Elsevier

Agricultural and Forest Meteorology

Volume 184, 15 January 2014, Pages 188-203
Agricultural and Forest Meteorology

An image-based four-source surface energy balance model to estimate crop evapotranspiration from solar reflectance/thermal emission data (SEB-4S)

https://doi.org/10.1016/j.agrformet.2013.10.002Get rights and content

Abstract

A remote sensing-based surface energy balance model is developed to explicitly represent the energy fluxes of four surface components of agricultural fields including bare soil, unstressed green vegetation, non-transpiring green vegetation, and standing senescent vegetation. Such a four-source representation (SEB-4S) is achieved by a consistent physical interpretation of the edges and vertices of the polygon (named T  fvg polygon) obtained by plotting surface temperature (T) as a function of fractional green vegetation (fvg) and the polygon (named T  α polygon) obtained by plotting T as a function of surface albedo (α). To test the performance of SEB-4S, a T  α image-based model and a T  fvg image-based model are implemented as benchmarks. The three models are tested over a 16 km by 10 km irrigated area in northwestern Mexico during the 2007–2008 agricultural season. Input data are composed of ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) thermal infrared, Formosat-2 shortwave, and station-based meteorological data. The fluxes simulated by SEB-4S, the T  α image-based model, and the T  fvg image-based model are compared on seven ASTER overpass dates with the in situ measurements collected at six locations within the study domain. The evapotranspiration simulated by SEB-4S is significantly more accurate and robust than that predicted by the models based on a single (either T  fvg or T  α) polygon. The improvement provided with SEB-4S reaches about 100 W m−2 at low values and about 100 W m−2 at the seasonal peak of evapotranspiration as compared with both the T  α and T  fvg image-based models. SEB-4S can be operationally applied to irrigated agricultural areas using high-resolution solar/thermal remote sensing data, and has potential to further integrate microwave-derived soil moisture as additional constraint on surface soil energy and water fluxes.

Introduction

Evapotranspiration (ET) plays a crucial role in predicting soil water availability (Oki and Kanae, 2006), in flood forecasting (Bouilloud et al., 2010), in rainfall forecasting (Findell et al., 2011) and in projecting changes in the occurence of heatwaves (Seneviratne et al., 2006) and droughts (Sheffield and Wood, 2008). The partitioning of ET into its surface components including soil evaporation, plant transpiration and canopy evaporation is important for modeling vegetation water uptake, land–atmosphere interactions and climate simulations. Large bare or partially covered soil surfaces especially occur in many cultivated areas. The soil evaporation term corresponds to the portion of ET that is unusable for crop productivity (Wallace, 2000) and its participation as a component of water balance may become dominant over bare or partially vegetated soils (Allen et al., 1998). Moreover, knowledge of ET partitioning would provide a powerful constraint on the physics of land surface models (Gutmann and Small, 2007). However, field measurements of both soil evaporation and plant transpiration are very sparse, and the current solar/thermal remote sensing techniques do not fully address the partitioning issue. This is notably due to the difficulty in separating the soil and vegetation components at the different phenological stages of crops from reflectance and thermal infrared data alone (Moran et al., 1994, Merlin et al., 2010, Merlin et al., 2012a).

A number of models have been developed to estimate ET from thermal remote sensing data (Courault et al., 2005, Gowda et al., 2008). Actual ET has been estimated by weighting the potential ET using reflectance-derived fractional photosynthetically active (green) vegetation cover (fvg) (Allen et al., 1998, Cleugh et al., 2007). fvg-Based modeling approaches are useful to provide ET estimates over integrated time periods e.g. the agricultural season. The point is that fvg is not sensitive to vegetation water stress until there is actual reduction in biomass or changes in canopy geometry (Gonzalez-Dugo et al., 2009). As a result fvg-based ET methods are not adapted to operational irrigation management when the objective is to detect the onset of water stress. Instead, canopy temperature can detect crop water deficit (Idso et al., 1981, Jackson et al., 1981). Operational ET models have hence been developed to monitor ET and soil moisture status from remotely sensed surface temperature (T) (Boulet et al., 2007, Hain et al., 2009, Anderson et al., 2012). Note that T-based ET models may also use fvg to partition soil and vegetation components (Norman et al., 1995), and surface albedo (α) as additional constraint on net radiation (Bastiaanssen et al., 1998). Among the T-based ET methods reviewed in Kalma et al. (2008) and Kustas and Anderson (2009), one can distinguish the single-source models (e.g. Bastiaanssen et al., 1998, Su, 2002) and the two-source models (e.g. Moran et al., 1994, Norman et al., 1995), which implicitly and explicitly represent soil evaporation and plant transpiration, respectively. Although both model representations may perform similarly in terms of ET estimates given they are correctly calibrated (Timmermans et al., 2007), the two-source models are of particular interest for ET partitioning.

Among T-based two-source ET models, one can distinguish the residual-based models (e.g. Norman et al., 1995, Anderson et al., 2007, Cammalleri et al., 2012), which estimate ET as the residual term of an aerodynamic resistance surface energy balance equation, and the image-based models (e.g. Moran et al., 1994, Roerink et al., 2000, Long and Singh, 2012), which estimate ET as a fraction (named surface evaporative efficiency or EE) of potential ET (Moran et al., 1994), or as a fraction (named surface evaporative fraction or EF) of available energy (Roerink et al., 2000, Long and Singh, 2012). In image-based models, EF (or EE) is estimated as the ratio of the maximum to actual surface temperature difference to the maximum to minimum surface temperature difference. In Moran et al. (1994) and Long and Singh (2012), maximum and minimum temperatures are estimated over the dry and wet surface edges of a polygon drawn in the T  fvg space, respectively. In Roerink et al. (2000), maximum and minimum temperatures are estimated over the dry and wet surface lines drawn in the T  α space, respectively. As clearly stated by Tang et al. (2010), the advantages of image-based models over the residual-based models are (1) absolute high accuracy in remotely sensed T retrieval and atmospheric correction are not indispensable, (2) complex parameterization of aerodynamic resistance and uncertainty originating from replacement of aerodynamic temperature by remotely sensed T is bypassed, (3) no ground-based near surface measurements are needed other than remotely sensed T, fvg and α, (4) a direct calculation of EF (or EE) can be obtained without resort to surface energy balance, and (5) estimations of EF (or EE) and available energy (or potential ET) are independent from each other by this method. Therefore, the overall errors in ET can be traced back to EF (EE) and available energy (potential ET) separately. Limitations of image-based models mainly lie in the determination of the maximum and minimum surface temperatures. Specifically, the dry and wet edges can be placed accurately in the T  fvg or T  α space if (1) the full range of surface (soil moisture and vegetation cover) conditions is met within the study domain at the sensor resolution, (2) meteorological conditions are uniform in the study domain (Long et al., 2011, Long et al., 2012), (3) the study domain is flat. In the case where all three conditions are not satisfied, alternative algorithms can be implemented to filter outliers in the T  fgv space (Tang et al., 2010), to estimate the maximum vegetation temperature from the T  α space (Merlin et al., 2010, Merlin et al., 2012b), to estimate extreme temperatures using a formulation of aerodynamic resistance (Moran et al., 1994, Long et al., 2012), or to correct remotely sensed T for topographic effects (Merlin et al., 2013).

Moran et al. (1994) proposed the T  fvg image-based water deficit index (WDI) to estimate a most probable range of crop water stress over partially vegetated pixels. The different steps of the WDI method are: (1) the temperatures of the four vertices of the T  fvg polygon are estimated using an energy balance model, (2) the minimum and maximum probable vegetation temperatures are estimated from the measured composite T, together with the maximum and minimum simulated soil temperatures, and (3) the minimum and maximum probable water stress indices are computed by normalizing the minimum and maximum probable vegetation temperatures from the vegetation temperature extremes simulated by the energy balance model. Note that the WDI approach does not allow estimating a single crop water stress index value because the canopy temperature retrieval problem is ill-posed using solely T and fvg. As mentioned in Moran et al. (1994) and Merlin et al. (2012a), knowledge of soil temperature would remove the underdetermination of the T  fvg polygon approach. A second limitation of the T  fvg polygon approach is that fvg does not allow distinguishing between soil and senescent vegetation, whereas the energy fluxes over bare soil and full-cover senescent vegetation may significantly differ. Separating vegetated areas according to the fraction of green versus senescent vegetation could be done by introducing additional information based on α (Merlin et al., 2010) or a vegetation index such as the Cellulose Absorption Index (Nagler et al., 2003, Krapez and Olioso, 2011). Note that optical data provide information on the surface skin only, which inherently prevents from separating green and senescent vegetation in the vertical dimension.

Roerink et al. (2000) proposed the Simplified Surface Energy Balance Index (S-SEBI) to estimate ET from the T  α space. S-SEBI determines the wet and dry lines by interpreting the observed correlations between T and α (Menenti et al., 1989). The wet line is defined as the lower limit of the T  α space. It generally has a positive slope as a result of an evaporation control on T. The dry line is defined as the upper limit of the T  α space. It generally has a negative slope as a result of a radiation control on T (Roerink et al., 2000). One main advantage of the T  α space over the T  fvg space is that α is sensitive to the total vegetation cover including green and senescent vegetation, whereas fvg is sensitive to the green vegetation only (Merlin et al., 2010). One drawback is that unstressed green vegetation, non-transpiring vegetation and senescent vegetation are not easily separable in the T  α space, which makes identifying green crop water stress more difficult than using the T  fvg space. Moreover the slope of both wet and dry lines may be difficult to determine when the full physical range of α (∼0.1–0.4) is not covered within the study domain.

Although T  fvg and T  α image-based models have been applied separately (Choi et al., 2009), or intercompared (Galleguillos et al., 2011), there is no model that combines the strength of each polygon notably in terms of ET partitioning. The objective of this study is thus to develop an image-based surface energy balance model (SEB-4S) that builds on advantages of both T  fvg and T  α spaces by (1) adequately constraining four surface components of agricultural fields including bare soil, unstressed green vegetation, non-transpiring green vegetation and standing senescent vegetation, (2) partitioning ET into soil evaporation and unstressed green vegetation transpiration, and (3) developing an automated algorithm for estimating temperature endmembers from joint T  fvg and T  α spaces. The modeling approach is tested over a 16 km by 10 km irrigated area in northwestern Mexico using ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) and Formosat-2 data collected on seven dates during the 2007–2008 agricultural season. Experimental data are described in Section 2. SEB-4S is described in Section 3, and two common (T  fvg and T  α) image-based models are reminded in Section 4. In Section 5, the surface fluxes simulated by SEB-4S, the T  fvg image-based model and the T  α image-based model are compared with in situ measurements at six locations.

Section snippets

Data collection and pre-processing

The Yaqui experiment was conducted from December 2007 to May 2008 over an irrigated area (27.25°N, 109.88°W) in the Yaqui valley (Sonora State) in northwestern Mexico. The campaign focused on a 4 km by 4 km area including 50% of wheat, the other 50% being composed of beans, broccoli, chickpea, chili pepper, corn, orange, potatoes, safflower and sorghum. The objective of the experiment was to characterize the spatial variability of surface fluxes from the field (hectometric) to kilometric scale.

SEB-4S model

SEB-4S is based on the classical surface energy balance equation applied to four surface components: bare soil, unstressed green vegetation, non-transpiring green vegetation and senescent vegetation. ET is computed as the sum of the four component latent heat fluxes. A key step in the modeling approach is therefore to estimate the component fractions. While Sections 3.1 Surface energy balance, 3.2 Model assumptions set the physical basis of SEB-4S, the following Sections 3.3 Estimating albedo

Image-based models

Two common image-based models are implemented as benchmarks to evaluate the performance of SEB-4S in estimating EF/ET. Although the T  α image-based model is similar to S-SEBI and the T  fvg image-based model similar to WDI, the objective is not to intercompare SEB-4S, S-SEBI and WDI, but rather to compare SEB-4S with image-based ET models having the same general structure as SEB-4S. In particular, the wet and dry edges are determined from the same temperature endmembers set in each case, and

Application

The simulation results of SEB-4S, the T  α image-based model, and the T  fvg image-based model are compared with the in situ measurements collected by the six flux stations. The objective is to evaluate model performances in terms of ET estimates in a range of surface conditions. Comparisons are made at the pixel scale by extracting the ASTER thermal pixels including a flux station.

Conclusions

An operational image-based surface energy balance model (SEB-4S) is developed from a consistent physical interpretation of the polygons obtained in the T  α and T  fvg spaces. The strength of the modeling approach relies on the synergy between both T  α and T  fvg polygons. Specifically, the combination of T  α and T  fvg image-based approaches allows to explicitly separate the energy fluxes of four surface components of agricultural fields including bare soil, unstressed green vegetation,

Acknowledgments

The participants of the Yaqui 2007–2008 experiment are gratefully acknowledged. This study is part of the MIXMOD-E project (ANR-13-JS06-0003-01) funded by the French ANR (Agence Nationale de la Recherche).

References (67)

  • S.B. Idso et al.

    Normalizing the stress-degree-day parameter for environmental variability

    Agric. Meteorol.

    (1981)
  • W.P. Kustas et al.

    Advances in thermal infrared remote sensing for land surface modeling

    Agric. Forest Meteorol.

    (2009)
  • D. Long et al.

    A two-source trapezoid model for evapotranspiration (TTME) from satellite imagery

    Remote Sens. Environ.

    (2012)
  • M. Menenti et al.

    Linear relationships between surface reflectance and temperature and their application to map actual evaporation of groundwater

    Adv. Space Res.

    (1989)
  • O. Merlin et al.

    Disaggregation of MODIS surface temperature over an agricultural area using a time series of Formosat-2 images

    Remote Sens. Environ.

    (2010)
  • O. Merlin et al.

    Self-calibrated evaporation-based disaggregation of SMOS soil moisture: an evaluation study at 3 km and 100 m resolution in Catalunya, Spain

    Remote Sens. Environ.

    (2013)
  • M.S. Moran et al.

    Estimating crop water deficit using the relation between surface-air temperature and spectral vegetation index

    Remote Sens. Environ.

    (1994)
  • P.L. Nagler et al.

    Cellulose absorption index (CAI) to quantify mixed soil-plant litter scenes

    Remote Sens. Environ.

    (2003)
  • J.M. Norman et al.

    A two-source approach for estimating soil and vegetation energy fluxes in observations of directional radiometric surface temperature

    Agric. Forest Meteorol.

    (1995)
  • L. Prihodko et al.

    Estimation of air temperature from remotely sensed surface observations

    Remote Sens. Environ.

    (1997)
  • G.J. Roerink et al.

    S-SEBI: a simple remote sensing algorithm to estimate the surface energy balance

    Phys. Chem. Earth

    (2000)
  • T.J. Schmugge et al.

    Recovering surface temperature and emissivity from thermal infrared multispectral data

    Remote Sens. Environ.

    (1998)
  • R. Tang et al.

    An application of the Ts-VI method with enhanced edges determination for evapotranspiration estimation from MODIS data in arid and semi-arid regions: implementation and validation

    Remote Sens. Environ.

    (2010)
  • M. Tanguy et al.

    A new parameterisation scheme of ground heat flux for land surface flux retrieval from remote sensing information

    J. Hydrol.

    (2012)
  • W.J. Timmermans et al.

    An intercomparison of the Surface Energy Balance Algorithm for Land (SEBAL) and the Two-Source Energy Balance (TSEB) modeling schemes

    Remote Sens. Environ.

    (2007)
  • J.S. Wallace

    Increasing agricultural water use efficiency to meet future food production

    Agric. Ecosyst. Environ.

    (2000)
  • Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration – guidelines for computing crop water...
  • M.C. Anderson et al.

    A climatological study of evapotranspiration and moisture stress across the continental united states based on thermal remote sensing: 1. Model formulation

    J. Geophys. Res.

    (2007)
  • L. Bouilloud et al.

    Coupling the ISBA land surface model and the TOPMODEL hydrological model for Mediterranean flash-flood forecasting: description, calibration, and validation

    J. Hydrometeorol.

    (2010)
  • G. Boulet et al.

    Monitoring water stress using time series of observed to unstressed surface temperature difference

    Agric. Forest Meteorol.

    (2007)
  • W. Brutsaert

    On a derivable formula for long-wave radiation from clear skies

    Water Resour. Res.

    (1975)
  • G. Campbell et al.

    An Introduction to Environmental Biophysics

    (1998)
  • J. Chirouze et al.

    Inter-comparison of four remote sensing based surface energy balance methods to retrieve surface evapotranspiration and water stress of irrigated fields in semi-arid climate

    Hydrol. Earth Syst. Sci. Discuss.

    (2013)
  • Cited by (67)

    • Estimating soil available water capacity within a Mediterranean vineyard watershed using satellite imagery and crop model inversion

      2022, Geoderma
      Citation Excerpt :

      This was explained by the strong dependence of vegetation transpiration and growth upon root zone SAWC and components, especially when vegetation faced water shortages. Including ET as a constraint variable permitted to increase the prediction of soil properties related to the root zone layer, which underlines the importance of developing robust methods to estimate ET from remote sensing, where current challenges are related to discontinuous canopies, heterogeneous landscapes and hilly areas (Aouade et al., 2020; Bellvert et al., 2021; Boudhina et al., 2018; Merlin et al., 2014; Zitouna et al., 2012; 2015; 2018). Similarly, including SSM as a constraint variable permitted to increase the prediction of soil moisture at field capacity and wilting point in the soil surface layer, which motivates continued efforts on the retrieval of surface soil moisture from remote sensing (Babaeian et al., 2019; Paolini et al., 2021), and especially over complex crop canopies such as vineyard (Fernandez-Moran et al., 2015).

    View all citing articles on Scopus
    View full text