.

Conclusion: The model performance in cropland and wetland /grass was better than the model performance over open shrub land or savannah, and the model performance over forested areas was acceptable. Misclassification in LUT leads to the selection of the wrong parameters for vapour pressure deficit (VPD) and minimum air temperature (Tmin) for stomatal (cs) and canopy (cc) conductance constraints, resulting in less accurate ET estimates. Model results are most accurate in the middle of the season where LAI and Fc are high. Other uncertainties in the MODGroETa algorithm may also arise from: i) ECMWF re-analysis data, which are validated at the global scale, and may require more detailed analysis when used at regional scales; (ii) resampling of reanalysis data with spatial resolution of ∼75 km to 10 km, (iii) infilling missing and contaminated LAI and FC values with low quality data, and (iv) ground-based measurements by the eddy covariance system and in the tower footprint used to validate the model.

Concluding remarks

Recommendations for enhancing model accuracy:

1. Use native high resolution for inputs.
2. LUT used in the model validation is for the year of 2015; use a suitable LUT based on the year processed.
3. Update model land cover ID number if the used LUT is changed.

References

Chen, Y., J. Xia, S. Liang, J. Feng, J. B. Fisher, X. Li, X. Li et al. (2014) Comparison of satellite-based evapotranspiration models over terrestrial ecosystems in China. Remote Sensing of Environment 140 (2014): 279-293.

Cleugh, H. A., Leuning, R., Mu, Q., and Running, S. W. (2007). Regional evaporation estimates from flux tower and MODIS satellite data. Remote Sensing of Environment, 106(3), 285-304.

Jovanovic, N., Mu, Q., Bugan, R. D., and Zhao, M. (2015). Dynamics of MODIS evapotranspiration in South Africa. Water SA, 41(1), 79-90.

Jung, M., Reichstein, M., Ciais, P., Seneviratne, S. I., Sheffield, J., Goulden, M. L., ... and Dolman, A. J. (2010). Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature, 467(7318), 951-954.

Monteith, J. L. (1965). Evaporation and environment. Symposium of the society of experimental biology, 19, 205−224

Mu, Q., Heinsch, F. A., Zhao, M., and Running, S. W. (2007). Development of a global evapotranspiration algorithm based on MODIS and global meteorology data.
Remote Sensing of Environment, 111(4), 519-536.

Mu, Q., Jones, L. A., Kimball, J. S., McDonald, K. C., and Running, S. W. (2009). Satellite assessment of land surface evapotranspiration for the pan-Arctic domain. Water Resources Research, 45(9).

Mu, Q., Zhao, M., and Running, S. W. (2011). Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sensing of Environment, 115(8), 1781-1800.

Mu, Q., Zhao, M., and Running, S. W. (2011). Evolution of hydrological and carbon cycles under a changing climate. Hydrological Processes, 25(26), 4093-4102.

Mu, Q., Zhao, M., & Running, S. W. (2012). Remote Sensing and Modeling of Global Evapotranspiration. In Multiscale Hydrologic Remote Sensing: Perspectives and Applications (pp. 443-480), edited by Chang, N. B., Hong, Y., CRC Press.

Mu, Q., Zhao M., Kimball J. S., McDowell N., and Running S. W., (2013) A Remotely Sensed Global Terrestrial Drought Severity Index, Bulletin of the American Meteorological Society, 94: 83–98.
Renzullo, L. J., Barrett, D. J., Marks, A. S., Hill, M. J., Guerschman, J. P., Mu, Q., and Running, S. W. (2008). Multi-sensor model-data fusion for estimation of hydrologic and energy flux parameters. Remote Sensing of Environment, 112(4), 1306-1319.

Ruhoff, A. L., Paz, A. R., Aragao, L. E. O. C., Mu, Q., Malhi, Y., Collischonn, W., ... and Running, S. W. (2013). Assessment of the MODIS global evapotranspiration algorithm using eddy covariance measurements and hydrological modelling in the Rio Grande basin. Hydrological Sciences Journal, 58(8), 1658-1676.

Smettem, K. R., Waring, R. H., Callow, J. N., Wilson, M., and Mu, Q. (2013). Satellite- derived estimates of forest leaf area index in southwest Western Australia are not tightly coupled to interannual variations in rainfall: implications for groundwater decline in a drying climate. Global Change Biology, 19(8), 2401-2412.

Wang, S., Pan, M., Mu, Q., Shi, X., Mao, J., Brümmer, C., ... and Black, T. A. (2015). Comparing Evapotranspiration from Eddy Covariance Measurements, Water Budgets, Remote Sensing, and Land Surface Models over Canada. Journal of Hydrometeorology, 16(4), 1540-1560. MOD16 User's Guide MODIS Land Team
Version 1.5, 6/1/2017 Page 24 of 27

Yao, Y., Liang, S., Li, X., Hong, Y., Fisher, J. B., Zhang, N., ... and Jiang, B. (2014). Bayesian multimodel estimation of global terrestrial latent heat flux from eddy covariance, meteorological, and satellite observations. Journal of Geophysical Research: Atmospheres, 119(8), 4521-4545.

Zhang, K., Kimball, J. S., Mu, Q., Jones, L. A., Goetz, S. J., and Running, S. W. (2009).
Satellite based analysis of northern ET trends and associated changes in the regional water balance from 1983 to 2005. Journal of Hydrology, 379(1), 92-110.

Related articles

• Page:
  Climate variables for cities in Europe from 2008 to 2017 documentation
• Page:
  Climate Suitability for Tourism Indicators (CST): Dataset description
• Page:
  Fisheries and marine properties
• Page:
  Fish Abundance and Catch Dataset: Products User Guide
• Page:
  Ocean Fronts Dataset: User Guide for Products