International Journal of Applied Earth Observation
and Geoinformation 5 (2004) 129140
Evaluation of the temporal variability of the evaporative
fraction in a tropical watershed
H.O. Farah
a,
, W.G.M. Bastiaanssen
b
, R.A. Feddes
c
a
Department of Civil and Structural Engineering, Moi University, P.O. Box 3900, Eldoret, Kenya
b
Garstsraat 23, 4021 AB Maurik, The Netherlands
c
SubDepartment of Water Resources, Department of Environmental Sciences, Wageningen Agricultural University,
Nieuwe Kanaal 11, 6700 PA Wageningen, The Netherlands
Received 4 September 2003; accepted 26 January 2004
Abstract
Evaporation exhibits diurnal variation in response to the changes in the available energy at the land surface. This requires
continuous measurements of evaporation to determine daily total evaporation. This is not feasible without sophisticated eld
equipment, which at the end, only provides eld scale evaporation rates. Remote sensing methods are a good alternative but
these give snapshot measurements. If the partitioning of available energy into the different surface uxes can be assumed to
be diurnally constant, then instantaneous remotely sensed measurements could be used to derive daily total evaporation. In
situ evaporation measurements were obtained for about a year at a grassland and woodland site in the Lake Naivasha basin,
Kenya. These measurements were used to test the validity of the diurnal constancy of the partitioning of the available energy,
expressed as evaporative fraction, and the extrapolation of evaporation from instantaneous to daily totals. A good relationship
between midday and average day evaporative fraction was obtained at the two sites. Estimated daily evaporation from midday
evaporative fraction was within 10% of measured evaporation for both sites. The deviation reduced if evaporation is further
integrated in time. The seasonal progression of evaporative fraction is gradual at both sites although grassland evaporative
fraction responds faster to changes in rainfall and moisture availability. The results provide a basis for the determination of
regional evaporation across a season in tropical watersheds if evaporative fraction is determined instantaneously at intermittent
intervals of 510 days.
2004 Elsevier B.V. All rights reserved.
Keywords: Evaporation; Evaporative fraction; Remote sensing; Lake Naivasha basin
1. Introduction
Evaporation is required on a daily as well as longer
time scales for applications in hydrology, agriculture,
forestry and environmental studies in general. How
ever in practice, continuous daily evaporation mea
surements are rarely available. Daily reference or po
Corresponding author.
Email address: highlandeld@africaonline.co.ke (H.O. Farah).
tential evaporation can be estimated from mean daily
values of available meteorological variables such as
temperature, solar radiation, humidity and wind speed
(Allen et al., 1998). More recently, one or more in
stantaneous measurements of evaporation have been
used to estimate daily total evaporation (Brutsaert
and Suigita, 1992). There has been a growing inter
est in this approach because of its attractiveness for
remote sensing applications. Remote sensing offers
a means of estimating actual evaporation at a large
03032434/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jag.2004.01.003
130 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
spatial scale, which is not possible with the tradi
tional point methods. Many techniques have been
proposed to solve the surface energy balance from re
motely sensed surface temperature, surface reectance
and vegetation indices (Moran and Jackson, 1991;
Kustas and Norman, 1996; Bastiaanssen et al., 1999).
Remote sensing data are however instantaneous mea
surements and a method is required to temporally
integrate instantaneous estimates of evaporation.
Latent heat ux (L) and other components of the
energy balance display considerable diurnal variation
over land surfaces. However several ratios of the uxes
have been shown to be relatively constant during day
light hours (Jackson et al., 1983; Shuttleworth et al.,
1989; Bastiaanssen et al., 1996). The classical energy
partitioning indicator is the Bowen ratio (), which is
a ratio of the sensible heat ux (H) and L. The pit
fall of applying for time integration is that it shows
distinct diurnal variation features. More recently the
evaporative fraction () has been found to have little
variations during daytime, although it is directly re
lated to (Crago and Brutsaert, 1996). Evaporative
fraction is dened as:
=
L
R
n
G
=
L
L +H
=
1
1 +
(1)
where, R
n
is the net radiation and G the soil heat
ux. Shuttleworth et al. (1989), were the rst to no
tice the constancy of during daylight hours. They
analyzed 4 clear sky days data from the rst ISLSCP
eld experiment (FIFE) over relatively homogeneous
grasslands and found that midday is nearly equal to
the average daylight . Nichols and Cuenca (1993),
used 72 days data from Hydrologic Atmospheric
Pilot ExperimentModelisation du Bilan Hydrique
(HAPEXMOBILHY) experiment and showed that
the midday was highly correlated with average
daytime but that the midday and daytime are
not statistically equal. Crago (1996a), evaluated 77
days data from FIFE. He used the data irrespective of
weather conditions of a particular day and concluded
that midday is signicantly different from the av
erage daytime value, the reason being the concaveup
shape of the diurnal progression of .
The central question is whether an instantaneous
value of can be used to estimate daily actual evap
oration (E) as:
E
d
=
ins
(R
n
G)
d
(2)
where, the subscripts d and ins indicate total daytime
and instantaneous values respectively. This way of ex
pressing E is a simple approach to integrate E on a
daily basis and across a season, if at least the tem
poral variations of are known. However, Eq. (2)
may not be valid under nonclear sky conditions be
cause the diurnal constancy of may not be satised
under cloudy conditions (Zhang and Lemeur, 1995).
For areas with persistent cloud cover, such as in the
humid tropics, it is important to test the validity of
Eq. (2). In order to assess the performance of the ap
proach, long term data series of measurements are re
quired so that a wide range of different conditions are
encountered. Most of the previously published stud
ies have used data from relatively short time periods
as reported above. In this study, eld data collected
over a period of about 1 year in Lake Naivasha basin
in Kenya is used to investigate the applicability of
the method to estimate E at daily scale and for
a season. Continuous daily E measurements at two
sites were compared with daily E estimated by using
Eq. (2).
The objective of this paper is to demonstrate the ca
pability of instantaneous measurements of to esti
mate the average day and E throughout a season in
tropical watersheds with data scarcity problems. Al
though only eld data was used in this study, the re
sults are expected to establish a sound basis for the es
timation of E from instantaneous remote sensing data
and routine daily weather data. The theoretical back
ground of and reasons for its stable diurnal behav
ior are discussed in Section 2. The eld measurements
carried out are detailed in Section 3. In Section 4, the
diurnal stability of is discussed. The results of the
comparison between instantaneous and average day
are presented in Section 5, while the results of es
timating time integrated E is presented in Section 6.
Finally the seasonal variations of are described in
Section 7.
2. Theoretical background
2.1. Reasons for the diurnal stability of
The diurnal behavior of can be understood from
its relationship with atmospheric conditions and sur
face characteristics. The PenmanMonteith equation
H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140 131
of L combines these conditions and is expressed as:
L =
(R
n
G) +C
p
[e
(z) e(z)]/r
a
+(1 +r
s
/r
a
)
(3)
where, is the slope of the saturation vapor pressure
curve, e*(z) and e(z) are the saturation vapor pressure
and actual vapor pressure at height z, C
p
the specic
heat of air at constant pressure, the air density,
the psychrometric constant, r
s
the surface resistance
to water vapor transport and r
a
is the aerodynamic
resistance to vapor transport. can be obtained by
dividing both sides of Eq. (3) by R
n
G giving the
following expression:
=
1
+(1 +r
s
/r
a
)
+
C
p
(e
(z) e(z))/r
a
R
n
G
(4)
Eq. (4) shows that is a function of vapor pressure
decit (VPD = e
(z) e(z)), r
a
and r
s
, besides avail
able energy R
n
G.
The transfer equations for heat and water vapor be
tween the surface of the earth and the atmosphere can
also be used to express without the explicit involve
ment of R
n
G:
0
0,2
0,4
0,6
0,8
1
0 200 400 600 800
r
s
(m s
1
)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
0
0,2
0,4
0,6
0,8
1
0 200 400 600 800
RnG( W m
2
)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
0
0,2
0,4
0,6
0,8
1
0 2 4 6 8 10 12
ToTa(
o
C)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
0
0,2
0,4
0,6
0,8
1
0 5 10 15 20 25 30 35
VPD(hP)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
(a) (b)
(d) (c)
Fig. 1. Evaporative fraction as a function of available energy, R
n
G, surface resistance, r
s
, from Eq. (4) and surface and air temperature
difference, T
0
T
a
and vapour pressure decit, VPD, from Eq. (7), with the following conditions prevailing on 28th October 1998
at a grassland site: (a) r
s
= 300 s m
1
, r
a
= 70 s m
1
, VPD = 15 mb; (b) R
n
G = 300 wm
2
, r
a
= 70 s m
1
, VPD = 15 mb; (c)
r
s
= 300 s m
1
, r
a
= 70 s m
1
, VPD = 15 mb (d) r
s
= 300 s m
1
, r
a
= 70 s
1
, T
0
T
a
= 2
C.
H =
C
p
(T
0
T
a
)
r
a
(5)
LE =
C
p
(e
(T
0
) e(T
a
)
(r
s
+r
a
)
(6)
where, T
0
and T
a
are the surface temperature and air
temperature, respectively. By Further expressing as
L/(L+H) (see Eq. (1)), an alternative expression for
becomes:
=
L
L +H
= 1
1
(1 +[r
s
((e
T
0
) e(T
a
))]/
(r
a
+r
s
)(T
0
T
a
)
(7)
For ideal conditions with no cloud obstructions and no
heat or moisture advection, R
n
G, r
s
, and VPD fol
low a regular diurnal cycle. Rowntree (1991), showed
that is more sensitive to R
n
G when R
n
G is
small. Fig. 1a shows as a function of R
n
G. It can
be seen that up to a value of 200 Wm
2
, decreases
rapidly with increasing R
n
G. then remains almost
constant with further increase in R
n
G. Available
132 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
energy greater than 200 Wm
2
, usually occurs be
tween 9.00 and 16.00 h. This means that variations in
is largest in the mornings and the evenings when
R
n
G is small (<200 Wm
2
). Rowntree (1991) also
demonstrated that variations of due to r
s
are larger
for small values of r
s
(wet conditions) than for larger
r
s
values (dry conditions). Fig. 1b illustrates as a
function of r
s
. decreases rapidly when r
s
increases
from 20 s m
1
to approximately 150 s m
1
, but de
creases at much smaller rate afterwards. Because
R
n
G often exceeds 200 Wm
2
during midday con
ditions, can be expected to behave temporally sta
ble especially for moderately wet and to dry surface
conditions with r
s
larger than 150 ms
1
.
Eq. (7) introduces (T
0
T
a
) as an important vari
able in the determination of . The diurnal trend of
T
0
T
a
follows closely that of solar radiation reaching
the earth surface (K), T
0
and T
a
. Crago (1996b), il
lustrated the dependence of on T
0
T
a
for different
surface conditions. He used the formulation of Eq. (7)
for to show that is most sensitive to T
0
T
a
when
T
0
T
a
is small (<2
C). Fig. 1c shows the variation
of with T
0
T
a
. remains fairly stable for T
0
T
a
larger than 3
C. Such values of T
0
T
a
occur in the
middle of the day under clear sky conditions. Fig. 1d
presents the relationship between and VPD using
Eq. (7). increases with increasing VPD, however
increases at lower rate for VPD values larger than
10 mb. Values of VPD larger than 10 mb usually pre
vail during day light hours.
2.2. Computation of and E
In this study is derived from measurements.
is determined from the difference in vapor pressure
and temperature between the two observational levels:
=
H
L
=
dT
de
=
Ta1 Ta2
e1 e2
(8)
where, the subscripts 1 and 2 indicate the lower and
upper levels, respectively. under eld conditions is
then computed as follows:
=
1
1 +
(9)
Daytime E is calculated as
E =
t2
t1
(R
n
G)dt (10)
where the time difference t
2
t
1,
represents the time
from 8.00 to 17.00 h in the present study. Daytime E
can in a simplied manner be estimated from midday
(
mid
) and morning (
mor
) as follows:
E =
mid
t2
t1
(R
n
G)dt (11)
For
mid
and
mor
measurements conducted between
12.00 and 13.00 h and 9.00 and 10.00 h, respectively
have been used. The daily net radiation is given by:
R
n
= (1 )K +Ln (12)
where, is the surface reectance and Ln is the net
longwave radiation. K was obtained from direct mea
surements of solar radiation and Ln was evaluated
from T
a
and Relative Humidity (RH) by using empiri
cal functions (Holtslag and Van Ulden, 1983). G is es
timated as10% of R
n
during daytime hours (de Bruin
and Holtslag, 1982) and ignored on a daily basis.
3. Field experiment and study area
The study area comprises the Lake Naivasha basin
located in central Kenya (Fig. 2). Two sites, namely
Ndabibi and Eburu, with different canopy cover and
at different altitudes were selected for in situ measure
ments. The topography of the Ndabibi site varies from
slightly undulating to at terrain and is at an altitude
of 1900 m. The vegetation consists mainly of annual
grasses. The Eburu site is at an altitude of 2200 m and
being a hilly terrain, is covered by woodland and for
est. MaximumT
a
is approximately 32
Cin the months
of January and February and the minimum day T
a
of
approximately 16
C occur in July and August. RH
(midday) varies from 70% in the JulyAugust months
to approximately 10% in the JanuaryFebruary
months. The average annual rainfall at the grassland
site is approximately 700 and 950 mm in the woodland
site. There is one main rainy season between March
and June, with peaks occurring in April and May. The
driest months are January, February and December.
Two Bowen ratio towers were erected at the exper
imental sites. T
a
and RH were measured at two levels
(0.3 and 2 m) with temperature and humidity sensors
having an accuracy of 0.2
C and 1% relative humid
ity. K was measured with a pyranometer with a sen
sitivity of 0.5%. Rainfall was measured with a tipping
H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140 133
Kenya
Lake Naivasha
Basin
woodland
grassland
200 km
Nairobi
Study area
20 km
Fig. 2. Location of study area showing the grassland and woodland sites where micrometeorological measurements were carried out.
Table 1
Measured meteorological variables which were used to determine evaporative fraction and evaporation
Measured variable Height above
surface (m)
Measurement
interval
Period grassland Period woodland
Air temperature, T
a
0.5, 2 20 min 14th May 9814th April 1999 27th September 9814th April 1999
Air relative humidity, RH 0.5, 2 20 min 14th May 9814th April 1999 27th September 9814th April 1999
Shortwave incoming
radiation, K
4 20 min 14th May 9814th April 1999 27th September 9814th April 1999
Shortwave reected
radiation, K
2 1 h (once
a month)
14th May 9814th April 1999 27th September 9814th April 1999
Rainfall 0.3 20 min 27th September 9814th
April 1999
27th September 9814th April 1999
bucket rain gauge. These measurements were col
lected by a data logger and recorded as twenty minute
averages. The surface reectance, was measured 1
day in each month at 1 h intervals at both sites. Table 1
shows the details of the measurements. Malfunction
ing instruments caused a period of 36 days in February
and March 1999 with missing data for the grassland
site.
4. Diurnal stability of
The standard deviation of measured (SD) be
tween 8.00 and 17.00 h was calculated and used as an
indicator of the diurnal stability of . The mean SD
for the grassland site is 0.071 at an average of 0.40
yielding a coefcient of variation of 0.18. SD varies
considerably during the study period. The months of
MarchJune, have the largest diurnal variations with
mean standard deviation of 0.082 with minimum 0.02
and maximum 0.17 values occurring on single days.
The remaining period had a mean standard deviation
of 0.060 with a minimum of 0.01 and maximum of
0.15. For the woodland site, the mean SD is 0.045
at an average of 0.33, hence a coefcient of vari
ation of 0.14 arises. The months of March and April
had the highest SD of 0.060. At both sites the peri
ods of largest SD coincide with rainy season. Dur
ing the rainy days R
n
G, T
a1
T
a2
and VPD are
small. It was shown on theoretical basis that is
most sensitive to variations in R
n
G, T
a1
T
a2
and
VPD when these variables are small values. These af
fect the diurnal cycle of the surface energy uxes and
the stability of . In comparison, the SD of the
woodland site is much lower than that of the grass
land site. This indicates that the diurnal stability is site
dependent.
An analysis of the relationship between SD and
T
a
, RH and the degree of cloudiness was undertaken
to see if routinely collected weather data could be used
to understand the diurnal stability of . The degree of
cloudiness is more accurately expressed as a shortwave
134 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
Table 2
Relationship between the daytime standard deviation of evaporative fraction and meteorological variables used to explain the diurnal
stability of the evaporative fraction
Meteorological variables 1 day, r
2
10 day, r
2
Grassland n = 304 Woodland n = 204 Grassland Woodland
Shortwave transmittance, 0.05 0.07 0.27 0.33
Relative humidity, RH 0.11 0.10 0.31 0.21
Air temperature, T
a
0.10 0.09 0.34 0.18
transmittance ():
=
K
K
TOA
(13)
where K
TOA
is the solar radiation incident on the top
of the atmosphere which can be calculated on the basis
of standard astronomical equations (e.g. Iqbal, 1983).
Table 2 shows the coefcient of determination (r
2
) of
the relationships. The relationships were modeled by
polynomial curves having an order 2. The daily SD
has a very weak relationship with T
a
, RH and . The
relationship between 10day average SD and 10day
average T
a
, RH and was also weak (Table 2).
To examine the effect of cloudiness on the stability
of , the days were stratied according to the daily
average values and put into three groups. The groups
were dened as cloudy ( < 0.5), partly cloudy (0.5 >
< 0.65) and clear ( > 0.65). Table 3 shows that
the average SD for the three groups is almost the
same indicating that cloudiness is not related to sta
bility of . Hence, the stability of the diurnal cycle
of can not be adequately explained by micromete
orological state variables only. There is no consensus
in the literature on the effects of clouds on the diur
nal cycle of . While Hall et al. (1992) conclude that
variations in R
n
due to cloudiness should not affect
signicantly. Suigita and Brutsaert (1991), attribute
Table 3
Average daytime standard deviation of evaporative fraction grouped according to shortwave transmittance, , in order to understand the
relationship between cloudiness and diurnal stability of evaporative fraction
Number of days Mean standard deviation of evaporative fraction,
Grassland Woodland Grassland Woodland
114 40 <0.5 0.075 0.049
91 61 0.5 and 0.65 0.078 0.045
99 103 >0.65 0.068 0.041
daytime changes in to changes in cloudiness. They
attribute increase in to decrease in R
n
as clouds pass
over. Crago (1996b), observes that cloud elds tend to
change R
n
Gand surface temperature erratically and
thereby cause changes in . However, he concludes
that the effect on may not be observed in practice
as it may masked by coincident changes in RH and
wind speeds. This implies that diurnal variability of
is a complex phenomenon and other factors inu
encing the variations of in Eqs. (3) and (7) need
to be considered more carefully. The other variables
that control , are r
s
and r
a
(see Eq. (4)), of which
r
s
is the dominant surface variable, which regulates
. r
s
depends on micrometeorological variables, soil
moisture and plant physiology (Jarvis, 1976; Stewart,
1988). Surface resistance has a diurnal trend. Model
ing of surface resistance is therefore required in or
der to understand better the diurnal dynamics of ,
but considered outside the scope of the present paper
where a large divergence of time scales is discussed.
5. Relationship between midday and morning
and daytime
The relationships between
mid
and average day
time are presented in Fig. 3a and b. All days were
H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140 135
(a)
(b)
R
2
= 0.75
0,1
0,2
0,3
0,4
0,5
0,6
0 0,1 0,2 0,3 0,4 0,5 0,6
Evaporative fraction(1213hrs)
D
a
y
t
i
m
e
e
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
R
2
= 0.74
0
0
0,2
0,4
0,6
0,8
1
0 0,2 0,4 0,6 0,8 1
Evaporative fraction(1213hrs)
D
a
y
t
i
m
e
e
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
Fig. 3. Relationship between midday and daytime evaporative
fraction at (a) grassland site for the period May 1998April
1999 and (b) woodland site for the period October 1998April
1999.
used irrespective of weather conditions. There is a
strong relationship between
mid
and daily . The
r
2
for the regression lines through the origin are 0.74
and 0.75 while the root mean square error (RMSE) are
0.095 and 0.070 for the grassland and woodland site
respectively. The 1:1 line (Fig. 3a) shows that
mid
larger than 0.65 are higher than corresponding day
time values while
mid
values smaller than 0.30 are
less than the daytime values, which reveals a slight
concave type of relationship. values larger than 0.65
occur in the rainy months of May, June and April.
During these wet periods, when there is no moisture
decit, evaporation is highest at midday when solar ra
diation is highest. is therefore expected to be higher
at midday as compared to the rest of the day. In con
trast values less than 0.3 mostly occur in the dry
months of January, February and December. Evapora
tion is signicantly reduced for the whole day, how
ever available energy (R
n
G) is highest at midday.
values will therefore tend to be lower at midday as
compared to the rest of the day and under estimate the
daytime .
The relationships between average
mor
between
9.00 and 10.00 h and average daytime was deter
mined to study the potential of using satellite remote
sensing based data acquired during the morning hours.
The r
2
for the 9.0010.00 h period is lower with 0.64
and 0.65 for the grassland and woodland sites, respec
tively as compared to the midday conditions. Poorer
RMSE of 0.112 and 0.106 were also obtained at the
grassland and woodland site, respectively. The impli
cation of the results for remote sensing studies is that
midday satellite passes (e.g. NOAAAVHRR) will give
better average daily than the morning satellite passes
(e.g. Landsat).
6. Seasonal variations of actual evaporation
Daytime E estimated from
mid
and
mor
simu
late the results of E obtainable from the satellite data
with morning (e.g. Landsat) or afternoon (e.g. NOAA
AVHRR) over passes at the equator. Fig. 4 shows the
comparison of measured E and estimated E from
mid
Fig. 4. Comparison of measured evaporation, E and estimated E
by midday evaporative fraction at (a) grassland site for the period
May 1998April 1999 and (b) woodland site for the period October
1998April 1999.
136 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
Table 4
Root mean square error of E on daily, 10 day and monthly time
scales at the two sites for the whole study period
RMSE ET (mm) Grassland Woodland
Daily 0.17 0.14
10 day 0.12 0.06
20 day 0.05 0.04
for the two sites. The r
2
and RMSE are also pre
sented in Fig. 4 and Table 4 respectively. The values
of measured and estimated E compare very well at
both sites. The RMSE for daily values are 0.17 and
0.14 mm at the grassland and woodland sites, respec
tively. These results are for the whole study period,
however on individual monthly basis the largest RMSE
for daily values obtained are 0.21 and 0.18 mm for the
month of April for the grassland and woodland sites
respectively. With respect to r
2
, the lowest values are
0.77 for the month of January at the grassland site
and 0.66 for the month of February at the woodland
site. The months of January and February are the dri
est months in the year and therefore E is very small
during this period. Although the comparison between
measured and estimated E may appear poorer for the
drier months, the RMSE are comparable to the other
months. Table 4 shows the RMSE of estimated E on
daily, 10 day and monthly scales. It can be seen that
the RMSE reduces with longer time scales. This indi
cates that accumulated E is more accurate than daily
E if estimated from instantaneous evaporation. It can
also be seen that the relationship between measured
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
15Apr
98
4Jun98 24Jul
98
12Sep
98
1Nov98 21Dec
98
9Feb99 31Mar
99
20May
99
Date
E
v
a
p
o
r
a
t
i
v
e
F
r
a
c
t
i
o
n
(

)
woodland
grassland
Fig. 5. Seasonal progression of evaporative fraction at the grassland site for the period May 1998April 1999 and woodland site for the
period October 1998April 1999.
and estimated E is better than the relationship be
tween average day and
mid
. This is because more
weight is given to the midday period in the calculation
of daytime E, when R
n
G is large and is more
stable.
The daytime E estimated by
mor
gave poorer re
sults than for
mid
. The RMSE values are 0.37 and
0.29 mm at the woodland and grassland sites, respec
tively. These values are about two times larger than
those obtained when
mid
was used. The r
2
obtained
are 0.33 and 0.65 for the grassland and woodland sites
respectively. This implies that in remote sensing stud
ies, data from satellites with afternoon overpass will
give better estimate of E compared to those with morn
ing overpass.
7. Seasonal variations of
The pattern of the seasonal variation of is pre
sented in Fig. 5. Each of the points represents the
average value of between 8.00 and 17.00 h. The
seasonal variation of is a reection of the climate
of the area, in particular of rainfall and soil moisture.
Superimposed on this trend are uctuations of from
day to day caused by variations in the micrometeoro
logical conditions elucidated in the previous sections.
It can be seen that for the grassland site, drops
quickly from 0.7 at the end of May to 0.3 in approxi
mately 60 days. The reduction in may be attributed
to the reduction of soil moisture availability in the
root zone due to sharply reduced rainfall rates.
H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140 137
uctuates around 0.3 for about 100 days between the
end of July and beginning of November. This is fol
lowed by a sharp decline in , reaching virtually zero
within 45 days. This indicates that responds to soil
moisture condition when a certain critical level of
moisture and soil water potential is reached and plant
stress is triggered. Periods when is zero imply that
all of the available energy is partitioned into sensible
heat ux. There is an increase of in the month of
January from 0 to 0.4 in response to a rainfall event
(see Fig. 5). However declines to zero in a few
days. , nally increases from zero at the end of Jan
uary to 0.8 by April in the response to the rain period
starting at the end of March. Although no data is
available in the month of February and the beginning
of March, rainfall data was available. During this
period there was only 0.5 mm of rainfall recorded.
It is therefore expected that remains in the range
between 0 and 0.1 between February and March.
For the woodland site, remains fairly constant
at about 0.4 from the end of September for about 80
days. then begins to decline steadily to reach zero
in about 70 days. The declination takes a longer pe
riod as compared to the grassland site. This, may be
related to the differences at the two sites. These dif
ferences are caused by differences in rooting depth of
the vegetation at the two sites besides that the forest
receives more rainfall annually. For the grassland site
vegetation, can only get moisture from the top soil
surface and as soon as the soil surface dries vegeta
tion stress emerges. Furthermore, the grasses at this
site begin to senescence, just before the dry season.
Evaporation from soils is the dominant component of
evaporation at this time. Evaporation therefore stops a
day or two after a rainfall event. The woodland site has
vegetation with deeper roots, which can extract mois
ture from deeper soil layers. The vegetation continues
to transpire even after the surface soils have dried up
two months after the last rainfall event.
The value of , nally increases in response to rain
fall and soil moisture replenishment in early March.
However, increases to a maximum of 0.5 by the end
of April as compared to 0.8 in the grassland site. This
could be ascribed to the lower VPD prevailing in the
woodland site which causes lower degrees of parti
tioning of R
n
G into L and hence limits evaporation.
The seasonal progression of is gradual at both
sites. The implication of this for the monitoring of
is that it would be sufcient to measure say ev
ery 510 days to capture the seasonal evolution of .
Interpolation between the measurements can be done
to estimate on days when there are no mea
surements. This means that for remote sensing pro
grams processing of daily images is not necessary to
estimate the seasonal variations of for large water
sheds, albeit daily acquisition might be required to se
lect the qualitative best cloud free image for a given
period.
7.1. Estimation of by standard
meteorological data
Soil moisture dynamics and thus indirectly the rain
fall events, control the long term seasonal variations
of . The seasonal trends of micrometeorological
variables such as T
a
, RH and follow the annual
rainfall regime. These variables obtained from stan
dard weather stations could be used to estimate the
seasonal variations of , rather than the daily pro
cessing of satellite images. A regression analysis
between and T
a
, RH and was performed on
the basis of 1 and 10day average values. Multiple
linear regression between and all the three mi
crometeorological variables was performed as well.
The relationships between and the variables at
the grassland site at the seasonal scale are presented
in Fig. 6, while the coefcient of determination, r
2
,
of the relationships at the two sites are shown in
Table 5.
The maximum value of coincides with T
a
of
25
C (see Fig. 6). The optimum RH for both sites
is 50%. These agree with the optimum meteorolog
ical condition for evaporation for vegetated surfaces
found by Stewart (1988). RH best explains the av
Table 5
Relationship between evaporative fraction and meteorological vari
ables at the two sites for the whole study period
Meteorological
variables
1 day average, r
2
10 day average, r
2
Grassland Woodland Grassland Woodland
0.25 0.23 0.45 0.31
RH 0.62 0.63 0.74 0.83
T
a
0.57 0.49 0.74 0.81
RH T
a
0.64 0.62 0.82 0.83
RH T
a
0.67 0.64 0.87 0.86
138 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
y = 0.5865x + 0.6892
r
2
= 0.25
0
0,2
0,4
0,6
0,8
1
0 0,2 0,4 0,6 0,8 1
Transmission()
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
y = 0.0038x
2
+ 0.1754x  1.5256
r
2
= 0.57
0
0,2
0,4
0,6
0,8
1
10 15 20 25 30 35 40
Temperature(
o
C)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
y = 0.00x
2
+ 0.02x  0.26
r
2
= 0.63
0
0,2
0,4
0,6
0,8
1
0 20 40 60 80 100
Relative Humidity(%)
E
v
a
p
o
r
a
t
i
v
e
f
r
a
c
t
i
o
n
(

)
(a)
(b)
(c)
Fig. 6. Relationship between evaporation fraction and air temperature, relative humidity and atmospheric transmission at the grassland site
for the period May 1998April 1999.
erage day with an r
2
of 0.69. As expected, there
is an improvement in the relationships if 10day
average values are considered due to smoothing ef
fects (see Table 5). In the multiple linear regression
an r
2
of 0.67 for daily averages and 0.87 for the
10day average values for the grassland site was
obtained. It is worthwhile to note that standard me
teorological data can explain 87% of the variations
of in the 10day average values. This has im
portant implications for hydrological applications
requiring 10day average . During periods when
satellite data is not available, readily available stan
dard meteorological data could be used to estimate
empirically, once the site specic relationships are
established.
8. Conclusions
The objective of this study was to investigate the
use of the diurnal constant behavior of to estimate
daytime average and daytime total of E throughout
a complete season. The results presented show that
the diurnal stability of varies signicantly during
H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140 139
the study period. The daily standard deviation of
varies from as low as 0.01 to as high as 0.16 for in
dividual days indicating that is not stable under all
vegetation, soil and atmospheric conditions. The re
sults also show that on the daily time scale, the vari
ations of cannot be explained well by meteoro
logical variables and cloudiness alone. The variations
could be due to other causes such as the diurnal vari
ation of surface resistance and energy and moisture
advection. The evaporative fraction is more unstable
during the cloudy and rainy period (AprilJune) as
compared to the other months due to low R
n
G,
VPD and T
a1
T
a2
values. The evaporative fraction is
more temporally stable at the woodland site than at the
grassland site.
The data presented showed that there is a strong
relationship between
mid
and daytime with the
average r
2
of the regression lines through the ori
gin at the two study sites being 0.74 and 0.75. The
changes of over an annual period are gradual.
It can be concluded that for remote sensing pro
grams, an acquisition of images say every 510
days may be able to capture the seasonal evolu
tion of for large watersheds. Furthermore the
interpolation of , between remote sensing days,
can be accomplished by routinely collected weather
data.
The estimated daytime E from
mid
compare very
well with measured daytime E (RMSE = 0.17 mm,
r
2
= 0.88 for the grassland). For the whole study
period the average daily difference between the esti
mated E and the observed E was within 10%. The dif
ferences reduced even further if 10 day and monthly
integrated E values are considered. Poor E results were
obtained from
mor
(RMSE = 0.37 mm, r
2
= 0.33
for the grassland). This indicates that the use of data
from satellites with morning overpasses will give less
accurate daily E values in the environmental condi
tions of Kenya. NOAA AVHRR satellite images with
afternoon over pass are preferred although a loss of
spatial scale accuracy should be accepted. The impor
tant conclusion from this study is that the hypothe
sis of quasiconstant to estimate seasonal variations
of evaporation is valid for tropical watersheds under
general weather conditions. This provides a basis for
the use of remote sensing methods in applied regional
hydrology in tropical watersheds with data scarcity
problems.
References
Allen, R.G., Preira, L.S., Raes, D., Smith, M., 1998. Crop
evapotranspiration: guidelines for computing crop water
requirements. FAO irrigation and Drainage Paper 56.
Bastiaanssen, W.G.M., Pelgrum, H., Menenti, M., Feddes, R.A.,
1996. Estimation of surface resistance and priestley and
Taylor parameter at different scales. In: Stewart, J.B.,
Engman, E.T., Feddes, R.A., Kerr, Y., (Eds.), Scaling Up in
Hydrology Using Remote Sensing. Inst. of hydrology, 93
111.
Bastiaanssen, W.G.M., Sakthivadivel, R., van Dellen, A., 1999.
American Geophysical Union, Monograph.
de Bruin, H.A.R., Holtslag, A.A., 1982. A simple parameterization
of the surface uxes of sensible and latent heat during daytime
compared to PenmanMonteith concept. J. Appl. Met. 21,
16101621.
Brutsaert, W., Suigita, M., 1992. Application of self preservation
in the diurnal evolution of the surface energy budget to
determine daily evaporation. J. Geophys. Res. 97 (D17), 18377
18382.
Crago, R.D., 1996a. Comparison of the evaporative fraction and
PriestlyTaylor for the parameterizating daytime evaporation.
Water Resour. Res. 32 (5), 14031409.
Crago, R.D., 1996b. Conservation and variability of the evaporative
fraction during the daytime. J. Hydrol. 180, 173194.
Crago, R., Brutsaert, W., 1996. Daytime evaporation and the
selfpreservation of the evaporative fraction and the Bowen
ratio. J. Hydrol. 178, 241255.
Hall, F.G., Huemmrich, K.F., Goetz, S.J., Sellers, P.J., Nickeson,
J.E., 1992. Satellite remote sensing of the surface energy
balance: success, failures and unresolved issues in FIFE. J.
Geophys. Res. 97 (D17), 1906119089.
Holtslag, A.A.M., Van Ulden, A.P., 1983. A simple scheme for
daytime estimates of the surface uxes from routine weather
data. J. Appl. Met. 27, 684704.
Iqbal, M., 1983. An Introduction to Solar Radiation. Academic
Press, Toronto.
Jackson, R.D., Hateld, J.L., Reginato, R.J., Idiso, S.B., Pinter
Jr., P.J., 1983. Estimation of daily evapotranspiration from one
time day measurements. Agric. Water Manage. 7, 351362.
Jarvis, P.G., 1976. The interpretation of the variations in leaf water
potential and stomatal conductance found in canopies in the
eld. Phil. Trans. R. Soc. London B273, 593610.
Kustas, W.P., Norman, J.M., 1996. Use of remote sensing for
evapotranspiration monitoring over land surfaces. Hydrol. Sci.
J. 41 (4), 495515.
Moran, S.M., Jackson, R.D., 1991. Assessing the spatial
distribution of evaporation using remotely sensed inputs. J.
Environ. Qual. 20, 725737.
Nichols, W.E., Cuenca, R.H., 1993. Evaluation of the evaporative
fraction for the parameterization of the surface energy balance.
Water Resour. Res. 29 (11), 36813690.
Rowntree, P.R., 1991. Atmospheric parametrization schemes for
evaporation over land: basic concepts and climate modeling
aspects. In: Schumugge, T.J., Andre, J.C., (Eds.), Landsurface
Evapoartion. Springer, New York, pp. 529.
140 H.O. Farah et al. / International Journal of Applied Earth Observation and Geoinformation 5 (2004) 129140
Shuttleworth, W.J., Gurney, R.J., Hsu, A.Y., Ormsby, J.P., 1989.
FIFE: the variation in energy partition at surface ux sites.
IAHS Publ. 186, 6774.
Suigita, M., Brutsaert, W., 1991. Daily evaporation over a region
from lower boundary layer proles measured with radiosondes.
Water Resour. Res. 27 (5), 747752.
Stewart, J.B., 1988. Modeling surface conductance of pine forest.
Agric. Forest Met. 43, 339353.
Zhang, L., Lemeur, R., 1995. Evaluation of daily evapotranspiration
estimates from instanteneous measurements. Agric. Forest Met.
74, 139154.