Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

WATER RESOURCES RESEARCH, VOL. 45, W11422, doi:10.

1029/2009WR007900, 2009

A model of surface heat fluxes based on the theory of maximum


entropy production
J. Wang1,2 and Rafael L. Bras1,2
Received 23 February 2009; revised 4 August 2009; accepted 25 August 2009; published 19 November 2009.
[1] A model of heat fluxes over a dry land surface is proposed based on the theory of
maximum entropy production (MEP) as a special case of the maximum entropy principle
(MaxEnt). When the turbulent heat transfer in the atmospheric boundary layer is
parameterized using a Monin-Obukhov similarity model, a dissipation function or entropy
production function may be expressed in terms of the heat fluxes following the MEP
formalism. A solution of the heat fluxes can be obtained by finding the extreme of the
dissipation function under the constraint of conservation of energy for a given energy
input (i.e., net radiation) at the surface. The MEP solution of the surface heat fluxes is
tested using observations from fields experiments.
Citation: Wang, J., and R. L. Bras (2009), A model of surface heat fluxes based on the theory of maximum entropy production, Water
Resour. Res., 45, W11422, doi:10.1029/2009WR007900.

1. Introduction that need to be described probabilistically for making statistical


[2] Modeling heat fluxes remains a major challenge in the inferences. The MaxEnt states that ‘‘Out of all the possible
study of energy and water balance at the land surface. probability distributions which agree with the given constraint
Recent studies have concluded that no single land surface information, select the one that is maximally noncommittal
model is capable of capturing all features of the surface with regard to missing information’’ [Gregory, 2005, p. 185].
energy balance under all conditions [Desborough et al., The MaxEnt distribution is interpreted as the most probable
1996; Henderson-Sellers et al., 2003]. One major problem and macroscopically reproducible state among all physically
is identified as nonclosure of the surface energy balance, possible states. The generality and power of the MaxEnt are
indicating that existing models of surface heat fluxes still rooted in the Bayesian interpretation of probability that allows
have room to improve. The difficulty is, to a large extent, the concept of probability to be applicable to any situation
due to lack of a mature theory for nonequilibrium systems. where statistical inference based on incomplete information is
The classical nonequilibrium thermodynamics [de Groot sought. In that sense, the MaxEnt is much more than a physical
and Mazur, 1984; Kondepudi and Prigogine, 1998] is most law. Yet, it carries physical significance. For example, the
successful for the near-equilibrium systems, but inadequate (information) entropy, a central concept in the MaxEnt,
for the far-from-equilibrium processes [Jaynes, 1980] reduces to the familiar thermodynamic entropy for thermody-
including turbulence in the atmospheric boundary layer. namic systems at equilibrium [Tribus, 1961]. The link between
Even though the classical treatment of turbulent transfer the information entropy and the thermodynamic entropy
in the atmosphere [Priestley, 1959] works reasonably well becomes even more intimate and intuitive in the MEP theory
in practice, further improvement in modeling the surface [Dewar, 2003]. Nonetheless, it is important to emphasize that
energy balance needs a better tool than offered by the the entropy (production) in the MEP theory is not necessarily
classical nonequilibrium thermodynamics. related to the thermodynamic entropy (production).
[3] The recent advances in the nonequilibrium thermo- [4] Since Paltridge [1975] used the MEP argument to
dynamics including the fluctuation theorem and the hypoth- explain the basic patterns of global climate in terms of the
esis of maximum entropy production (MEP) [e.g., Dewar, maximum rate of thermodynamic dissipation toward a
2005] offer fresh approaches to modeling surface fluxes. steady state, there have been a number of applications of
The MEP theory is a derivative of the principle of maximum the MEP theory in earth and planetary climatology and
entropy (MaxEnt) first formulated as a general method to hydrodynamics [Ozawa et al., 2003]. During the last
assign probability distributions in statistical mechanics decade, the MEP theory has been tested as an organizing
[Jaynes, 1957]. The theoretical foundation of MaxEnt is principle governing biological and ecological systems. For
Bayesian probability theory [Jaynes and Bretthorst, 2003] example, Kleidon and Fraedrich [2004] have shown that
where the concept of entropy is defined as a quantitative global biotic productivity corresponds to the states of MEP.
measure of information [Shannon, 1948] for any systems Juretic and Zupanovic [2003] found that the MEP may
govern the photosynthetic processes. Evidence accumulated
so far suggests that the MEP theory could explain transport
1
Department of Civil and Environmental Engineering, University of phenomena in the far from equilibrium systems over a wide
California, Irvine, California, USA. range of space and time scales [Kleidon and Schymanski,
2
Formerly at Ralph M. Parsons Laboratory, Massachusetts Institute of
Technology, Cambridge, Massachusetts, USA. 2008]. We intend to show that the MEP not only explains the
observed behavior of nonequilibrium thermodynamic pro-
Copyright 2009 by the American Geophysical Union. cesses, but also can provide a predictive tool in modeling the
0043-1397/09/2009WR007900 transport processes quantitatively. This paper investigates the
W11422 1 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

case of heat fluxes over a dry land surface as a proof-of- where lk (1  k  m) are the Lagrange multipliers
concept study, which ultimately could lead to a MEP model associated with the constraints Fk, which are related through
of surface energy balance including latent heat flux. the well known Legendre transform.
[5] Traditionally, ground heat flux is derived from the [8] The MEP theory results from the particular situation
gradient of soil temperature across a small depth near the of antisymmetric functions fk in equation (2) satisfying
surface given the thermal conductivity of the soil materials fk(xi+) = fk(xi) when all potential outcomes, xi, can be
[e.g., Sellers, 1965]. Ground heat flux may also be derived grouped in pairs (i+, i). For anti-symmetric functions
from the time history of skin temperature as the heat transfer (whose physical meaning becomes clear later in the paper),
in the soil is described by a diffusion equation given the the corresponding MaxEnt distribution satisfies the generic
thermal inertia of the soil materials [Wang and Bras, 1999]. ‘‘fluctuation theorem’’ (FT),
Turbulent sensible heat is often derived from the gradient
 X m 
of air temperature according to the well-known Monin- piþ
Obukhov similarity theory [Monin and Obukhov, 1954]. ¼ exp 2 lk fk ðxiþ Þ ; ð4Þ
pi k¼1
Also based on the Monin-Obukhov similarity theory is the
flux variance model [Tillman, 1972; Katul et al., 1995]
The FT implies that some microscopic states or phase space
where sensible heat flux is derived from the standard
trajectories are overwhelmingly favored over the others,
deviation of turbulent fluctuations of air temperature (at a
leading to macroscopic transport of heat and matter, unless
single level). In these classical models, ground and sensible
the exponent of the exponential function in equation (4) is
heat fluxes are estimated using either single-level or multi-
always close to zero.
ple-level temperature records (with other necessary input
[9] The behavior of the MaxEnt distribution for the
parameters). As shown later in the paper, the proposed MEP
antisymmetric functions is determined by the properties of
model of heat fluxes over a dry soil is different from the
the exponent of the exponential function in equation (4),
existing models in that the MEP model of ground and
whose mathematical expectation
sensible heat fluxes does not use soil and air temperature
records if net radiation is available. X
m
[6] The paper is organized as follows. We first review the D2 lk Fk ; ð5Þ
MEP formalism following Dewar [2005] in section 2. The k¼1
formulation and validation of the MEP solution of ground
and sensible heat fluxes is presented in section 3. Further is referred to herein as the ‘‘dissipation function’’ or
understanding of the partition between ground and sensible ‘‘entropy production function’’ for a reason explained by
heat fluxes from the perspective of an optimality principle is Dewar [2005].
the subject of section 4, followed by concluding remarks [10] D in equation (5) satisfies orthogonality conditions
(section 5). [see Dewar, 2005, equations (18) and (19)], which imply
that D reaches an extremum given certain constraint on the
2. MEP Formalism Fks. D is maximum when the constraint is a certain
nonlinear (e.g., quadratic) function of Fk, and minimum
[7] The details of the MEP formalism are described in when that constraint is a linear function. Therefore, ‘‘max-
the original work of Dewar [2005]. Here we summarize imum entropy production’’ is in fact a misnomer in the
the key results that are most relevant to the proposed sense that D could reach either a maximum or a minimum
MEP model below. The MEP is a derivative of the MaxEnt. or a saddle point, theoretically, depending on the functional
In the MaxEnt formalism, the probabilities pi of variables xi form of the constraint on Fks. The well-known Prigogine’s
(1  i  n where n is a positive integer) are derived by minimum entropy production theorem can be viewed as a
maximizing the Shannon information entropy, SI special case of the general linear constraint, leading to a
minimum D. Hopefully, MEP will be replaced by a more
X
n
SI   pi ln pi ; ð1Þ suitable term in the future to avoid confusion. In the
i¼1 development of MEP model below, we deal with the
situation of a linear constraint on the unknown surface heat
subject to the constraints fluxes through the surface energy balance equation.
[11] We emphasize that the orthogonality conditions of D
X
n may not be exact since the derivation invokes a quadratic
hfk i  pi fk ðxi Þ ¼ Fk ; 1  k  m; ð2Þ approximation of the exponent function in equation (4).
i¼1 Grinstein and Linsker [2007] argued that this approximation
implies that the orthogonality conditions are valid only for
where fk are functions of xi, Fk are given parameters (as the near-equilibrium systems where Fks are close to zero [see
constraints) representing available information about xi, and Grinstein and Linsker, 2007, equation (2)]. A careful
m ( n) is a given integer. The brackets stand for examination of Grinstein and Linsker’s derivation indicates
mathematical expectation. The MaxEnt distributions pi that Grinstein and Linsker’s conclusion is in part due to a
can be derived by maximizing SI in equation (1) subject mathematical artifact. (It is incorrect to differentiate lk with
to the constraints in equation (2), leading to respect to Fn as in equation (2) of Grinstein and Linsker
X
m  [2007] because equation (15) of Dewar [2005] is only an
pi / exp lk fk ðxi Þ ; ð3Þ expression of lk evaluated at given ~ F. @lk/@Fn must be
k¼1 obtained from equation (10) or equation (9) where S(~ F) is

2 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

specification of fk indeed can be done for the case of land


surface energy balance, of interest in this paper. Below we
show how to use an idealized case (described as the toy
model below), where the analytical function D is known, to
formulate D involving turbulent heat flux through analogy.

3. Application of MEP in Land Surface Energy


Balance
[13] The surface heat fluxes predicted by the MEP theory
are referred to as the MEP solution of surface heat fluxes
herein. Using the MEP framework outlined above, it would
be more realistic to formulate D without getting into
microscopic details of molecular conduction in the soil
and turbulent diffusion in the air. Anticipating some com-
monality of transport processes in nonequilibrium systems,
we start by building a toy model of the surface energy
balance where heat transfer is assumed to be through
conduction. The MEP solution of the toy model not only
reveals the functional form of D, but also elucidates the
physical significance of the Lagrange multipliers when the
constraints are imposed on fluxes.
3.1. A Toy Model
[14] Consider one-dimensional heat conduction in one
(vertical) semi-infinite column (labeled 1) on top of the
Figure 1. Graphical illustration of the toy model config-
other (labeled 2) with different thermal properties. Figure 1
uration. The top graph refers to z > 0, and the bottom graph
illustrates the configuration of the system. A source of heat
refers to z < 0. T1(z, t) and T2(z, t) are temperature at
is located at the interface z = 0 with a prescribed input heat
location z and time t within the top graph and the bottom
flux F0 varying with time. Conservation of energy requires
graph, respectively. F1 and F2 are heat fluxes at the
F0 to be distributed between two heat fluxes into the two
boundary of the top graph and the bottom graph,
columns, F1 and F2, which are defined as positive when
respectively, which are defined as positive going toward
heat flows away from the interface. F1 and F2 can be
the interior of the two media. The thermal properties of the
determined by solving the corresponding diffusion equa-
corresponding medium include bulk density r, specific heat
tions governing the heat transfer within the media as
c, thermal conductivity m, thermal diffusivity k, and thermal
described in Appendix A. An analytical solution of F1
inertia I.
and F2 is given in equations (A25) and (A26),
given in equation (8) of Dewar [2005]). Grinstein and Ik
Linsker are correct that the quadratic approximation is less Fk ¼ F0 ; k ¼ 1; 2;
I1 þ I2
accurate than hoped by Dewar [2005]. Nonetheless, the
formulation of MEP for the case of Fk not close to zero put where I1 and I2 are the thermal inertia of the two media
forward by Dewar [2005] does not invalidate the MEP defined in equation (A19), respectively.
hypothesis for far from equilibrium systems. A case study of [15] The solution of Fk (k = 1, 2) can be obtained without
land surface energy balance presented below provides solving the differential equations. It is straightforward to
evidence that the MEP hypothesis may hold for a far from verify that the exactly the same solution of Fk results from
equilibrium system where constitutive relations such as minimizing D defined as
turbulent heat transfer are highly nonlinear. This is a topic 2F12 2F22
for ongoing and future research. D þ ; ð6Þ
I1 I2
[12] The key to apply the MEP method is to obtain an
expression of D where lk must be expressed as explicit under the constraint equation (A9), i.e., F1 + F2 = F0, with
functions of Fk according to the Legendre transformation lk in equation (5) expressed as
once fks in equation (2) are identified. For the case of heat Fk
exchange processes at the land surface, fk may include the lk ¼ ; k ¼ 1; 2; ð7Þ
Ik
kinetic energy of molecules or turbulent eddies, etc. Iden-
tification of fk requires deep understanding of the physical which is recognized as the orthogonality conditions given
processes at a microscopic level and characterized in by Dewar [2005, equations (18) and (19)].
mathematically tractable forms. Yet most of the microscopic [16] The physical meaning of lk becomes clear once Fk is
details of the physical processes, difficult to fully capture, expressed in terms of temperature at the interface [Wang
are irrelevant to the macroscopic observable properties. and Bras, 1999]
Therefore, it is possible to find D without knowing fk
Zt
exactly. The situation is analogous to the classical equilib- Ik dTk ðt; 0Þ
rium thermodynamics where the states of individual mole- Fk ¼ pffiffiffi pffiffiffiffiffiffiffiffiffiffi ; k ¼ 1; 2 ð8Þ
p tt
cules play no role in the ideal gas law. Finding D without 0

3 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

where Tk(t, 0) is the surface temperature and t is the [20] Following the eddy diffusivity defined in equation
integration variable. Comparing equation (8) with equation (B12) and the proposed extremum solution based on the
(7) reveals that lk is the half-order time derivative of the MOST summarized in Table B1, we obtain an expression of
temperature at the interface. The validity of the MEP Ia ,
argument is confirmed by the physical reality of a
continuous distribution of temperature throughout the two pffiffiffiffiffiffiffiffiffiffi 1
kzg 6 16 1
columns at all times. The straightforward application of the Ia ¼ rCp C1 kz C2 jHj  I0 jHj6 ; ð10Þ
rCp T0
MEP demonstrates its power in deriving highly nontrivial
results from seemingly trivial propositions.
[17] The toy model is an example where the concept of where r, Cp, k, g and z carry their usual meanings, and T0 is
‘‘entropy production’’ is not related to the production of a reference temperature. C1 and C2 are coefficients related
thermodynamic entropy expressed as the ratio of flux to to the universal constants in the empirical functions (a, b,
temperature that the term ‘‘MEP’’ may allude to. g 1 and g 2 in Table B1) representing the effect of the stability
on the mean profiles of wind speed and (potential)
3.2. Ground and Sensible Heat Fluxes Over Dry Soil temperature within the surface layer [Businger et al., 1971],
[18] Consider the energy budget over a dry nonvegetated  pffiffiffi
land surface. The major difference between the above toy 3=a; unstable
C1 ¼
model and the energy budget over a land surface is due to 2=ð1 þ 2aÞ; stable
the transport mechanism, i.e., heat transfer in the atmo- 
g 2 =2; unstable
sphere results from turbulent diffusion instead of conduction C2 ¼
2b; stable:
in solid media. Yet it remains true that the temperature is
continuous at the land-atmosphere interface. Since this I0 defined in equation (10),
property plays a key role in the MEP solution of the heat
fluxes in the toy model, the energy budget over a dry land
pffiffiffiffiffiffiffiffiffiffi 1

surface is expected to be solved in the same way using the kzg 6


I0 ¼ rCp C1 kz C2 ; ð11Þ
MEP framework. By analogy, the dissipation function of rCp T0
ground and sensible heat fluxes, G and H, is defined as
is referred to as the ‘‘apparent thermal inertia of the air,’’
2G2 2H 2 only depends on external parameters such as z and T0. Note
D þ ; ð9Þ that1 I0 does not have the same unit as Is due to the factor of
Is Ia
jHj6 in the expression of Ia that does have the same unit as
where Is is the thermal inertia for heat conduction in the soil that of Is. It is important to point out that parameterization of
as a composition of density, specific heat and thermal Ia based on equation (B11) by no mean implies linearity
diffusivity. Ia is the ‘‘thermal inertia’’ for turbulent heat between heat flux and temperature gradient, which is true
transfer in the air. Is is a well-defined physical property of only for conduction. Turbulent heat flux in general is a
the soil material, while Ia is yet to be defined. nonlinear function of the corresponding temperature
[19] Assuming Ia has the same composition as Is defined gradient, leading to a flux-dependent thermal inertia
in equation (A19), an eddy diffusivity must be parameter- parameter.
ized using a turbulent transfer model where an eddy [21] The dissipation function D with Ia parameterized in
diffusivity appears as the coefficient formally relating the terms of equation (10) is
turbulent heat flux to the gradient of mean (potential)
temperature. The most successful models of turbulent trans- 2G2 2H 2 1
D¼ þ jHj6 ; ð12Þ
port in the atmospheric boundary layer are those based on Is I0
Monin-Obukhov similarity theory (MOST) [e.g., Arya,
1988]. In the original MOST, two dimensionless equations which is not quadratic in H. Minimizing D under the
are established to relate the gradients of mean temperature constraint of conservation of energy at the land surface,
and wind velocity to the fluxes of heat and momentum
according to the Buckingham p theorem [Buckingham, G þ H ¼ Rn ; ð13Þ
1914]. Consequently, the eddy diffusivity must be expressed
in terms of two of the four state variables (i.e., the gradients of for a given net radiation Rn requires
mean wind velocity and temperature and the fluxes of heat
and momentum). Our goal is to reduce the degree of freedom 11 Is 1
G¼ H j H j6 : ð14Þ
from two to one so that Ia can be formulated as a function of 12 I0
the heat flux alone. This can be done by using an extremum
solution based on the MOST briefly described in Table B1 in Substituting equation (14) into equation (13) leads to a
Appendix B. The extremum solutions were derived by nonlinear algebraic equation for H,
removing the nonuniqueness in the relationships between
wind shear/temperature gradient and momentum/heat flux 11 Is 1

described by the similarity equations, equations (B1) and H j H j6 þH ¼ Rn : ð15Þ


12 I0
(B2). The extremum hypothesis leads to a third equation
linking wind shear, temperature gradient, momentum flux G and H according to equations (14) and (15) are the MEP
and heat flux, allowing any three of the four parameters to be solution of the energy budget over a dry land surface.
expressed in terms of the other [Wang and Bras, 2009].
4 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 2. Observed energy balance Rn = G + H for (a) Owens Lake and (b) Lucky Hill.

[22] The solution of H, hence G, is unique since equation the experiment site, instrument setup, sampling schemes,
(15) has only one real root for realistic values of the etc. have been reported in an earlier publication [Katul,
parameters (i.e., Is, I0 and Rn). G and H have the same sign 1994]. Thermal inertia of the soil material at this location
according to equation (14) following the usual sign con- was estimated as the (regression) coefficient in equation (8),
vention for G and H. As a special case, H = 0 and G = 0 Is ’ 0.83  103 J m2 K1 s1/2, using the independently
when Rn = 0. measured skin temperature and ground heat flux. I0, defined
[23] Equation (14) would be formally identical to equa- in equation (10), is computed as
tion (7) were not for the 11/12 factor. The difference is due 
to the fact that D defined in equation (12) is not a quadratic 0:35  103 ; unstable
function of G and H, while D in equation (6) for the toy I0 ’
0:23  103 ; stable:
model is a quadratic function of Fk. This numerical example
suggests that the orthogonality conditions given by Dewar where T0 = 300 K is used. It turns out that I0 is insensitive to
[2005], which hold exactly for near-equilibrium systems, T0 due to the 1/6 power dependence. For example, the
are indeed a good approximation for far from equilibrium extreme diurnal variation of temperature T0  280 – 330 K
systems. Equation (14) without the 11/12 coefficient is at the two sites causes no more than 3% change in I0, which
recognized as the formula obtained by Priestley [1959, p. is small enough to be negligible. The observed energy
105, equation (8.10)] based on the classical theory, balance at this site is shown in Figure 2a.
pffiffiffiffiffiffiffi   [26] Figures 3 and 4 compare the MEP solutions of G and
H rCp KH Ia H with the observed fluxes. The agreement is excellent. In
¼ pffiffiffiffiffi ¼ : ð16Þ
G rs cs ks Is Figure 3, the MEP G and H are computed with the observed
Rn as input to equation (15). In Figure 4, the MEP G and H
The MEP theory justifies Priestley’s result based on a more are computed with the observed total fluxes, G + H, to
fundamental principle except that Priestley’s equation is replace Rn in equations (15). Since the observed energy
difficult to apply without the parameterization of KH based balance is closed without significant biases (see Figure 2a),
on the newly derived extremum solution of MOST the agreement between the two MEP solutions (under
summarized in Table B1. The new parameterization of KH different inputs) are comparable to that between observed
based on the equations given in Table B1 is briefly Rn and G + H. In addition, the agreement between the MEP
described in the end of Appendix B. To our knowledge, predictions and the observed fluxes is as good during the
there has been no application of Priestley’s formula in land nighttime as during the daytime. Notice the observed G and
surface models. H (as well as Rn not shown here) cross zeroes at the same
times predicted by the MEP. The second test using the
3.3. Validation of the MEP Solution
Lucky Hill data demonstrates the behavior of the MEP
[24] The MEP solution of G and H in equations (14) and solution when the observed energy balance has some biases.
(15) will be compared with observations from two field 3.3.2. Lucky Hill
experiments at Owens Lake, California in 1993 and at [27] The site (31°44.5640N, 110°3.2510W) is located in
Lucky Hill near Tombstone, Arizona in 2008. The measured the Walnut Gulch watershed, Arizona. The field experiment
energy balance at the two sites is shown in Figure 2. was carried out 2– 17 June 2008 during the premonsoon
3.3.1. Owens Lake season with strong insolation and occasional clouds. An
[25] The data at this bare soil site was collected during 20 instrument tower was set up at a flat spot with sparse dry
June to 2 July 1993. Sensible heat flux was measured using shrubs. The topsoil layer of tens of centimeters was com-
an eddy covariance device, including a sonic anemometer pletely dry after an extended rain-free period. Ground heat
mounted at z = 2.5 m above the ground, colocated with flux was measured using heat flux plates placed  1 cm
other sensors to measure radiative fluxes, ground heat flux, below the surface. The observed energy balance at this site
soil skin temperature among other variables. Twenty minute is shown in Figure 2b. Skin temperature was measured
averaged variables are used in this study. More details about using an infrared thermometer mounted on the tower  1 m
5 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 3. (a) Time series of the MEP solution of G and H (dashed lines) versus the observed fluxes
(solid lines) with the observed Rn as the input to equation (15); MEP solution of (b) G and (c) H versus
observations. Twenty minute average data collected at Owens Lake, California, 20 June to 2 July 1993.

above the ground. Sensible heat fluxes were derived from the observed G (Figures 5a and 5b). According to Figure 2b,
10 Hz samples of a RM Young sonic anemometer and a the observed Rn is underestimated at high Rn, and over-
temperature sensor mounted at z = 4.3 m above the ground. estimated at low Rn relative to the observed G + H. Figure
Four components of radiative fluxes were measured using 5b shows that these biases in the observed energy budget
Epply Precision Spectral Pyranometers and Precision Infra- lead to the same biases in the MEP G relative to the
red Radiometers (Pyrgeometers). All variables were sam- observed G due to the energy balance constraint. Figure
pled at 0.1 Hz except those measured by the eddy 5b confirms G in phase with Rn predicted by the MEP (the
covariance device. Twenty minute averaged variables are scatterplot of the observed G versus Rn not shown). Figure
used below. The thermal inertia of the soil at this site has 5c shows the biases as well as phase differences between the
been estimated as Is ’ 0.83  103 J m2 K1 s1/2, in the MEP H and the observed H. The biases in the MEP H are
same way as that at Owens Lake. I0 is computed as due to the same reason for those in the MEP G. The phase

differences, which do not exist between the MEP and
0:50  103 ; unstable observed G, are caused by the phase differences between
I0 ’ the observed Rn and H (figure not shown) since the MEP
0:33  103 ; stable:
predicts H in phase with Rn (e.g., peaking at the same
[28] Figure 5 compares the MEP predictions with the times). This test demonstrates that the MEP is able to use
observed fluxes corresponding to the case of Figure 3. They the available information effectively to pick up true signals
agree qualitatively but have visible differences. The MEP G and highlight the inconsistency in the inputs.
has reduced diurnal amplitudes relative to, but in phase with [29] Using the observed total flux G + H input, the MEP
G and H are in much closer agreement with the observed G
6 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 4. (a) Time series of the MEP solution of G and H (dashed lines) versus the observed fluxes
(solid lines) with the observed total fluxes G + H as the input to equation (15) (to replace Rn); MEP
solution of (b) G and (c) H versus observations. Same data set as that of Figure 3.

and H as shown in Figure 6. The bias in G and phase where no field measurements of H (e.g., using eddy
difference in H are reduced according to Figures 6b and 6c, covariance device) are available to compare with. Note that
respectively. It appears that the MEP G has slight phase shift the MOST model for the surface layer is only valid for a
relative to the observed G. This is due to the phase limited range of z. z cannot be too small due to the
difference between the observed G and H (figure not assumption of the MOST that mean wind speed and
shown). The signal of relative phase in the input is pre- temperature profiles are not affected by the characteristics
served but divided between the MEP G and H. of the underlying surface such as roughness. z cannot be too
[30] The MEP solution of heat fluxes appears to be a large either because H at large z could be substantially
function of z even though the physical process of the different from H at the surface [e.g., Wang and Bras, 2001].
partition of radiative energy occurs at the land-atmosphere It would be ideal if z is set at the lower bound of the surface
interface. This is due to the parameterization of Ia (and H) layer, zs, where H is expected to be the closest to H at the
using the MOST model. In reality, H has always been surface (which has never been measured directly). The
measured at a certain distance above the ground in field theoretical value of zs is rather difficult to determine.
experiments. Since z has appreciable effect on the MEP Nonetheless, the good agreement between the MEP H with
solution due to the 2/3 power dependence of I0 on z, the the observed H indicates that zs at the two sites is close to
value of z must be appropriately chosen for the locations the measurement height of H. Our results suggest that,
7 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 5. (a) Time series of the MEP solution of G and H (dashed lines) versus the observed fluxes
(solid lines) with the observed Rn as the input to equation (15); MEP solution of (b) G and (c) H versus
observations. Twenty minute average data collected at Lucky Hill in Walnut Gulch watershed, Arizona,
2– 17 June 2008.

depending on the surface roughness, zs  2 – 3 m for flat FT (see equation (4)). The odds that the MEP prediction
nonvegetated surfaces, while zs  4 – 5 m for surfaces fails is estimated as 1 in 1017 for common situations, e.g.,
covered with sparse vegetation of 1 m in height. D  40 for G, H  100 and Is, Ia  103. Certainly, the
[31] The only meteorological input to the MEP model is performance of the MEP method depends on the input
net radiation. The model parameters are either physical (i.e., information. The power of the MEP is reflected in its
thermal inertia of the soil) or universal constants. The model capability, as an inference tool, of extracting the relevant
does not need calibration (i.e., no tuning parameters). No information about G and H given the net energy, Rn.
classical theories would give a unique solution of G and H [32] An intriguing aspect of the analysis is that the surface
(over a dry soil) based on conservation of energy only (i.e., temperature seems to be irrelevant since neither the dissi-
G + H = Rn for given Rn) since there are more unknowns pation function nor the test of the MEP solution needs the
than governing equations. The problem is underdetermined information of surface temperature. Yet, surface temperature
using conventional methods. This is a typical situation is expected to be an important diagnostic for the surface
where an inference is sought based on incomplete informa- energy balance as the emitted long-wave radiation, a com-
tion. In that sense, the MEP solution is a ‘‘guess,’’ arguably ponent of Rn, strongly depends on the skin temperature. In
the best one, instead of a guaranteed prediction. Our fact, the effect of the surface temperature on the MEP
confidence on the MEP solution may be measured by the solution of G and H is included through Rn even though a
8 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 6. (a) Time series of the MEP solution of G and H (dashed lines) versus the observed fluxes
(solid lines) with the observed total fluxes G + H as the input to equation (15) (to replace Rn); MEP
solution of (b) G and (c) H versus observations. Same data set as that of Figure 5.

given Rn does not uniquely determine the surface temper- maximized among all possible values allowed by the
ature. In addition, the surface heat fluxes are directly related available energy for the case of dry soil. The hypothesis
to space-time variations in the surface temperature instead of maximum H, or equivalently minimum G due to conser-
of temperature itself. Nonetheless, the surface temperature vation of energy H + G = Rn for a given Rn, can be tested
does contain essential information about the surface energy using the observations of heat fluxes and skin temperature.
budget that we turn our attention to next. [34] In the following analysis, G = G(Ts) is a functional of
the skin temperature time history [Wang and Bras, 1999],
4. A Stationary Hypothesis of Energy Balance where Ts is understood as the skin temperature at current
time while the skin temperature at earlier times is treated as
[33] Many nonequilibrium systems can be understood given parameters. The necessary condition for H to be
and described by extremum principles [Weinstock, 1952; extremum is that all partial derivatives of the Lagrangian
Sieniutycz and Salamon, 1990]. The idea that nature takes function,
an optimal path (or fastest approach) to reach an equilibrium
state under certain constraints has led the hypotheses of fh  H þ lh ðRn  H  GÞ; ð17Þ
maximum evaporation and transpiration [Wang et al., 2004,
2007] governing the energy budget over the land surface. It
with respect to the state variables Ts, H and the Lagrangian
has also been suggested that turbulent sensible heat flux is
multiplier lh vanish,
the next most effective (to evaporation) heat removal
mechanism [Kim and Entekhabi, 1998]. If that is true,
@fh @fh @fh
which as we show below it is not, H is expected to be ¼ 0; ¼ 0; ¼ 0; ð18Þ
@Ts @H @lh

9 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 7. Validation of equation (19) using the observed Ts, G and Rn measured at the Owens Lake site.

for a given Rn. The choice of H as a state variable explicitly cause large numerical errors in estimating the partial deriv-
leads to a partition of net radiation as a result of the atives. Second, the nonzero derivatives may result from the
extremization procedure without specification of arbitrary violation of energy conservation as the observed energy
models for turbulent heat transfer. Heat fluxes are fluxes tend to be more out of balance, especially at the
commonly used as independent state variables in the Lucky Hill site (see Figure 2), at very high and very low Rn.
nonequilibrium thermodynamics [Sieniutycz and Salamon, Hence we view the results as solid evidence supporting the
1990, p. 3]. Equations (17) and (18) yield the necessary hypothesis that H and G reach stationary saddle point in
condition of a potential extremum H, balancing net radiation.
 Rn [36] The hypothesis, if further confirmed to be true, is a
@G general governing principle for the energy budget over a dry
¼ 0; ð19Þ
@Ts soil since the test of the hypothesis is (turbulent transport)
model independent. Note that equation (19) holds for the
lh ¼ 1: ð20Þ case considered in section 3.1 where an analytical solution
The superscript Rn in equation (19) emphasizes that the of the heat fluxes exists (see Appendix A). Temperature
derivative is evaluated under the condition of fixed Rn. The independent F1 and F2 in equations (A25) and (A26)
corresponding Hessian matrix Hfh (see Appendix C) has guarantees vanishing derivatives for prescribed F0. But
three (nonzero) eigenvalues, neither can be maximum or minimum because F1 and F2
 2 0
are symmetric. Then the only possibility is that the station-
@ G ary point is a saddle. It would not be surprising that the
 ; 1; 1; conclusion for F1 and F2 would be also valid for G and H if
@Ts2
one thinks of heat conduction as a limiting case of turbulent
where the superscript ‘‘0’’ indicates the derivatives are diffusion.
defined at the stationary point of fh under the constraint [37] The above analysis indicates that the energy budget
G(Ts) + H = Rn. The eigenvalues do not have the same sign over a dry soil depends on change of temperature instead of
since the curvature of G function is not zero in general. absolute level of temperature. This is not a new finding as
Then the stationary point of H is a saddle point instead of an all existing models of G and H [e.g., Sellers, 1965; Wang
extremum due to the indefiniteness of Hfh. If equation (19) and Bras, 1999; Monin and Obukhov, 1954; Tillman, 1972;
holds, neither is H maximum, nor is G minimum due to the Katul et al., 1995] use either spatial or temporal variations
constraint G + H = Rn. Validation of equation (19) implies of temperature. The MEP solution of heat fluxes is consis-
that Rn is partitioned in such as way that both fluxes reach a tent with this property of energy budget over dry soil: the
stationary saddle point. surface heat fluxes are not informative about the surface
[35] The estimated @G/@Ts terms using the observed Ts, G temperature, at least not directly, which gives a physical
and Rn for the Owens Lake and the Lucky Hill site are meaning to equation (19).
shown in Figures 7 and 8, respectively. They are close to [38] The stationary hypothesis and the MEP model are
zero, meaning that the values are small compared to the intended to tell a full story about the partition of net
ratio of the amplitude of variation in G to that of Ts, 10 W radiation into surface heat fluxes over a dry soil seen from
m2 K1. At both sites, the observed G ranges from 100 complementary point of view. The stationary hypothesis
to 500 W m2 and Ts from 10 to 60 C (Ts data not shown). describes a certain aspect of the process that the MEP model
The estimated derivatives only significantly depart from does not, and vice versa. The MEP model of surface heat
zero value at high and low levels of Rn. There are two fluxes is formulated using an epistemological argument
plausible explanations for this departure. First, there are based on the Bayesian probability theory. The result is a
fewer data points at high and low values of Rn that may useful algorithm, as a statistical inference, for computing the
10 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Figure 8. Validation of equation (19) using the observed Ts, G and Rn measured at the Lucky Hill site.

surface heat fluxes using the available information. The only needs net radiation input without requiring other
stationary hypothesis of surface heat flux is developed using meteorological variables such as temperature and wind
an ontological argument based on our experience about the speed. The MEP method will not replace the existing ones,
physical world. The result is a rationale, as a fundamental instead it offers an alternative method that would enhance
principle, for explaining the underlying physical mecha- the models of land surface energy balance.
nisms leading to the observed the surface heat fluxes. [40] The significance of the proposed MEP model is that
Together they not only shed new light on the process of the concept of entropy production is not limited to the
surface energy balance but also predict the energy budget thermodynamic entropy production. In general, the MaxEnt
with information considered to be incomplete from tradi- and MEP formalisms provide a tool through extremization
tional viewpoint. The MEP hypothesis has little direct of ‘‘entropy production’’ to make predictions (as statistical
bearing on the stationary hypothesis except that the latter inference). Depending on the quantities of interest, the
may explain why the MEP model of heat fluxes does not ‘‘entropy (production)’’ defined in the MaxEnt and MEP
need temperature input. Logically, the stationary hypothesis formalisms is not always related to the thermodynamic
is completely independent of the MEP model of heat fluxes entropy. This study presents an example where the ‘‘entropy
in the sense that (1) testing the stationary hypothesis does production’’ is different from both thermodynamic entropy
not require models of sensible and ground heat flux, which production and (thermal) energy dissipation that are often
makes the hypothesis most general; (2) the stationary used in the models of nonequilibrium systems. We expect
hypothesis does not lead to models of sensible and ground the MEP method to be applicable to the energy balance over
heat flux either (at least not yet). Nonetheless, the stationary a wet or a vegetated land surface where the challenge is to
hypothesis supported by the observational evidence (i.e., formulate the entropy production or dissipation function
@G/@Ts = 0) conditioned on conservation of energy (i.e., including the latent heat flux term; an ongoing research
G + H = Rn) is consistent with the MEP model that the subject. This study offers new possibilities of improving
solution of surface heat fluxes does not depend on Ts, at hydrology models using the MEP theory. The encouraging
least not directly. results offer genuine hope that the next generation of land
surface models will be capable of capturing all features of
5. Conclusions the surface energy balance under all conditions.
[39] The proposed model of the surface heat fluxes based
on the principle of maximum entropy production and the Appendix A: Analytical Solution of Heat
stationary hypothesis of the energy balance are two com- Conduction
plementary components of a new view of the heat exchange [41] The heat transfer in the two semi-infinite columns in
at the land-atmosphere interface. The stationary hypothesis contact at z = 0 are described by
adds another example of natural phenomena understood
from the perspective of an optimality argument. The @T1 @ 2 T1
hypothesis was tested under the most general conditions r1 c1 ¼ m1 2 ; 0<z<1 ðA1Þ
@t @z
even though it does not lead to a model of the heat fluxes.
That task is fulfilled by the MEP theory that allows the heat
fluxes to be parameterized once transport models are @T2 @ 2 T2
selected. The MEP method for modeling and estimating r2 c2 ¼ m2 2 ; 1 < z < 0 ðA2Þ
@t @z
surface heat fluxes complements the existing methods as it
11 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

where T is temperature and r, c and m are constant density, where C1(s) and D1(s) are two arbitrary functions. Zero flux
specific heat and heat conductivity associated with self- boundary condition equation (A4) requires C1(s) = 0.
explanatory indices. t and z carry their conventional Therefore,
meaning. The prescribed initial and boundary conditions are
 rffiffiffiffiffi 
s
T^1 ðs; zÞ ¼ D1 ðsÞexp  z ; ðA15Þ
T1 ð0; zÞ ¼ T2 ð0; zÞ ¼ T0 ; t > 0; ðA3Þ k1

where D1(s) is yet to be determined. Similarly, L transform


@T1 of equation (A2) with the initial condition equation (A5)
¼ 0; z!1 ðA4Þ
@z leads to
rffiffiffiffiffi 
s
@T2 T^2 ðs; zÞ ¼ C2 ðsÞexp z ; ðA16Þ
¼ 0; z ! 1 ðA5Þ k2
@z
where k1 = m2/(r2c2) and C2(s) is an arbitrary function to be
T1 ðt; 0Þ ¼ T2 ðt; 0Þ; t > 0: ðA6Þ determined.
[44] Continuous temperature at the interface z = 0 through
where T0 is a constant. A second boundary condition at z = equation (A6) requires
0 must be specified for solving T1 and T2. Assume a source
of heat is located at z = 0 (the contact point of the two D1 ðsÞ ¼ C2 ðsÞ ¼ AðsÞ: ðA17Þ
columns) with a given heat flux input, F0, partitioned into
F1 and F2,
Substituting equations (A15) – (A17) into L-transformed
@T1 ðt; 0Þ equations (A9), we obtain
F 1 ¼ m1 ; ðA7Þ
@z
1 ^0
F
@T2 ðt; 0Þ AðsÞ ¼ pffiffi ðA18Þ
F 2 ¼ m2 ; ðA8Þ I1 þ I2 s
@z
^ 0 is the L transform of F0 with thermal inertia
where F
satisfying conservation of energy, parameters defined as
F1 þ F2 ¼ F0 ðt Þ; ðA9Þ pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffi
Ij  rj cj mj ¼ rj cj kj ; j ¼ 1; 2: ðA19Þ
where appropriate sign is understood. F1 and F2 are
unknowns to be determined from equations (A1) – (A9). Then,
[42] Solutions of equations (A1) and (A2) under initial
 rffiffiffiffiffi 
and boundary conditions equations (A3) – (A9) may be 1 F ^ 0 ðsÞ s
deduced by means of Laplace transform (L transform) with T^1 ðs; zÞ ¼ pffiffi exp  z ; 0<z<1 ðA20Þ
I1 þ I2 s k1
respect to time t. Denote
T^1 ðs; zÞ ¼ Lt;s fT1 ðt; zÞg; ðA10Þ rffiffiffiffiffi 
1 F ^ 0 ðsÞ s
T^2 ðs; zÞ ¼ pffiffi exp z ; 1 < z < 0: ðA21Þ
I1 þ I2 s k2
T^2 ðs; zÞ ¼ Lt;s fT2 ðt; zÞg; ðA11Þ
[45] Inverse L transform of equations (A20) and (A21)
where L is the Laplace transform operator, gives the solution of temperature at the interface z = 0,
Z1
Zt
T^ ðs; zÞ ¼ est ½T ðt; zÞ  T0 dt: ðA12Þ 1 1 F0 ðt Þdt
T1 ðt; 0Þ ¼ T2 ðt; 0Þ ¼ T0 þ pffiffiffi pffiffiffiffiffiffiffiffiffiffi : ðA22Þ
0 I1 þ I2 p tt
0
[43] L transform of equation (A1) with the initial condi-
tion equation (A3) leads to [46] According to equations (A7) and (A8), L transform
of F1 and F2 are
@ 2 T^1 ^
sT^1 ¼ k1
@z2
; ðA13Þ ^1 ¼ m1 @ T1 ðt; 0Þ ;
F ðA23Þ
@z
where k1 = m1/(r1c1). Equation (A13) has a general
^
solution, ^2 ¼ m2 @ T2 ðt; 0Þ :
F ðA24Þ
@z
rffiffiffiffiffi   rffiffiffiffiffi 
s s
T^1 ðs; zÞ ¼ C1 ðsÞexp z þ D1 ðsÞexp  z ; ðA14Þ Then, inverse L transform of equations (A23) and (A24)
k1 k1
where @ T^ 1(t, 0)/@z and @ T^ 2(t, 0)/@z are obtained using

12 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Table B1. Summary of the Extremum Solution Based on forms of fm and fh have been suggested by other authors
Monin-Obukhov Similarity Theorya [e.g., Beljaars and Holtslag, 1991], but they do not differ
qualitatively from those of Businger et al. According to
Stable Unstable
  u  Businger et al.,
T0 u* 2
Qz (a + 12)2
1
g z  p2ffiffi3 2 Tg0 *
H
z z
Cp u3 u3 fm ¼ 1 þ b ; ðB6Þ
1 T0 * 2 T0 * L
 2 g z 2 g z
a
The constants are taken as a  0.75 or 1, b  4.7, g 1  15, g 2  9.
z
fh ¼ a þ b ; ðB7Þ
L

equations (A20) and (A21) by differentiating with respect to under stable conditions, L > 0; and
z first and then letting z = 0 leads to the desired solution,
 z 14
@T1 ðt; 0Þ I1 fm ¼ 1  g 1 ; ðB8Þ
F 1  m1 ¼ F0 ; ðA25Þ L
@z I1 þ I2
 z 12
fh ¼ a 1  g 2 ; ðB9Þ
@T2 ðt; 0Þ I2 L
F 2  m2 ¼ F0 : ðA26Þ
@z I1 þ I2
under unstable conditions, L < 0. The empirical constants in
the above expressions are estimated as

Appendix B: An Extremum Solution of Surface a  0:75; b  4:7; g 1  15; g 2  9:


Layer Turbulence
[47] According to Monin and Obukhov [1954], mean [49] It can be shown that equations (B1) and (B2) do
wind shear Uz and temperature gradient Qz within a not always lead to unique relationships between Uz, Qz, u*
stationary and homogeneous surface layer can be described and H. To remove the nonuniqueness, we seek those
by the following equations: among all possible solutions allowed by equations (B1)
kz z and (B2) that satisfy the requirement that ‘‘momentum
U z ¼ fm ; ðB1Þ flux always reaches such values that heat flux and wind
u* L
shear are minimized under stable condition; and that
kz z heat flux and temperature gradient are minimized under
Qz ¼ fh ; ðB2Þ unstable condition.’’ Then we obtain expressions of Qz
q* L
and H in terms of u* shown in Table B1, called the
extremum solution based on Monin-Obukhov similarity
where z is the vertical coordinate (i.e., distance from the theory (MOST).
surface), k is the von Karman constant. Velocity scale u* [50] By combining the extremum solutions of Qz and H
(friction velocity) and temperature scale q* are defined in terms of u* given in Table B1, H can be formally
through shear stress t and heat flux H as expressed in the form
t
¼ u*2 ; ðB3Þ H ¼ rCp KH Qz ; ðB10Þ
r

with the eddy-diffusivity KH defined as


H
¼ u* q* ; ðB4Þ
rCp KH  C1 kzu* ; ðB11Þ

where r is the (constant) air density, Cp is the heat capacity


of the air at constant pressure. L is the Obukhov length, where C1 is defined in section 3.2. The ‘‘thermal inertia’’ for
the air, Ia, may be defined formally,
 1 pffiffiffiffiffiffiffi
kg H Ia  rCp KH ;
L ¼ u3* ; ðB5Þ
T0 rCp pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðB12Þ
¼ rCp C1 kzu* :

where g is the gravitational acceleration, and T0 is a


representative temperature.
[48] fm and fh in equations (B1) and (B2) are empirical
functions introduced by Monin and Obukhov to represent Appendix C: Sufficient Conditions for Extremum
the effect of the stability on the mean profiles of wind speed [51] Positivity of the Hessian matrix of a multivariate
and temperature. The most popular functional forms of fm function f(x1, x2,. . ., xn) evaluated at the stationary point
and fh are proposed by Businger et al. [1971]. Alternative x0 = (x01, x02,. . ., x0n) determines whether f(~
~ x0) is an
13 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

extremum [Dineen, 2001]. The Hessian matrix of f is Desborough, C. E., A. J. Pitman, and P. Irannejad (1996), Analysis of the
relationship between bare soil evaporation and soil moisture simulated by
defined as 13 land surface schemes for a simple non-vegetated site, Global Planet.
Change, 13, 47 – 56.
0 1 Dewar, R. C. (2003), Information theory explanation of the fluctuation
@2f @2f @2f theorem, maximum entropy production and self-organized criticality
B @x2 ; @x1 @x2
; :::;
@x1 @xn C
B 1 C in non-equilibrium stationary states, J. Phys. A Math. Gen., 36,
B @2f @2f @2f C 631 – 641.
B ; ; :::; C
Hf ð~
xÞ ¼ B C; ðC1Þ Dewar, R. C. (2005), Maximum entropy production and the fluctuation
B @x2 @x1 @x22 :::
@x2 @xn C
B 2 C theorem, J. Phys. A Math. Gen., 38, L371 – L381.
@ @ f @2f @2f A Dineen, S. (2001), Multivariate Calculus and Geometry, 254 pp., Springer,
; ; :::; New York.
@xn @x1 @xn @x2 @x2n
Gregory, P. C. (2005), Bayesian Logical Data Analysis for the Physical
Sciences, 468 pp., Cambridge Univ. Press, New York.
Grinstein, G., and R. Linsker (2007), Comments on a derivation and
and the stationary point of f, ~
x0, is determined by application of the ‘maximum entropy production’ principle, J. Phys. A
Math. Theor., 40, 9717 – 9720.
Henderson-Sellers, A., P. Irannejad, K. McGuffie, and A. J. Pitman (2003),
rf ð~
x0 Þ ¼ 0; ðC2Þ Predicting land-surface climates—Better skill or moving targets?, Geo-
phys. Res. Lett., 30(14), 1777, doi:10.1029/2003GL017387.
Jaynes, E. T. (1957), Information theory and statistics mechanics, Phys.
where r is the gradient operator. The criteria for extremum Rev., 106, 620 – 630.
x0) are
f(~ Jaynes, E. T. (1980), The minimum entropy production principle, Annu.
Rev. Phys. Chem., 31, 579 – 601.
[52] 1. f(~x0) is a local minimum (maximum) if Hf is Jaynes, E. T., and G. L. Bretthorst (Eds.) (2003), Probability Theory—The
positive (negative) definite at ~
x0, Logic of Science, 755 pp., Cambridge Univ. Press, New York.
[53] 2. Hf is positive (negative) definite if and only if all Juretic, D., and P. Zupanovic (2003), Photosynthetic models with maximum
eigen values of Hf are positive (negative), entropy production in irreversible charge transfer steps, Comput. Biol.
Chem., 27, 541 – 553.
x0) is a saddle point if Hf is indefinite, i.e., its
[54] 3. f(~ Katul, G. G. (1994), A model for sensible heat flux probability density
eigen values change signs. function for near-neutral and slightly-stable atmospheric flows, Boundary
[55] The Hessian matrix of fh in equation (17) is Layer Meteorol., 71, 1 – 20.
Katul, G. G., S. T. Goltz, C.-I. Hsieh, Y. Cheng, F. Mowry, and
0 1 J. Sigmon (1995), Estimation of surface heat and momentum fluxes
@2G @G using the flux-variance method above uniform and non-uniform terrain,
B lh @T 2 ; 0; 
@Ts C Boundary Layer Meteorol., 74, 237 – 260.
B s C
Hfh ðTs ; H; lh Þ ¼ B
B 0; 0; 1 CC: ðC3Þ Kim, C. P., and D. Entekhabi (1998), Feedbacks in the land-surface and
@ @G A mixed-layer energy budgets, Boundary Layer Meteorol., 88, 1 – 21.
 ; 1; 0 Kleidon, A., and K. Fraedrich (2004), Biotic entropy production and
@Ts global atmosphere-biosphere interaction, in Non-equilibrium Thermo-
dynamics and the Production of Entropy, pp. 173 – 189, Springer,
At the stationary point, Heidelberg, Germany.
Kleidon, A., and S. Schymanski (2008), Thermodynamics and optimality of
 0 the water budget on land: A review, Geophys. Res. Lett., 35, L20404,
@G doi:10.1029/2008GL035393.
l0h ¼ 1; ¼ 0; ðC4Þ Kondepudi, D., and I. Prigogine (1998), Modern Thermodynamics, 486 pp.,
@Ts John Wiley, New York.
Monin, A. S., and A. M. Obukhov (1954), Basic turbulence mixing laws in
then Hfh has three eigenvalues, the atmospheric surface layer, Tr. Inst. Teor. Geofiz. Akad. SSSR, 24(151),
163 – 187. (English translation, pp. 164 – 194 in Turbulence and Atmo-
 2 0 spheric Dynamics, edited by J. L. Lumley, translated by V. N. Bespalyi,
@ G Cent. for Turbul. Res., Palo Alto, Calif., 2001.)
 ; 1; 1:
@Ts2 Ozawa, H., A. Ohmura, R. D. Lorenz, and T. Pujol (2003), The second law
of thermodynamics and the global climate system: A review of the max-
imum entropy production principle, Rev. Geophys., 41(4), 1018,
doi:10.1029/2002RG000113.
[56] Acknowledgments. This work was supported by ARO under Paltridge, G. W. (1975), Global dynamics and climate—A system of mini-
project W911NF-07-1-0126 and NSF under grant EAR-0309594. We are mum entropy exchange, Q. J. R. Meteorol. Soc., 101, 475 – 484.
grateful to Gabriel Katul of Duke University for providing the Owens Lake Priestley, C. H. B. (1959), Turbulent Transfer in the Lower Atmosphere,
data used in this study. We thank David Goodrich and John Smith of 130 pp., Univ. of Chicago Press, Chicago, Ill.
USDA-ARS for their support during the field experiment involving Sellers, W. D. (1965), Physical Climatology, 272 pp., Univ. of Chicago
graduate students at MIT, Ryan Knox and Gajan Sivandran. Press, Chicago, Ill.
Shannon, C. E. (1948), A mathematical theory of communication, Bell Syst.
Tech. J., 27, 379 – 423.
References Sieniutycz, S., and P. Salamon (1990), Non-equilibrium Theory and Extre-
Arya, S. P. (1988), Introduction to Micrometeorology, 307 pp., Academic, mum Principles, 547 pp., Taylor and Francis, New York.
New York. Tillman, J. E. (1972), The indirect determination of stability, heat, and
Beljaars, A. C. M., and A. A. M. Holtslag (1991), Flux parameterization momentum fluxes in the atmospheric boundary layer from simple
over land surfaces for atmospheric models, J. Appl. Meteorol., 30, 327 – scalar variables during dry unstable conditions, J. Appl. Meteorol., 11,
341. 783 – 792.
Buckingham, E. (1914), On physically similar systems; illustrations of the Tribus, M. (1961), Thermostatics and Thermodynamics—An Introduction
use of dimensional equations, Phys. Rev., 4, 345 – 376. to Energy, Information and States of Matters, With Engineering Applica-
Businger, J. A., J. C. Wyngaard, Y. Izumi, and E. F. Bradle (1971), Flux- tions, 649 pp., Van Nostrand Reinhold, Princeton, N. J.
profile relationships in the atmospheric surface layer, J. Atmos. Sci., 28, Wang, J., and R. L. Bras (2001), Effect of temperature on surface energy
181 – 189. balance, Water Resour. Res., 37(12), 3383 – 3386.
de Groot, S. R., and P. Mazur (1984), Non-equilibrium Thermodynamics, Wang, J., and R. L. Bras (1999), Ground heat flux estimated from surface
510 pp., Dover, New York. soil temperature, J. Hydrol., 216, 214 – 226.

14 of 15
W11422 WANG AND BRAS: MEP AND SURFACE HEAT FLUXES W11422

Wang, J., and R. L. Bras (2009), An extremum solution of the Monin- Weinstock, R. (1952), Calculus of Variations With Applications to Physics
Obukhov similarity equations, J. Atmos. Sci., in press. and Engineering, 326 pp., McGraw-Hill, New York.
Wang, J., G. D. Salvucci, and R. L. Bras (2004), An extremum principle of
evaporation, Water Resour. Res., 40, W09303, doi:10.1029/
2004WR003087. 

Wang, J., R. L. Bras, M. Lerdau, and G. D. Salvucci (2007), A maximum R. L. Bras and J. Wang, Department of Civil and Environmental
hypothesis of transpiration, J. Geophys. Res., 112, G03010, doi:10.1029/ Engineering, University of California, Irvine, CA 92697, USA. (jingfenw@
2006JG000255. uci.edu)

15 of 15

You might also like