19.2 Remote Sensing of Surface Energy Fluxes and
19.2.2 Overview of estimation of Turbulent Fluxes by remote Sensing
Several algorithms have been developed in the last 4 decades for estimating the exchange of moisture and heat between the surface and the atmosphere from space- or airborne sys- tems; these are often used in combination with ancillary surface and atmospheric observa- tions� Table 19�1 summarizes some of the available satellite sensors providing data suitable for the retrieval of surface energy fluxes and soil water content� Overviews of the available methodologies can be found in studies by Diak et al� (2003), Courault, Seguin, and Olioso (2005), and Verstraeten, Veroustraete, and Feyen (2008)�
The vast majority of remote sensing–based methods employed today for estimating energy fluxes are fundamentally residual-based approaches, working based on the principle
of energy conservation� In such methods, net radiation (Rn) and soil heat flux (G) are com- puted using well-established approaches (see the review of Diak et al� 2003), whereas H flux is estimated based on the evaluation of difference between the surface radiometric tem- perature (Ts) and the Tair gradient at a single time, using only Ts (or a derivative quantity) as the surface boundary condition� Subsequently, LE flux is computed from the difference between Rn and G and H fluxes, based on the principle of energy conservation�
The simplest scheme was originally proposed by Seguin and Itier (1983) and elabo- rated on by Carlson, Capehart, and Gillies (1995)� This scheme uses the midday Ts and the net radiation to estimate the mean daily evapotranspiration� The two empirical coef- ficients for the simple equation are themselves variables, depending on the wind speed and fractional vegetation cover (Fr)� More physically based (but equally simple) estima- tion schemes propose the so-called one-layer models (Hall et al� 1992; Inoue and Moran 1997)� In these models, energy balance, temperature, and vapor pressure regimes of the vegetation canopy and the soil are not distinguished� Those models typically use Ts in place of the aerodynamic temperature and link H flux to the difference between Ts and Tair through a single aerodynamic resistance� However, with these models major problems exist in the interpretation of the derived results when the soil surface is partially covered by vegetation (e�g�, discussions by Kustas et al� 1989; Moran et al� 2005)� Because of par- tial plant cover, the surface temperature measured by a thermal infrared sensor will be a composite temperature between that of the vegetation and the soil substrate� Relevant studies have shown that these models tend to overestimate the H flux term, especially over sparse canopies, because the resistance to heat transport from the soil component within the sensor’s field of view is often significantly larger than the resistance above the canopy�
Table 19.1
Examples of Spaceborne Sensors Currently in Orbit Providing Observations Appropriate to Derive Surface Heat Fluxes and Soil Surface Moisture
Sensor Name Manufacturer Platform
Spatial Resolution
Spectral
Resolution Revisit Period
ASTER NASA/ERSDAC Terra VNIR: 15 m
SWIR: 30 m TIR: 90 m
VNIR: 4 SWIR: 6 TIR: 5
16 days
Landsat
TM/ETM+ NASA/U�S�
Department of Defense
Landsat VNIR: 30 m SWIR: 30 m TIR: 120 m (TM)/60 m (ETM+)
VNIR: 4 SWIR: 2 TIR: 1
16 days
MODIS NASA Terra and Aqua VNIR:
250 m/500 m SWIR: 500 m TIR: 1 km
VNIR:
18 SWIR:
2 TIR: 16
2 daytime/
2 nighttime
AVHRR NASA NOAA VNIR: 1�1 km
SWIR: 1�1 km TIR: 1�1 km
VNIR: 2 SWIR: 1 TIR: 2/3
2 daytime/
2 nighttime
AATSR ESA ENVISAT VNIR: 1 km
SWIR: 1 km TIR: 1 km
VNIR: 3 SWIR: 1 TIR: 3
2 daytime/
2 nighttime
SEVIRI EUMETSAT/ESA Meteosat-2 VNIR: 1�1
and 3�0 km, SWIR: 3 km
VNIR: 4
TIR: 8 96 scenes per day (every 15’)
Another drawback in the use of such methods is that no distinction is made between soil and vegetation components, which in turn makes impossible the identification of vegeta- tion stress conditions (see the review by Schmugge et al� 2002)�
As an improvement to these simple “one-layer” models, “two-layer (two-source)” mod- els have been developed� These models have included treatment for the temperature and energy balance regimes separately for the vegetation canopy and soil surface components, accounting for the variation in surface resistance due to the variation in vegetation cover and surface roughness (Norman, Kustas, and Humes 1995; Anderson et al� 1997; Brasa et al� 1998; Norman et al� 2000; Chehbouni et al� 2000; French, Schmugge, and Kustas 2002)� Validation of the LE and H estimates from such models has demonstrated a vary- ing degree of accuracy, despite the complexity with which the soil and vegetation com- ponents are treated in these types of models (Norman et al� 2003)� As noted by Kustas, Zhan, and Schmugge (1998), an important advantage of two-layer models is that they can be useful in interpreting aggregated flux estimates using bulk atmospheric boundary layer approaches over heterogeneous surfaces (Hipps, Swiatek, and Kustas 1994; Kustas and Norman 1996)� Furthermore, another important advantage of some of these two-layer models (e�g�, that of Mecikalski et al� 1999) is that they accommodate a view-angle depen- dence of surface brightness temperature� Consideration of the latter has shown that it can have a pronounced effect on the accuracy of the LE and H retrievals, especially over sparse vegetation where changes in the view angle cause large differences in the fractions of veg- etation and bare soil visible within the radiometer footprint� Nonetheless, one of the main drawbacks of two-layer models in comparison with one-layer models is their increased architectural complexity, which results in difficulties in their implementation, requiring a larger number of shelter-level meteorological inputs (primarily wind speed, aerodynamic resistance, friction velocity, air and aerodynamic temperature) and introducing errors in their representativeness (Beven and Fisher 1996; Jacob et al� 2002)�
A different approach for deriving spatial maps of land-surface heat fluxes from remote sensing observations—and in some cases also soil water content—is to place theoreti- cal boundary lines on the observed inverse relationship between an estimate of the land radiometric temperature and a spectral vegetation index (VI)� VI is an index related to the amount of vegetation present, often taken as Fr (Jiang and Islam 1999; Sandhold, Rasmussen, and Andersen 2002)� In such a scatter plot, the boundary lines may resemble a triangle (or trapezoid)� Even though it has been demonstrated that the derivation of spatially distrib- uted estimates of turbulent heat fluxes using the Ts/VI triangular scatter plot is feasible without the use of a boundary layer model (Moran et al� 1994, 1996), more sophisticated approaches (Gillies et al� 1997; Carlson 2007a) have proposed the retrieval of the above parameters from the combined use of the contextual interpretation of the Ts/VI domain with thermodynamic principles embodied in a two-layer surface/boundary layer energy balance (in particular, soil–vegetation–atmosphere transfer [SVAT]) models� This type of approach has certain advantages over one- and two-layer models, including a potentially improved ability to deal with surface heterogeneity (because this can be encapsulated in the VI measure), their potential to provide easier transformation between instantaneous and daytime average fluxes (which is often based on the conservation of a flux ratio, the Bowen ratio, during the day), ability to avoid dependence on external surface and meteoro- logical parameters, and that the key input data are relatively easy to obtain from space- or airborne data over large areas (i�e�, VI and surface radiometric temperature and nominal Tair), and that they allow for the correlation between the input (i�e�, Fr, Ts) and output vari- ables (soil water content and surface heat fluxes) to be nonlinear, contrary to the majority of all other analogous methods that are based on the Ts/VI pixel envelope or one- or two-layer
models that assume linear interpretations of the Ts/VI domain� Nevertheless, an important limitation of these approaches is the assumption that the temperature of live leaf surfaces is close to that for potential transpiration� This restriction, however, does not significantly affect the results as long as the vegetation is not seriously stressed� Another minor diffi- culty that conceivably restricts the use of these methods relates to the difficulty in choos- ing the appropriate parameters for the SVAT model, as their use requires some familiarity by the users; the latter will be shown to be relatively unimportant, however� Finally, the Ts/ VI methods require a large number of pixels to be sampled over a varied terrain�
Despite these potential impediments, the method of Gillies and Carlson (1995) and Gillies et al� (1997), the so-called triangle method, has been applied in a number of studies that have highlighted its potential for mapping surface heat fluxes and that have shown its abil- ity to provide distributed estimates of LE and H with an accuracy of around 25–55 W ⋅ m−2,
or about 10–30% (Gillies et al� 1997, Brunsell and Gillies 2003)� This accuracy is generally comparable to other Ts/VI scatter plot–based methods (Jiang and Islam 2001) and/or also some two-layer models (Norman et al� 2003)� These numbers are to be compared with the accuracy in the measurement of these fluxes using ground instrumentation, which is around 10–15% (as referred to in Jiang, Islam, and Carlson 2004; and also in Kustas and Norman 1996; Wilson et al� 2002)� Thus, current methods are pressing the limit of accuracy in the use of remote measurements to estimate LE, H, and soil surface water content�