You are on page 1of 12

VOL. 14, NO.

12

JOURNAL OF CLIMATE

15 JUNE 2001

The Estimation of Regional Precipitation Recycling. Part I: Review of Recycling Models


G. I. BURDE

AND

A. ZANGVIL

The Jacob Blaustein Institute for Desert Research, Ben-Gurion University of the Negev, Sede Boqer Campus, Israel
(Manuscript received 27 September 1999, in final form 27 September 2000)
ABSTRACT
This review preceding a description of a new recycling model (Part II of this paper) discusses one aspect of
the regional precipitation recycling studies, namely, the mathematical modeling of the recycling process. Several
recycling models developed in recent decades are discussed within a unified framework of equations of the
conservation of atmospheric water vapor mass. Most of the recycling models may be considered as modifications
and generalizations of Budykos model. In spite of different forms, into which authors present their models,
every recycling model is based on two equations of conservation of mass for two fractions of precipitable water
in the atmospherewritten under some common assumptionsand the equation expressing the condition that
these two fractions are well mixed. If any of these three equations is missing the model is incomplete and its
results inconsistent. Different models differ by additional assumptions made in order to simplify either the
equations or their solution or to circumvent a solution.
The unified approach permits the showing of the principles and assumptions that are common to all the models
and identify additional assumptions made by a specific model and related limitations. In addition, it makes the
derivation of the results more straightforward and frequently less complicated than in the original works.

1. Introduction
Precipitation over a land region is composed of two
components: the advective component resulting from
the flux of external water vapor and the internal component resulting from the flux of water vapor from local
evaporation. Precipitation recycling is defined as the
contribution of the locally evaporated water to the precipitating water in the same region. Studying precipitation recycling over land areas provides useful information on the possible interactions of hydrology and
climate. The degree of precipitation recycling for a land
region determines the role for land surface hydrology
in the regional climate, on the one hand, and the role
of climate in the formation of surface and subsurface
water resources, on the other hand. The recycling rate
is a measure of soil moistureprecipitation feedback effects that are of particular importance for the soilatmosphere system.
Our view on the problem of precipitation recycling
changed dramatically during the last decades (see Brubaker et al. 1993 and Eltahir and Bras 1996 for reviews).
The theory, that the contribution of evaporation from a
land region to precipitation on the same region is very
significant, widely accepted until the late 1930s and
supported by some estimates, was later criticized and
Corresponding author address: Dr. G. I. Burde, The Jacob Blaustein Institute for Desert Research, Ben-Gurion University of the
Negev, Sede Boqer Campus, 84990 Israel.
E-mail: georg@bgumail.bgu.ac.il

q 2001 American Meteorological Society

different estimates supporting the alternative view were


presented (e.g., Sellers 1965; Budyko 1974; Brubaker
et al. 1993; Eltahir and Bras 1994; Trenberth 1998; see
also Eltahir and Bras 1996 for a summary).
Estimates of precipitation recycling, even though they
were based on observational data, inevitably represented
indirect estimates since no possibility existed to separately observe water molecules of local origin and those
of external origin in precipitating water. Another approach is the use of stable isotopes of water (e.g., Salati
et al. 1979). There are, however, some difficulties in
interpreting isotope analysis for water recycling studies
because the isotope data reflect a number of processes
simultaneously. Thus, estimation of precipitation recycling necessarily involved some underlying assumptions
and limitationsoften referred to as a recycling model.
The most common method of estimating precipitation
recycling over large regions is that developed by Budyko and Drozdov (1953) and described in Budyko
(1974)Budykos model. This model yields a simple
formula expressing the recycling ratio as a function of
regional evaporation, inflow of atmospheric moisture,
and the linear scale of the region measured along streamlines. Almost all the aforementioned estimates of regional precipitation recycling were made with the use
of Budykos formula.
Numerical atmospheric models were utilized to estimate the continental-scale water cycle and, in particular, to analyze the precipitation response to soil-mois-

2497

2498

JOURNAL OF CLIMATE

ture conditions. One of the first experiments of this type


with a general circulation model (GCM) was that of
Shukla and Mintz (1982) who detected a strong sensitivity of the precipitation in response to change in surface evaporation. Similar experiments have been performed with a wide range of numerical modelsboth
GCM and regional climate models (e.g., Rowntree and
Bolton 1983; Yeh et al. 1984; Oglesby and Erickson
1989; Shukla et al. 1990; Beljaars et al. 1996; Schar et
al. 1999). Most of these studies have documented the
important role of land evaporation in rainfall on continental spatial scales.
An alternative, more direct way to estimate precipitation recycling with numerical models is to incorporate
tagged water experiments into GCM simulations to trace
the origin and transport of water substances. It was initiated by Joussaume et al. (1984) who used a GCM to
simulate the behavior of isotopes of water, but they did
not sufficiently discuss the relationships with the water
cycle. Later, Koster et al. (1986) and Joussaume et al.
(1986) attempted to estimate sources of precipitating
water by using a GCM. A similar methodology (isotope
modeling or tagged water experiments with GCMs) was
used in a range of subsequent studies (e.g., see Druyan
and Koster 1989; Jouzel et al. 1997; Numaguti 1999).
Dirmeyer and Brubaker (1999) applied a quasi-isentropic back-trajectory algorithm to trace the transport and
estimate the surface sources of moisture supplying rainfall. Generally, the results indicated the importance of
the recycling process and yielded higher values of the
recycling ratio than earlier estimations.
The discrepancies may be related to an inaccurate
description of the precipitation recycling process by Budykos model, which was used in earlier estimations.
Note, however, that numerical results could also suffer
from incompleteness of the model climatology and parameterization schemes for soil, vegetation, radiation,
and convective processes. Several modifications and
generalizations of Budykos model have been proposed.
However, every model incorporated only part of the
effects neglected in Budykos model so that only part
of Budykos model shortcomings have been overcome,
and in some cases modifications introduce new additional assumptions and inconsistencies. Thus, there is
plenty of room for improvement in the problem of constructing an adequate recycling model.
The development of improved recycling models is
important both for observational studies and for numerical simulations of atmospheric and hydrological
processes. Studies with regional climate models, despite
the advantage of a more detailed surface physics incorporated, have to use a recycling model to estimate
the degree of regional precipitation recycling (e.g.,
Schar et al. 1999). Also studies incorporating tagged
water experiments into GCM simulations seem to feel
a necessity to compare their results with estimates by
an analytical formula even if it provides a lumped estimate as Budykos formula does (e.g., Numaguti 1999).

VOLUME 14

An analytical formula incorporating in a simple but adequate way the land surfaceatmosphere feedback effects through the precipitation recycling process could
be used in the physically based models of large-scale
moisture variability (e.g., Rodriguez-Iturbe et al. 1991;
Entekhabi et al. 1992). As Eltahir and Bras (1996) summarized in their review of the precipitation recycling
problem:
. . . we need to focus on developing of new analytical
formulae for estimating precipitation recycling.

In Part II of this paper we present a new recycling


model, which may be thought of as a generalization of
the existing models. Involving no additional assumptions it incorporates the effects of inhomogeneity of the
fluxes and the nonparallel flow effects and, at the same
time, it retains the advantage of an analytical presentation of the results. Here, in Part I, we present a discussion of the existing recycling models within a unified
framework of the equations of conservation of atmospheric water vapor mass. Such a review does not exist
in the literature. The recent review of the precipitation
recycling problem by Eltahir and Bras (1996) does not
include a review of recycling models; the authors restricted themselves to description of their recycling
model and Budykos model. A need in such a review
may be justified from several points of view. The review
preceding a description of a new recycling model should
show precisely its place among the existing models.
However, in our opinion, there is a general need in such
a review since in many cases references to existing recycling models contain unclear, and even wrong, information, and underlying principles and assumptions and
the limits of applicability of a model are incorrectly
assessed. The unified approach permits us to show the
principles and assumptions that are common to all models and identify additional assumptions made by a specific model and related limitations. In addition, it makes
derivation of the results more straightforward and frequently less complicated than in the original works.
2. Basic definitions, assumptions, and equations
a. Basic definitions
Consider an atmospheric control volume above the
land region of interest into which water vapor is brought
by air currents with a moisture influx F 1 through the
sides of the volume. The water vapor content w in the
air, moving across the region with a horizontal velocity
V, varies within the region decreasing due to precipitation with a vertical flux P and increasing due to evaporation with a vertical flux E. The water vapor content
w is the vertically integrated water vapor depth (precipitable water) represented by the vertical integral of
specific humidity q( p) from the surface to an elevation
where pressure p vanish, as follows

15 JUNE 2001

w5

1
rL g

q( p) dp,

(2.1)

p0

where r L is the liquid water density and g is the acceleration due to gravity. The velocity V 5 (u, y ) is defined
through the vertically integrated water vapor flux F 5
[F (x) , F ( y ) ], as follows
u5

y 5

2499

BURDE AND ZANGVIL

F (x)
,
w

F (x) 5

F (y)
,
w

F (y) 5

1
rL g
1
rL g

E
E

q( p)u ( p) dp,

p0
0

q( p)y ( p) dp,

(2.2)

p0

where u ( p) and y ( p) are the wind components in the x


and y directions, respectively.
The water vapor content is composed of an advective
portion w a and an evaporative portion w m :
w 5 wa 1 wm ,

(2.3)

and, correspondingly, the precipitation P is composed


of the parts P a and P m of advective and local (evaporative) origins:
P 5 Pa 1 Pm .

(2.4)

The problem consists in estimating the relative contributions of advective moisture and local evaporation to
precipitation for a given domain.
This relation may be characterized by the precipitation recycling ratio representing the fraction of precipitation due to local evaporative origin. The precipitation
recycling ratio for the total land region r is defined as
the ratio of total precipitation derived from evaporation
to the total area precipitation:

r5

E
E

Pm (x, y) dA

(2.5a)

P(x, y) dA

where A is the area of the region and the areal integrals


are taken to find the total area precipitation. The same
may be expressed as the ratio of areal-averages
r5

Pm
.
P

w (x, y) dA

Pm (x, y)
,
P(x, y)

(2.8)

which is a result of the passage to the limit

r (x, y) 5 lim

E
E

Pm (x, y) dA

DA

D A0

(2.9)

P(x, y) dA

DA

where DA is a small area reduced to the point (x, y) in


the limit.
It should be emphasized that in this definition P m (x,
y) is the contribution of evaporation from the total area
of the domain to precipitation at this specific point. In
other words, P m (x, y) is the water molecules that appeared in the atmosphere because of an evaporation
event from any point within the region, stayed in the
atmosphere for some time, and precipitated from the
atmosphere on the area DA including the point (x, y).
This definition should not be confused with the contribution of evaporation within the small area DA to precipitation in that same area, which evidently tends to
zero when the area is reduced to a point.
Having the local recycling ratio distribution r (x, y)
determined, the recycling ratio r for the total region may
be found from the relation

r5

r (x, y)P(x, y) dA

(2.10)

P(x, y) dA

(2.6)

in which w(x, y) is any variable.


Another characteristic, which is frequently used, is
Budykos recycling coefficient b representing the ratio
of total to advected average precipitation:
P
b5 .
Pa

r (x, y) 5

(2.5b)

Here and everywhere overbars denote horizontal averaging over the region, defined as
1
w 5
A

The recycling characteristics defined in such a way


depend on the size of the area considered. The recycling
ratio defined by (2.5) ranges from 0 to 1 being small
for small spatial scales and increasing for larger scales.
The limiting values 0 and 1 correspond to the extreme
cases of the area reduced to a point (the evaporation
contribution is zero), and the whole globe (all water
evaporated from the earth surface precipitates back to
the surface). For a given region of an intermediate scale,
the recycling ratio r depends on the processes involved
in the atmospheric branch of the regional hydrological
cycle.
It is also useful to define a local recycling ratio r (x,
y) for a specific point (x, y) of the region under consideration as

(2.7)

which is obtained by introducing P m from (2.8) into


(2.5).
b. Common assumptions
The following assumptions are commonly made in
precipitation recycling studies.
1) Time-averaged data is used, for example, monthly.

2500

JOURNAL OF CLIMATE

The time-averaged moisture fluxes represent the result of averaging the flux values obtained from daily
or twice-daily (sometimes four times daily) measurements. It is important to note that u and y , as
defined by (2.2), represent the normalized vertically
integrated moisture fluxes, and therefore their time
averages include contributions from both the mean
flow and the high-frequency transient eddies. Thus,
the transient eddy effect is included.
2) The next assumption states that at sufficiently long
timescales the change in storage of atmospheric water vapor is small compared with the atmospheric
water vapor fluxes (see Eltahir and Bras 1994 for
the data supporting this assumption).
3) The atmosphere is assumed to be well mixed, which
means that water molecules of external (advective)
origin and those of internal (evaporative) origin have
equal probabilities to be precipitated [such a formulation allows the inclusion of Lettau et al.s (1979)
model into the framework, see section 3a]. If all
evaporated moisture is mixed with the total precipitable water in the tropospheric column, this implies
that the ratio of locally evaporated and advected water molecules in the precipitation is the same as that
in the vertically integrated atmospheric moisture:
Pa
w
5 a.
Pm
wm

(2.11a)

This, with allowance for (2.3) and (2.4), may be also


expressed as
Pa
w
Pm
w
5 a or
5 m.
P
w
P
w

(2.11b)

The atmosphere above most land regions is well mixed


vertically so that the assumption of a well-mixed atmosphere is quite well justified (see, e.g., discussion in
Eltahir and Bras 1996).
The assumptions listed above are common to all recycling models (although there are some variations in
the form in which the assumption 3) of a well-mixed
atmosphere is applied), and they are also the underlying
assumptions of our model formulated in Part II. In specific recycling models discussed below, some additional
assumptions are made.
c. Equations
The underlying principle for all recycling models is
the conservation of atmospheric water vapor mass that
is expressed by the equations of conservation of the
total water vapor content (with sources and sinks due
to evaporation and precipitation)
](wu)
](wy )
1
5 E 2 P,
]x
]y
and its fraction of advective origin

(2.12)

VOLUME 14

](wa u)
](way )
1
5 2Pa .
]x
]y

(2.13)

It may be convenient to use the equation for the fraction


of moisture of local evaporative origin
](wm u)
](wmy )
1
5 E 2 Pm
]x
]y

(2.14)

obtained by subtracting (2.13) from (2.12). It is useful


to represent the last equation with the use of (2.8) and
(2.11b) as an equation for the recycling ratio r in the
form
](rwu)
](rwy )
1
5 E 2 rP,
]x
]y

(2.15a)

or, using the product rule of differentiation and Eq.


(2.12), in the form
wu

]r
]r
1 wy 5 E(1 2 r).
]x
]y

(2.15b)

If the velocity components u and y and the evaporation and total precipitation rates E and P are considered as given functions of the horizontal coordinates x
and y, then a complete system of equations for deriving
a recycling formula is formed by any two of Eqs. (2.12)
(2.15) together with the condition of a well-mixed atmosphere (2.11b). These equations have to be complemented by boundary conditions giving the moisture content in the air entering the region.
3. Recycling models
a. One-dimensional models
We will start with the one-dimensional models of Budyko and Drozdov (1953), Drozdov and Grigoreva
(1965), and Lettau et al. (1979) considered from a unified point of view that makes the derivations of the
results more straightforward and also facilitates extensions to two dimensions.
One-dimensional recycling models are based on the
assumption that a parallel airflow is traversing a land
region so that all the processes are considered along a
single straight streamline. If the x axis is chosen to be
parallel to the streamline, the one-dimensional counterparts of the two-dimensional equations of conservation
of water vapor (2.12) and (2.13) are
d(wu)
5E2P
dx
d(wa u)
5 2Pa .
dx

(3.1)
(3.2a)

The condition (2.11) of a well-mixed atmosphere completes the system. With the use of this condition in the
form (2.11b) the last equation becomes

15 JUNE 2001

2501

BURDE AND ZANGVIL

d(wa u)
P
52
(w u).
dx
(wu) a

(3.2b)

The one dimensional counterpart of Eq. (2.15b) for the


recycling ratio is represented as
d(1 2 r)
E
52
(1 2 r).
dx
(wu)

(3.3)

The boundary conditions for these equations giving the


moisture content in the air entering the region are

r 5 0 at x 5 0,

wu 5 wa u 5 w1 u 5 F11,
1

(3.4)

1
1

where w is the boundary value of w and F is the


moisture influx per unit length. The horizontal average
of a variable w(x) is calculated as

w 5

1
L

w (x) dx,

(3.5)

where L is the length of the region measured along a


streamline.
1) BUDYKOS

MODEL

In this model (first formulated in Budyko and Drozdov 1953 and described by Budyko 1974), the rectangular region traversed by a parallel uniform atmospheric
flow and located parallel to the streamlines is considered
(Fig. 1a) and all the vertical flux quantities P a , P, and
E are treated as constants equal to their average values.
Since P a is treated as a constant, the condition for a
well-mixed atmosphere can be imposed only on averages:
Pa
w
5 a.
Pm
wm

(3.6)

The recycling coefficient b defined by (2.7) may be then


calculated as

b5

w
.
wa

(3.7)

To calculate the quantities w and w a one needs to solve


Eqs. (3.1) and (3.2a) subject to the conditions of (3.4)
and find the averages by (3.5). The solutions of (3.1)
and (3.2a) are obtained in the forms
wu 5 F11 1 (E 2 P)x,

wa u 5 F11 2 Pa x,

(3.8)

with the averages


wu 5 F11 1

(E 2 P)L
,
2

w a u 5 F11 2

Pa L
,
2

(3.9)

and substituting into (3.7) yields

b5

F11 1 (E 2 P)L/2
,
F11 2 PL/(2b)

FIG. 1. Schematic representation of the region and the atmospheric


flow field for the recycling models: (a) Budykos model and Drozdov
and Grigorevas (1965) model; (b) Brubaker et al.s (1993) model;
(c) Burde et al.s (1996) model.

(3.10)

where the relationship P a 5 P/b has been used. Equation


(3.10) can be readily solved for b to give

b511

EL
.
2F11

(3.11a)

The recycling ratio r for the total land region defined


by (2.5b) is
r 5 1 2 b 21 5

EL
.
EL 1 2F11

(3.11b)

The main shortcomings of Budykos model, in addition to the one-dimensional formulation, originate
from the assumption of a constant flux P a . It is unreasonable from a physical viewpoint: while the fluxes P
and E can be, in principle, derived from observations,
the flux P a cannot be measured and so its distribution
should not be given but determined by the model equations. In addition, this assumption forces the use of the
condition of a well-mixed atmosphere in the form (3.6),

2502

JOURNAL OF CLIMATE

which does not follow from the local condition (2.11a)


justified by observations, and it is hardly reasonable for
large land scales. From a mathematical viewpoint the
assumption of constant P a is unnecessary: for given P
and E, Eqs. (3.1) and (3.2a) represent a system of two
equations for two variables (wu) and (w a u), and any
other relation is redundant.
Note, however, that this assumption is justified when
a territory of small size, L is considered, such that

L ]Pa
K1
Pa ]x

(3.12)

is satisfied. It is useful to define the spatial range of


validity of (3.12) (actually validity of Budykos model)
by solving Eqs. (3.1) and (3.2a) without imposing the
assumption of constant P a . We assume as before that P
and E distributions are replaced by their averages but
we do not impose any conditions on the advected part
of precipitation P a . The solutions of Eqs. (3.1) and
(3.2a) subject to the boundary conditions (3.4) are

Pa (x) dx.

MODEL

Drozdov and Grigoreva (1965) developed a generalization of Budykos model by waiving the assumption
that the evaporation and precipitation fluxes are constants along a streamline. Since in this model the condition of constant P a is not imposed, the condition of a
well-mixed atmosphere can be used in its local form
(2.11) and, correspondingly, equation (3.3) obtained
with the use of (2.11) may be applied. Then, the system
of Eqs. (3.1) and (3.3) may be solved to find r (x) for
given distributions E(x) and P(x) (note that such a derivation of the recycling formula is much simpler than
that presented in the original work by Drozdov and Grigoreva 1965). Equation (3.1) is solved and the solution
satisfying the boundary condition (3.4) is

(wu) 5 F11 1

(E 2 P) dx.

(3.18)

[F11 1 (E 2 P)x]Pa 5 F11 2

Pa (x) dx P,

(3.15)

(3.16)

This implies that Budykos formula (3.11b) is justified


only for the extreme case of small territories, and correspondingly small recycling ratios, when it is simplified to
EL
.
2F11

(3.14)

It is evident from (3.12) and (3.15) that the spatial range


of validity of Budykos model is defined by
EL K F11 .

r 5 1 2 exp 2

E dx
(wu)

(3.17)

Thus, the use of Budykos formula for an analysis of


the scale dependence of the recycling process is not
consistent since it is justified only close to an extremity
of small scale and for correspondingly small recycling
ratios where it gives the trivial linear dependence of r
on L.

F 1
1
1

.
x

(E 2 P) dx

E dx

5 1 2 exp 2

that after differentiating with respect to x and making


transformations may be expressed as
L dPa
EL
5 1
.
Pa dx
F1 1 (E 2 P)x

[E
E

(3.13)

Applying the condition of a well-mixed atmosphere in


its local form (2.11b) yields

r5

GRIGOREVAS (1965)

AND

Then the solution of (3.3) subject to the condition (3.4)


for r is

wu 5 F11 1 (E 2 P)x,
wa u 5 F11 2

2) DROZDOV

VOLUME 14

(3.19)

Note that Drozdov and Grigorevas (1965) resulting


formula is represented in a form differing from (3.19)
since they derived it in another way, which is equivalentbut still more complicatedto solving Eq. (3.2b)
for (w a u) instead of (3.3). Their formula can be reduced
to (3.19) by identical transformations.
For specific distributions E(x) and P(x) the recycling
ratio r for the total land region may be found by (2.10).
The important specific case is that of E and P independent of x, which implies that the evaporation and total
precipitation fluxes are taken to be equal to their average
values (but the fluxes P a and P m are not). Drozdov and
Grigoreva (1965) studied extensively the results for this
case, and the model considered in Brubaker et al. (1993)
as Drozdov and Grigorevas model, is just this case. In
this case integrating implemented in (3.19) yields

r 5 12

x(E 2 P)
11
F11

r 5 1 2 e 2Ex/F11

E/(P2E)

(E P)

(3.20a)

(E 5 P).

(3.20b)

The ratio for the whole region is then calculated as

15 JUNE 2001

r512

F11
F11
12 1
PL
F1 1 (E 2 P)L

P/(E2P)

(E P, F11 1 EL $ PL)
(3.21a)
r512

2503

BURDE AND ZANGVIL

1
1

F
(1 2 e 2PL/F1 )
PL

(E P).

(3.21b)

The condition placed in parentheses in (3.21a) providing


a positiveness of the expressions in square brackets expresses a physically required condition that the moisture
outflow should be positive.
In accordance with the remarks made at the end of
section 3a(1), one should expect that the formulas (3.21)
turn into Budykos formula at the limit of small scales
(3.16). Indeed, expanding the expressions in braces in
(3.21) with respect to a small parameter a 5 EL/F11 we
obtain Eq. (3.17). It is worth noting, however, that due
to the special structure of the formulas (3.21), where
braces are multiplied by a large quantity 1/a, the expansions are to be made up to the order of a 2 . This
indicates that the range of validity of Budykos formula
expands to larger scales than one could expect on the
basis of considerations given by (3.12)(3.17).
3) LETTAU

ET AL.S

(1979)

b. Two-dimensional models
1) BRUBAKER ET AL.S (1993)
BUDYKOS MODEL

P 5 p E E 1 r p w,

F1 5 2

(3.22)

where p E and r p are interregional parameters. The dimensionless parameter p E serves to express (by p E E)
the part of evaporation returned to the regional airsoil
interface by fast recycling, which refers to local
showers yielding rain before all cloud water is mixed
with the total precipitable water in the average tropospheric column above the region, while the dimensional
r p (yr 21 ) is the flushing frequency of atmospheric moisture. Lettau et al. (1979) use Eq. (3.1) jointly with (3.22)
to find the distribution of precipitable water w and the
corresponding distribution of precipitation P over the
region for some representative values of E, p E , r p , u,
and du/dx. To avoid using auxiliary variables introduced
in Lettau et al. (1979) we will show the result for du/
dx 5 0 and constant E, p E , r p , and u. Then the solution
of the equation yielded by introducing (3.22) into (3.1)
subject to the boundary condition (3.4) is
(1 2 pE )E
(1 2 e 2rp x/u ),
rp

(3.23)

and P distribution is obtained by substituting (3.23) into


(3.22). Lettau et al. (1979) use the results to calculate
several nondimensional characteristics. They consider
p E E as the recycling part of the regional evaporation

MODIFICATION OF

Brubaker et al. (1993) suggested to extend Budykos


model to a two-dimensional land region by defining the
influx F 1 as the integral of the atmospheric moisture
inflow over all the boundary of the region

MODEL

Lettau et al. (1979) developed a one-dimensional


model in which the depleting process is divided into
two parts

w 5 w1 e 2rp x/u 1

and r p w as precipitation derived from advected precipitable water and correspondingly treat the ratio P(r p w) 21
as the recycling coefficient. Evidently, it is not completely correct: Eq. (3.1) and its solution (3.23) incorporate replenishment of precipitable water w by the
evaporative flux and thus the quantity r p w includes also
part of the precipitation of evaporative origin (in addition to p E E). Therefore Lettau et al.s (1979) comparison of their results with the values of the recycling
coefficient b produced by Budykos formula is not valid.
To construct a consistent recycling model based on the
assumption (3.22) it is necessary to solve also Eq. (3.2a)
[where P a 5 r p w a in accordance with (3.22) and the
assumption of a well-mixed atmosphere] to find the distribution of advected moisture. Then the recycling coefficient b 5 P/P a can be defined.

wg Vg n g dg.

(3.24)

Gin

In the boundary integral (3.24), w g and V g 5 (u g , y g )


are, respectively, the moisture content and the flow velocity vector on the boundary, n g is the outward unit
normal vector, and Gin denotes a part or parts of the
boundary across which the atmospheric motion is inward. The gain and loss to regional atmospheric moisture due to evaporation and precipitation are calculated
by multiplying the evaporation and precipitation by the
area of the region, instead of the length of the streamline.
Next, the average fluxes of total and advected moisture
over the region are taken as the arithmetic means of the
incoming and outcoming moisture
Q5
Qa 5

F 1 1 (F 1 1 EA 2 PA)
(E 2 P)A
5 F1 1
,
2
2
F 1 1 (F 1 2 Pa A)
P A
5 F1 2 a ,
2
2

(3.25)

from which by invoking the condition of a well-mixed


atmosphere in the form Q/Q a 5 P/P a 5 b the formula
for the recycling coefficient is obtained in the same way
as in section 3a(1) but in a slightly different form, namely,

b511

EA
.
2F 1

(3.26a)

This formula is identical to (3.11a) with the area of the


region A replacing the length scale L and the influx F 1

2504

JOURNAL OF CLIMATE

replacing the influx per unit length F11 . The counterpart


of the formula (3.11b) for the recycling ratio will be
EA
r5
5
EA 1 2F 1

1
.
2F 1
11
EA

(3.26b)

This model cannot be regarded as an extension to the


two-dimensional case. The arguments are based on the
assumption (3.25) that the average moisture flux is the
arithmetic mean of the incoming and outcoming moisture, which, in general, is not correct when the flow is
two-dimensional. It is justified only in the one-dimensional case where it follows from the one-dimensional
equations of conservation of water vapor (3.1) and
(3.2a) having the linear distributions (3.8) as solutions.
Two-dimensional arguments should be based on the
two-dimensional equations of conservation of water vapor (2.12) and (2.13) and it is evident (see also the next
section) that solutions with the linear variation of moisture fluxes cannot be obtained if the fields u(x, y) and
y (x, y) differ from a parallel flow field. In other words,
the formula based on the assumption of a linear moisture
flux distribution cannot incorporate the effects of moisture redistribution due to streamline curvature and velocity variations.
Even though the modified Budykos formula (3.26)
cannot be considered as a correct extension of Budykos
model to the two-dimensional flow case, it [with the
influx calculated by (3.24)] provides a correct extension
when a two-dimensional region is traversed by the parallel airflow, but the region boundaries are not parallel
to streamlines (Fig. 1b). It also allows for a nonuniform
moisture distribution in the flow entering the region. To
show this we will apply the one-dimensional Eqs. (3.1)
and (3.2a), with x being a longitudinal coordinate, to
the depleting process along one of the streamlines,
which corresponds to an arbitrary value y of the transversal coordinate with the length L(y) confined inside
the region (see Fig. 1b). The arguments similar to those
presented in section 3a(1) lead to
w
(y)u 5 w1 (y)u 1

(E 2 P)L(y)
,
2

w
a(y)u 5 w 1 (y)u 2

Pa L(y)
,
2

VOLUME 14

wu 5 F 1 1
w a u 5 F1 2

where w and w a denote the streamline averages that


now may differ from areal averages. These streamline
averages and the moisture amount in the air entering
the region w1 depend on the streamline position y. Integrating these equations across streamlines [for the first
terms on the right-hand sides of Eq. (3.27a), dy is represented as 2dg V g n g /u to integrate along the boundary]
yields

Pa A
,
2

(3.27b)

where overbars now denote averaging over all the region and F 1 is the influx determined as in (3.24). Using
(3.27b) in (3.6) and (3.7) yields (3.26).
2) BURDE ET AL.S (1996)
BUDYKOS MODEL

EXTENSION OF

Burde et al. (1996) developed a two-dimensional recycling model that represents an exact extension of Budykos model to the two-dimensional flow case (Fig.
1c). As distinct from Budykos model, Burde et al.s
model is based on the two-dimensional equations of
conservation of water vapor (2.12) and (2.13), but it
retains the assumption of Budykos model that all the
vertical flux quantities P a , P, and E are constants equal
to their average values. Correspondingly the condition
for the atmosphere to be well mixed is imposed on the
averages as is in (3.6).
The model resultthe expression for the recycling
coefficient bmay be represented in the form resembling (3.26) and differing from it by a correction factor
that depends on the atmospheric flow structure. Such a
representation is based on the fact that solutions of Eqs.
(2.12) and (2.13) have the forms
w 5 w1 1 (E 2 P)G(x, y),
wa 5 w1 2 Pa G(x, y),

(3.28)

where the function G(x, y) is determined completely by


the velocity fields and condition G(0, y) 5 0 (we have
assumed the uniform moisture distribution at x 5 0 and
make some other simplifications below - see Burde et
al. 1996 for the general case). Equations (3.28) are twodimensional counterparts of Eqs. (3.8). Following the
scheme of derivation of Budykos formula in section
3a(1), we find by (2.6) the areal averages of distributions
(3.28) and substitutes them into (3.7), which with the
use of the relation P a 5 P/b gives the counterpart of
(3.10) in the form

b5
(3.27a)

(E 2 P)A
,
2

w1 1 (E 2 P)G
.
w1 2 PG/b

(3.29)

Solving for b we get

b511

EG
.
w1

(3.30)

Considering for simplicity the case when the moisture


enters a rectangular region only through the side x 5
0 and u(0, y) 5 u 0 is uniform, we may express the
moisture influx F 1 through the characteristic moisture
content w1 simply by F 1 5 w1 u 0 H and represent
(3.30) as

15 JUNE 2001

2505

BURDE AND ZANGVIL

b511

EA
K
2F 1

(3.31)

the cell. With the use of the relations expressing the


condition of a well-mixed atmosphere as
O m 5 r O, P m 5 r P

where
K5G

Eq. (3.32) may be represented in the forms

2u 0
.
L

E(DA) 1 I 5 O 1 P(DA),

This formula differs from (3.26) by the correction


factor K. This correction factor incorporates all the information concerning the flow structure and it also depends on the geometrical parameter of the region L/H.
It is easily seen that (3.31) transforms into Budykos
formula in Brubaker et al.s form (3.26) in the case of
a parallel air flow (and only in this case). Then G 5 x/
u 0 and G defined by (2.6) is G 5 L/2u 0 so that K 5 1.
The effects of the flow field on precipitation recycling
may be investigated on the basis of this analytical model
within the framework of Budykos approach. Evidently,
the model retains the shortcoming of Budykos model
connected with the assumption of a constancy of the
flux P a that prevents us from using the condition of a
well-mixed atmosphere in its local form. This shortcoming is overcome in the new model developed in Part
II.
3) ELTAHIR

AND BRASS
RECYCLING MODEL

(1994)

(3.33)

NUMERICAL

Eltahir and Bras (1994) developed a two-dimensional


numerical model for estimation of regional precipitation
recycling. The region is covered by a horizontal grid
dividing the whole area into small cells. The relations
expressing conservation of water vapor mass, together
with the condition of a well-mixed atmosphere, are applied to a grid cell to obtain a formula expressing the
recycling ratio at this cell through the evaporation rate
into the same cell and water vapor fluxes from adjacent
cells. An iterative procedure is applied to find the recycling ratio values satisfying this formula for all cells.
The recycling ratio for the total area of the region is
obtained by (2.5a).
Eltahir and Brass model is equivalent to a finite difference scheme for solving the differential equations of
the problem. Using integral forms of partial differential
equations (conservation laws) for a grid cell is a common way to construct their discrete representations. In
the precipitation recycling problem, there is freedom in
choosing a minimal set of equations from (2.12) to
(2.15). In Eltahir and Brass method the integral forms
of equations (2.12) and (2.14) are used. Integrating these
equations over the cell area DA and using the Gaussian
theorem to transform surface integrals into boundary
integrals yields
2I 1 O 5 E(DA) 2 P(DA),
2Im 1 Om 5 E(DA) 2 Pm (DA),

(3.32)

where I and O denote the moisture fluxes in and out of

E(DA) 1 Im 5 r [O 1 P(DA)]

(3.34)

from which Eltahir and Brass formula follows

r5

Im 1 e
Im 1 e
5
,
I1e
Ia 1 Im 1 e

(3.35)

where e 5 E(DA). It should be clarified that (3.35) is


not an analytical formula for the local recycling ratio
that can be directly used for estimation of regional precipitation recycling on the basis of regional parameters.
This formula includes internal variables I m and I a defined
for a given cell, which cannot be derived from observations but can be the result of a solution, namely, the
result of an iterative procedure adjusting the values of
I m and I a for different cells to each other. In fact, (3.35)
represents an implicit scheme since the data are available only for the total flux I, and it is partitioned into
I m and I a using the relation I m 5 r I, so that Eltahir and
Brass method corresponds to the following iterative
formula

r (s11) 5

r (s) I 1 e
,
I1e

(3.36)

where s is the iteration number. It is evident that


different iterative formulas can be constructed on the
basis of discrete representations of the problem equations but this scheme has already been proved to converge quickly by Eltahir and Bras (1994).
4) MODEL BASED ON
(IMB MODEL)

INTEGRAL MOISTURE BUDGET

In this approach (see, e.g., Schar et al. 1999) the


equations of water vapor balance for each fraction of
precipitable water are applied in their integral form to
all the region. To obtain the regional recycling ratio, the
condition of a well-mixed atmosphere is applied also in
an integral form; namely, it is assumed that the relative
content of evaporated moisture in integral regional precipitation is the same as that in the moisture flux out of
the region.
P m 5 r P, F2m 5 r F 2 ,

(3.37)

where F 2 is the total outflow of moisture and F2m is the


outflow of moisture evaporated within the domain. We
will not reproduce the derivation here as it is equivalent
to that made in the previous section for one grid cell
[Eqs. (3.32)(3.35)], but applied to the total region. One
has to set I m 5 0 in (3.35) for the case of the total
region, which leads to the following formula

2506

JOURNAL OF CLIMATE

r5

E
,
F 1E
1

(3.38)

where E 5 EA and F 1 replaces I in our notation. Note,


that Eltahir and Bras (1994) also considered a possibility
of applying their Eq. (3.35) to the entire region as one
grid cell in the form (3.38).
Evidently, such a model represents a rough approximation since it ignores completely the character of
moisture distribution within the region dictated by the
differential equations of conservation of water vapor.
Validity of the assumption (3.37) for a large area also
seem doubtful. Trenberth (1999) notes that this assumption is clearly not satisfied over the region considered by Eltahir and Bras (1994) as it follows from
their Fig. 12.
Eltahir and Bras (1994) reported that applying (3.38)
to the Amazon basin yields an estimate that is not significantly different from those obtained using 2.58 3
2.58 grid in their numerical method. They considered
this fact as a proof of an insensitivity of their scheme
to the spatial resolution of the grid. Such a result could
be incidental. The validity of the estimate by (3.38),
which produces higher values of recycling ratio as compared with (3.20), may rather indicate the existence of
a proper flow structure in the regionin accordance
with the results presented in Burde et al. (1996) and
Part II of this paper. However, since the degree of increase of the recycling ratio strongly depends on the
flow type and the geometrical parameter of the region,
Eq. (3.38) in no way can be universalit can be valid
under proper flow conditions and not be valid under
other conditions.
Note in this connection that the formula (3.38) had
been earlier suggested on the basis of water vapor budget considerations (tank model) and applied to a dataset over the midwestern United States by Zangvil et
al. (1992) to describe the tendency for the recycling
ratio to increase under complex flow conditions.
4. Discussion and conclusions
We have discussed several recycling models developed in recent decades. Actually most of them originate
from the approach developed in the work by Budyko
and Drozdov (1953; reproduced later in Budyko 1974),
and may be considered as modifications and generalizations of Budykos model. Mostly, the models are analytical and aimed at obtaining a formula for the regional
recycling ratio.
We have shown that all the models have the same
underlying principlesin spite of different forms into
which the authors present their models. These principles
are represented by equations for conservation of mass
for two fractions of water vapor in the atmosphere and
the equation expressing the condition that these two
fractions are well mixedall written under the common
assumptions listed in section 2. If any of these three

VOLUME 14

equations is missing the model is incomplete and its


results inconsistent. Different models differ by additional assumptions made in order to simplify either the
equations or their solution or to circumvent a solution.
Although the only way to test a recycling model is to
carry out the experiments (observational or numerical),
in which the fate of molecules of different origins is
traced (the isotope or tagged water experiments), a comparison of the estimates of precipitation recycling produced by different models is of interest as it shows to
what extent a model can improve the estimates.
The results of applying the modification of Budykos
formula by Brubaker et al. (1993) to the European part
of the former USSR do not differ significantly from those
obtained by Budyko (1974) on the basis of his formula.
Budyko (1974) estimated the local component of precipitation to be 12% of the total on an annual basis, and
similar results have been reported by Brubaker et al.
(1993). The results of applying the generalization of Budykos model by Drozdov and Grigoreva (1965) to some
specific regions also did not reveal a large difference
compared with the results of calculations using Budykos
formula. The computations of moisture cycle components
with the use of Drozdov and Grigorevas (1965) model
for the former European USSR, presented in Shiklomanov (1989), gave the same estimate of 11% for recycled precipitation as Budyko (1974) and Sellers (1965).
Drozdov and Grigoreva (1965) also reported that the
discrepancies between yearly estimates were within the
accuracy limits for the calculations. The comparative
study of the formulas produced by these two models
made in Brubaker et al. (1993) showed that the discrepancies should not be large if one does not consider extreme values of the model parameters. Lettau et al.s
(1979) modification should have increased the amount
of precipitating water of local origin but their model is
incomplete and therefore the calculation of the recycling
coefficient is not valid as it is shown in section 3. The
numerical character of the recycling model by Eltahir
and Bras (1994) does not give a possibility to analyze,
in general, the corrections it should make to estimations
by Budykos model; a comparison of the results produced
by these two models has not been made. Analysis of the
two-dimensional extension of Budykos model by Burde
et al. (1996) for specific flow structures shows that the
corrections to Budykos estimation may be significant if
the flow differs significantly from a parallel structure.
Numaguti (1999) considers that the cases of a significant
disagreement between the direct estimation of the recycling ratio by tagged water experiments and the indirect
estimations by Budykos formula may be related to the
tendency of underestimation suggested by Burde et al.
(1996) under complex flow conditions. Dirmeyer and
Brubaker (1999) calculated back trajectories of atmospheric moisture for the Mississippi River basin and its
subregions to trace and estimate the evaporative source
of precipitation falling over the regions and found that
the Budyko method gives lower values for the recycling

15 JUNE 2001

BURDE AND ZANGVIL

ratio for all basins and months considered. The discrepancies were too significant (for the Mississippi River basin as a whole, 27%47% by the back trajectories method
against 12%24% by the Budyko method) to explain
them, for example, by the differences in the representative height of advection between the two methods or
by a uniformity of the precipitation and evaporation fluxes assumed in Budykos model. At the same time, the
disrepancies between the estimates by the two methods
are of the order of magnitude of the corrections to Budykos estimation produced by Burde et al.s model.
To conclude, among a number of analytical models,
which extend Budykos model, the extension of the
model to a nonparallel and nonuniform flow made in
Burde et al. (1996) seems to correct to a large extent
Budykos model estimates. However, this extended
model, aimed at a consistent inclusion of two-dimensional effects into the framework of Budykos approach,
retains another limitation of Budykos model; namely,
the model assumes all the precipitation and evaporation
fluxes to be constants. This assumption cannot be treated
simply as an approximation (rather rough) of the regional evaporation and total precipitation data but it also
introduces some inconsistency into the model since it
imposes conditions on the regional distribution of precipitation of external origin that is not observed and
should be the model result. In addition, it forces us to
use the condition of a well-mixed atmosphere for horizontal averages instead of applying it to the ratios of
local precipitation rates and local moisture contents.
An analytical two-dimensional model, which would
overcome these inconsistencies by incorporating both
the effects of atmospheric flow structure and the inhomogeneity of the fluxes, should further improve estimations of precipitation recycling for land regions and
allow to more consistently size up effects of flow fields
in the recycling process. Such a model being free of
restrictions and assumptions of simpler models could
also provide an indication about a range of validity of
the simpler models. Such a new model is presented in
Part II of this paper.
Acknowledgments. We are grateful to two reviewers
for their useful comments that contributed to the improvement of this paper.
APPENDIX
List of Symbols
Most of the symbols are also used in Part II. The
symbols used only in Part I are marked by (PI).
5 EDA, where DA is the area of a grid cell
e
[(3.35)] (PI)
Acceleration due to gravity (PI)
g
Atmospheric pressure (PI)
p
Surface pressure (PI)
p0
Interregional parameter in Lettau et al.s model
pE
[(3.22)] (PI)

q( p)
r
rp
u ( p),
y ( p)
u, y
u0
w
w1
w
x, y
A
E

F
F1
F11
F2
G
H
I
K
L
O
P
Q
V
a

b
g
r
rL
w
Gin

2507

Specific humidity [(2.1), (2.2)] (PI)


Recycling ratio for the total land region [(2.5),
(2.10)]
Interregional parameter in Lettau et al.s model
[(3.22)] (PI)
The wind vector components (PI)
Vertical averages of the wind vector components
weighted according to the specific humidity profile [(2.2)]
Horizontal velocity scale
Vertically integrated water vapor (precipitable)
depth
Characteristic value of the moisture content
Streamline average of the moisture content
[(3.27a)] (PI)
Horizontal coordinates
Area of the region
Evaporation flux
5 EA [(3.38)] (PI)
5 (F (x) , F ( y ) ) Vertically integrated horizontal
flux of atmospheric moisture [(2.2)]
Influx of atmospheric moisture
Moisture influx per unit length
Outflux of atmospheric moisture
Auxiliary function in Eqs. (3.28)(3.31) (PI)
Dimension of the region along the y axis
Moisture flux in a grid cell in Eltahir and Brass
model [(3.32)(3.36)] (PI)
Correction factor to Budykos formula in Burde
et al.s model [(3.31)] (PI)
Dimension of the region along the x axis
Moisture flux out a grid cell in Eltahir and
Brass model [(3.32)(3.36)] (PI)
Precipitation flux
Average moisture flux in Brubaker et al.s model [(3.25)] (PI)
5 (u, y )
5 EL/F11 A parameter used in considerations
related to (3.21) (PI)
Budykos recycling coefficient [(2.7)]
Part of the boundary
Local precipitation recycling ratio [(2.8), (2.9)]
Liquid water density
Any variable
Part or parts of the boundary across which the
atmospheric motion is inward

Subscripts
a
m

Parts of the water vapor content or precipitation


of advective origin
Parts of the water vapor content or precipitation
of local (evaporative) origin
Moisture contents and flow velocities on the
boundary

Superscripts

Horizontal average of w (x, y) over the region

2508

JOURNAL OF CLIMATE
REFERENCES

Beljaars, A. C. M., P. Viterbo, M. J. Miller, and A. K. Betts, 1996:


The anomalous rainfall over the United States during July 1993:
Sensitivity to land surface parameterization and soil-moisture
anomalies. Mon. Wea. Rev., 124, 362383.
Brubaker, K. L., D. Entekhabi, and P. S. Eagleson, 1993: Estimation
of continental precipitation recycling. J. Climate, 6, 10771089.
Budyko, M. I., 1974: Climate and Life. Academic Press, 508 pp.
, and O. A. Drozdov, 1953: Zakonomernosti vlagooborota v
atmosfere (Regularities of the hydrologic cycle in the atmosphere). Izv. Akad. Nauk SSSR, Ser. Geogr., 4, 514.
Burde, G. I., A. Zangvil, and P. J. Lamb, 1996: Estimating the role
of local evaporation in precipitation for a two-dimensional region. J. Climate, 9, 13281338.
Dirmeyer, P. A., and K. L. Brubaker, 1999: Contrasting evaporative
moisture sources during the drought of 1988 and the flood of
1993. J. Geophys. Res., 104, 19 38319 397.
Drozdov, O. A., and A. S. Grigoreva, 1965: The Hydrologic Cycle
in the Atmosphere. Israel Program for Scientific Translations,
282 pp.
Druyan, L. M., and R. D. Koster, 1989: Sources of Sahel precipitation
for simulated drought and rainy seasons. J. Climate, 2, 1438
1446.
Eltahir, E. A. B., and L. B. Bras, 1994: Precipitation recycling in the
Amazon Basin. Quart. J. Roy. Meteor. Soc., 120, 861880.
, and , 1996: Precipitation recycling. Rev. Geophys., 34,
367378.
Entekhabi, D., I. Rodriguez-Iturbe, and R. L. Bras, 1992: Variability
in large-scale water balance with land surfaceatmosphere interaction. J. Climate, 5, 798813.
Joussaume, S., J. Jouzel, and R. Sadourny, 1984: A general circulation
model of water isotope cycles in the atmosphere. Nature, 311,
2429.
, R. Sadourny, and C. Vignal, 1986: Origin of precipitating water
in a numerical simulation of the July climate. OceanAir Inter.,
1, 4356.
Jouzel, J., K. Froehlich, and U. Schotterer, 1997: Deuterium and
oxygen-18 in present-day precipitation: Data and modelling. Hydrol. Sci. J., 42, 747763.
Koster, R., J. Jouzel, R. Souzzo, G. Russel, D. Rind, and P. S. Eagleson, 1986: Global sources of local precipitation as determined
by the NASA/GISS GCM. Geophys. Res. Lett., 13, 121124.
Lettau, H., K. Lettau, and L. C. B. Molion, 1979: Amazonias hy-

VOLUME 14

drologic cycle and the role of atmospheric recycling in assessing


deforestation effects. Mon. Wea. Rev., 107, 227238.
Numaguti, A., 1999: Origin and recycling processes of precipitating
water over the Eurasian continent: Experiments using an atmospheric general circulation model. J. Geophys. Res., 104 (D2),
19571972.
Oglesby, R. J., and D. J. Erickson, 1989: Soil moisture and the persistence of North American drought. J. Climate, 2, 13621380.
Rodriguez-Iturbe, I., D. Entekhabi, and R. L. Bras, 1991: Nonlinear
dynamics of soil moisture at climate scales: Stochastic analysis.
Water Resour. Res., 27, 18991906.
Rowntree, P. R., and J. A. Bolton, 1983: Simulations of the atmospheric response to soil moisture anomalies over Europe. Quart.
J. Roy. Meteor. Soc., 109, 501526.
Salati, E., A. DallOlio, E. Matsui, and J. R. Gat, 1979: Recycling
of water in the Amazon basin: An isotopic study. Water Resour.
Res., 15, 12501258.
Schar, C., D. Luthi, and U. Beyerle, 1999: The soil precipitation
feedback: A process study with a regional climate model. J.
Climate, 12, 722741.
Sellers, W. D., 1965: Physical Climatology. University of Chicago
Press, 272 pp.
Shiklomanov, I. A., 1989: Climate and water resources. Hydrol. Sci.
J., 34, 495529.
Shukla, J., and Y. Mintz, 1982: Influence of land-surface evapotranspiration on the Earths climate. Science, 215, 14981501.
, C. Nobre, and P. Sellers, 1990: Amazon deforestation and climate change. Science, 247, 13221325.
Trenberth, K. E., 1998: Atmospheric moisture residence times and
cycling: Implications for rainfall rates and climate change. Climatic Change, 39, 667694.
, 1999: Atmospheric moisture recycling: Role of advection and
local evaporation. J. Climate, 12, 13681381.
Yeh, T. C., R. T. Wetherald, and S. Manabe, 1984: The effect of soil
moisture on the short-term climate and hydrology changeA
numerical experiment. Mon. Wea. Rev., 112, 474490.
Zangvil, A., D. H. Portis, and P. J. Lamb, 1992: Interannual variations
of the moisture budget over the Midwestern United States in
relation to summer precipitation. Part II: Impact of local evaporation on precipitation. Symp. Programme and Abstracts, The
Yale Mintz Memorial Symp. on Climate and Climate Change,
Jerusalem, Israel, Israel Meteor. Soc., Amer. Meteor. Soc., and
National Science Foundation, 101.

You might also like