You are on page 1of 22

Journal of Hydrology, 44 (1979) 169---190 169

© Elsevier Scientific Publishing Company, Amsterdam -- Printed in The Netherlands

[3]

A SENSITIVITY ANALYSIS OF THE PENMAN--MONTEITH ACTUAL


EVAPOTRANSPIRATION ESTIMATES

KEITH BEVEN
Institute of Hydrology, Wallingford, Oxon OX10 8BB (Great Britain)
(Received January 10, 1979; revised and accepted May 21, 1979)

ABSTRACT

Beven, K., 1979. A sensitivity analysis of the Penman--Monteith actual evapotranspiration


estimates. J. Hydrol., 44: 169--190.

The evapotranspiration component, based on the Penman--Monteith equation, of the


distributed Syst~me Hydrologique Europ~en (SHE) model of catchment hydrology is
briefly described. The importance of evapotranspiration predictions to the operation of
the model is stressed. However, such predictions may be expected to be in error and
ideally it would be desirable to predict the error variance associated with estimates of
evapotranspiration. This is not yet possible but the sensitivity of such estimates to errors
in input data and estimated model parameter values can be investigated. It is shown that
for dry canopy conditions at three sites within a broadly humid temperate region, the
sensitivity of Penman--Monteith estimates of evapotranspiration to different input data
and parameters is very dependent on the values of the aerodynamic and canopy resistance
parameters that introduce the influence of vegetation type into the predictions. For
forest surfaces in particular, the evapotranspiration predictions are highly sensitive to
values of the canopy resistance, so that accurate estimation of this parameter is important.

INTRODUCTION

This paper reports a study of the sensitivity of the Penman--Monteith


equation for estimating actual evapotranspiration rates to errors in parameter
values and input data. The study has been carried out as part of continuing
work on the Syst~me Hydrologique Europ~en (SHE), a deterministic distributed
hydrologic modelling system which is being jointly developed by the Danish
Hydraulics Institute, SOGREAH (France) and the Institute of Hydrology (U.K.).
A brief outline of SHE is given here to place the present study in context.
The SHE model is physically-based in that it has been developed from
equations of flow for the processes of overland and channel flow, unsaturated
and saturated subsurface flow. Finite-difference methods are used to obtain
solutions to these non-linear flow equations. With present computer limita-
tions, it is not economically viable to produce an operational model fully
three-dimensional in space (see Freeze, 1978) that would allow the required
accuracy of discretisation in both horizontal and vertical planes. On the
170

assumption that in the unsaturated zone vertical flow is far more important
than lateral flow, the model has been rationally simplified such that indepen-
dent one-dimensional unsaturated flow components of varying depth are
used to link a two-dimensional groundwater flow component and a two-
dimensional surface-water flow component (Fig. 1). This structure has
important implications for the modelling of other components of the
hydrological cycle. It is simple, for example, to model spatial variations in
evapotranspiration rates due to changes in moisture conditions and vegetation
assemblages between different grid squares of the model. Moreover, it
becomes possible to attempt to model the effects of localised vegetation and
land-use changes over time in some more rigorous physically-based manner
than has been hitherto feasible using lumped catchment models.
The evapotranspiration component of SHE is necessarily of particular
importance since in many catchments in varied climatic zones, evapotranspira-
tion is the dominant hydrological process in volume or water balance terms.
Even when the response to storm rainfall of a catchment may be little af-
fected by evapotranspiration during the period of a storm, losses to the
atmosphere may be extremely important between storm periods insetting
up the initial soil-moisture conditions prior to the next event.
In developing the SHE model it has been explicitly recognised that, for a
model that is expected to have widespread application, it must always be the

R~modelzone ~z

Fig. 1. A diagrammatic representation of the Syst~me Hydrologique Europ~en (SHE)


model.
171

aim to make maximum use of the (often limited) data that will be available
for a given site. Thus it was felt that no clear-cut decisions as to the most
suitable evapotranspiration model could be taken independent of the
application. Thus flexibility has been retained in the model programming,
with a hierarchy of evapotranspiration components requiring different levels
of data availability and parametric input, but consequently with different
degrees of theoretical acceptability. At the top of this hierarchy, as the most
complex and physically realistic model considered for the purposes of SHE,
is the Penman--Monteith combination equation (Monteith, 1965) for the
prediction of actual evapotranspiration rates.
However, the choice of an evapotranspiration model is not simple. The
complexities of the processes of evapotranspiration and its interaction with
soil moisture are paramount. The current generation of predictive models of
these processes, including the Penman--Monteith equation, are broadly
physically-based, but remain simplistic and subject to important limiting
assumptions. The parameters of these models exhibit significant variations in
both space and time, and whereas there are now a reasonable number of
studies in which parameter values have been derived from measurements of
evapotranspiration, studies of the p r e d i c t i o n of parameter values in other
situations have been notably lacking. The evapotranspiration component
shares these problems with other model components, but in the case of
evapotranspiration the problems are compounded by the scale of variations
in time and space, and the physical scale of measurement studies, which are
both small relative to catchment scales of interest. Yet, the importance of
accurate predictions of evapotranspiration rates cannot be underestimated.
The use of a distributed model in itself goes some way towards diminishing
these problems but, in accepting the limitations of available evapotranspiration
models, an assessment of the effects of errors in both the measurement of
input data and the estimation of parameter values becomes of particular
importance. The aim of such a sensitivity analysis is to make clear what
range of accuracy is required for an input variable or parameter value, and
consequently where the greater effort should be expended in measurement
or model calibration. The acceptability of measurement or extrapolation
techniques, and the confidence in model predictions should reflect the inter-
pretation of sensitivity estimates reported from studies such as that in this
paper.

THE PENMAN--MONTEITH EVAPOTRANSPIRATIONEQUATION

The Penman--Monteith equation is a one-dimensional single-source model


of the evapotranspiration process. The assumptions on which it is based are
described in the development by Monteith (1965). Within the limitations of
these assumptions actual evapotranspiration (Ea) is predicted b y :

A s A + Pcp(qw, TD - q)/r a
E a ~--- (kg m -2 s -l ~ m m s-I ) (I)
~.[As + (Cp/~.)(1 + rc/ra)]
172

where
= latent heat of vapourisation of
water (= 2.47.106 J kg -l)
Cp = specific heat of air at constant
pressure (= 1.01 103 J kg -l °C -I)
P = density of air (= 1.2 kg m -2)
A = available energy given by A =
RN - G (W m -2)
RN = net radiation measured at the
reference height, z (W m -2)
G denotes the sum of energy fluxes
into the ground, to adsorption by
photosynthesis and respiration
and to storage between ground
level and z (W m -2)
qw, TD = saturated specific humidity at
dry-bulk temperature, TD (kg kg- ~)
(qW,TD - q) = specific humidity deficit (kg kg -~)
SHD = derived from measurements of
TD and wet~bulb temperature
depression, DEP
A s = slope of the specific humidity/
temperature curve between the
air temperature TD and the
surface temperature of the
vegetation T s (kg kg -1 °C-1)
ra = aerodynamic resistance to the
transport of water vapour from
the surface to the reference
level z (s m -l)
and
re = (Monteith) canopy resistance to
the transport of water from
some region within or below the
evaporating surface to the surface
itself, and is expected to be a
function of the stomatal resis-
tance of individual leaves. Under
wet-canopy conditions rc = 0 (s m -x)
Eq. 1 continues the one-dimensional vertical structure of SHE and assumes
further that all evapotranspiration within the complex soil--vegetation canopy
system takes place from a single representative source layer. Stewart (1979)
describes the development and assumptions underlying the Penman--Monteith
equation and its relationship to other similar evapotranspiration models. The
173

meteorological data required for the model are values of A (or RN if G is small),
TD and DEP (= TD -- TW ) where Tw is a wet-bulb temperature. The para-
metric data required are values of ra and rc.

PARAMETRIC AND METEOROLOGICAL INPUT DATA

This study has been designed to allow consideration of two vegetation


types and a range of meteorological conditions within a broadly temperate
maritime climatic regime. Data from three meteorological sites maintained
by the Institute of Hydrology, Wallingford, have been used, spaced across
England and Wales from west to east (Fig. 2). At each site one or two
automatic weather stations (AWS), as developed by the Institute of Hydrology
(Strangeways and McCulloch, 1965), collect data on incident solar radiation,
net radiation, air temperature, wet-bulb temperature, wind run and direction,
and rainfall on a 5-min. time base. The original records are routinely validated
and processed into the mean hourly values that are used in the present study.
The input data required by the Penman--Monteith equation can all be supplied
by the standard AWS measurements with the exception of G, which has been
neglected in the sensitivity analysis. Values of A s and SHD are calculated
from dry-bulb (TD) and wet~bulb (Tw) temperature measurements from
equations given in Appendix A. The characteristics of each measurement site
are as follows.

Fig. 2. A location diagram for the three meteorological stations used in this study. 1 =
Carregg Wen, Plynlimon; 2 = G r e n d o n U n d e r w o o d ; and 3 = Thetford Forest.
174

S i t e 1 - - P l y n l i m o n , Carreg Wen. This station lies at an altitude of 575 m


A.O.D. and is the highest of the meteorological stations in the Institute of
Hydrology's experimental catchments at Plynlimon, Wales. Vegetation at the
site is an upland grassland community.

Site 2 - - G r e n d o n U n d e r w o o d . The Grendon Underwood site lies in the River


Ray experimental catchment in the Oxford clay vale. The altitude of the site
is 67 m A.O.D. and the vegetation is short grass pasture.

Site 3 - - T h e t f o r d Forest. The site at Thetford Forest, Norfolk, was main-


tained as part of the Institute of Hydrology's Thetford Forest evaporation
project. The site is at an altitude of 47 m A.O.D. and is above the canopy of
Scots and Corsican pine (see Stewart and Thom, 1973).

400

Z~
A
~E o n
o
0 x
X

Z~
£
i I i I

30

2O
oo
x
x x

10

i I i i 1 i i
M J J A S 0 N

0 0
Z~
L~
,,X
4 0
X

0
X z',
£3 2

x
i i
3 J A
Fig. 3. Mean m o n t h l y i n p u t data for the three meteorological sites. Mean o f all hours w i t h
positive net radiation in each m o n t h , M a y - - N o v e m b e r 1 9 7 6 . x = Carreg Wen, P l y n l i m o n ;
o = G r e n d o n U n d e r w o o d ; and ~ = T h e t f o r d Forest.
175

Seven months of fully processed data were available for all three sites,
covering the period May--November 1976, and including the abnormally dry
British summer of that year. The calculation of evapotranspiration estimates
and the sensitivity analysis described below was carried out for all hours with
positive net radiation and a complete set of satisfactory meteorological
measurements, giving 2406 hr. for the Plynlimon site, 2056 hr. for Grendon
Underwood and 2191 hr. for Thetford Forest over the 7-month period. The
meteorological input to the present study for the three sites is summarised
and compared in terms of mean monthly values and mean hourly values over
the complete study period in Figs. 3 and 4. Fig. 3 shows some interesting dif-
ferences between the sites. The lowland sites are consistently warmer than
the upland Plynlimon site, with similar mean monthly temperature curves.
However, the central Grendon Underwood site has much lower mean net
radiation than the Thetford site, close indeed to the Plynlimon site (which

400 I [ I

A
A A
A
A
A
IE 00
A OXx z~
200 o x £
Z~
E A
0 ~ £
x
a~

X 0

!
0 6 12 18 24Hour

30 I I

2O
0
o

10
x~ x xxx~xxxx
:~ 0 0 0

I I t I I
6 12 18 24 H o u r
i

o
6i,o ? '
4
O
O@
' eOOOg .

2 0 xxxxxX XXo
0 x

oo ~ x x * *~oO°
0 xxx^ ] I I xxXl
0 6 12 18 24Hour

Fig. 4. Mean hourly input data for the three meteorological sites. Mean of hours with
positive net radiation, May--November 1976. Site key as Fig. 3.
176

must reflect the nature of the surface over which the measurements were
made) b u t the wet-bulb depression and consequently the specific humidity
deficit are both higher.
Fig. 4 demonstrates the mean diurnal pattern of change at the sites and
shows h o w dry-bulb temperatur6 wet-bulb depression and consequently A s
and SHD all continue to rise throughout the afternoon following the peak in
net radiation at 12h00 m GMT. The mean temperature characteristics of the
two lowland sites correspond quite closely, with the upland site having lower
air temperatures and wet-bulb depression.
The parametric data required essentially consists of the two resistance co-
efficients r a and re. Both resistances are expected to vary over time in some
complex way. One form in which a variable ra may be calculated is from:
[ln((z - d)/zo}] 2
ra = k2 u

where
u = mean wind speed (m s -l)
z = reference height of the anemometer (m)
d = zero plane displacement (m)
Zo = roughness length (m)
k = v o n Karman's constant (= 0.41)

This form was originally used by Penman and Long (1960), Monteith and
Szeicz (1962) and Van Bavel (1966). However, there is evidence that b o t h
z0 and d themselves vary with wind speed (e.g., Szeicz and Long, 1969), and
also that the improvement in model predictions b y allowing ra to vary with
wind speed, as opposed to using a constant value may be small (Calder, 1977).
For the purpose of the present study, constant values have been used, chosen
to be representative of grass and forest vegetation surface. The values used
were 46 s m -1 for the former surface (Thorn and Oliver, 1977), and 4 s m -1
for the latter (Calder, 1977).
Values of rc are known to have a diurnal variation (see, e.g., Van Bavel,
1967; Szeicz and Long, 1969; Stewart and Thorn, 1973; Tan and Black, 1976)
and there is evidence that they may also show important seasonal variation
(Calder, 1977). Diurnal variations in mean hourly Values of rc relative to an
assumed mid-day value for the grass and forest surfaces have been assumed
on the basis of measurements given in Szeicz and Long (1969), and Stewart
and Thom (1973) as shown in Fig. 5. The relative diurnal distribution has been
assumed constant over the s t u d y period and for both surfaces. Changes in
surface resistance as a result of a w e t canopy following rainfall (see, e.g.,
Stewart, 1978) have also been ignored for present purposes.
177

' I I I /
400 -,, //"

350 - \X I/
~ g 11
~oo- ', III I

250-
','|
/11/
Forest I I ) I
c , IV ,'
,oo1-. ', if,; ..

,~ol-
/",, \, ',,\',/ I/,"
I/I ,' ,'
,-'
~ . . , " !
too ~ /'-~--~--:-" /
/ /" Grass

5o

o I I I I
6 12 18 24Hour

Fig. 5. T h e a s s u m e d daily v a r i a t i o n in r c relative t o a m i d - d a y ( 1 2 h 0 0 m - - 1 3 h 00 m ) value o f


50 s m -1 for forest, in c o m p a r i s o n w i t h d a t a f r o m Szeicz a n d L o n g ( 1 9 6 9 } for grass,
S t e w a r t a n d T h o r n ( 1 9 7 3 ) for p i n e forest.
D o t t e d l i n e s = a s s u m e d values; s o l i d l i n e s = m e a s u r e d values.

SOURCES OF MODEL ERROR VARIANCE

We must accept that applications of the evapotranspiration c o m p o n e n t


of the SHE model will be subject to errors (perhaps significant) in both input
data and parameter estimation. Such errors must result in an error variance
associated with the estimates of evapotranspiration. At least four sources of
error contributing to the estimation error m a y be identified.

I n s t r u m e n t error

It is expected t h a t every measurement that forms the input data to the


model will be subject to error, such that:
rni(t) = m i ( t ) + E l ( t )
where ~ni(t ) is the ith recorded measurement at time t, mi (t) is the true
measurement, and E i ( t ) is an error term. In general E i will be a random
variable, but may be biased due to errors in instrument settings, etc. This
bias may show a trend over time due to instrument drift, with sudden changes
as the instruments are reset.
178

S i t e errors

When applying the evapotranspiration model over a catchment area we


may have meteorological measurements at only one site (possibly outside the
area of interest) which m u s t be taken as representative of the whole catch-
ment. Where there are satisfactory measurements at more than one site, we
must use some interpolation procedure to estimate evapotranspiration at
points within the catchment. In either case, evapotranspiration estimates may
be in error relative to the true amounts from the whole catchment, or an
elemental grid square. This may be particularly true under advective condi-
tions (see, e.g., McNaughton, 1976), and for measurement sites poorly
situated in respect of vegetational changes in the area. Site errors may exhibit
bias as well as a random c o m p o n e n t and m a y show correlation in both space
and time.

M e a s u r e m e n t m o d e l errors

Consider now the measurement (estimation) of evapotranspiration in the


field. Errors in the field estimates will be due to both instrument errors and
measurement model errors when the measurement model (say the water
balance equation or Bowen ratio estimation) is inappropriate to the situation
in which it is used and does n o t conform to the true nature of the physical
process. Such errors may have important consequences for the later develop-
mert of error variance in the evapotranspiration component, since the param-
eter values used will be based on estimates from field experiments. Again it
may be expected t h a t measurement model errors m a y have bias as well as a
random component, with a dependency structure over time.

M o d e l l i n g errors

Where a model such as the Penman--Monteith equation is being used to


fit measured evapotranspiration rates, it is (for this example) always possible
to alter the parameter values to fit the measured rates exactly at each t i m e
step. However, it may then be extremely difficult to predict the changes in
parameter values over time to again obtain a good fit under any given condi-
tions. In the general application of the model, parameter values must be
predicted a priori and it is expected t h a t the model will subsequently be in
error. Due to diurnal and seasonal variations in evapotranspiration this will
be particularly true when estimates of average parameter values are the best
that are available. Errors may then involve diurnal and seasonal time
dependency. Considerable reductions in apparent model error variance m a y
be achieved by optimising functions for expected parameter changes against
measured evapotranspiration rates (see Calder, 1977). However, the number
of such studies is so small as to make general application impossible at present.
At present we have little or no knowledge of the characteristics of the
179

error variance associated with the estimates of evapotranspiration. On the


basis of a consideration of the contributing sources of variance, we cannot
assume that the individual error terms are stationary, independent and non-
autocorrelated (with the possible exception of pure instrument error).
Ultimately it is desirable to predict the evolution of the error variance
around estimates of cumulative evapotranspiration as it changes over time,
and the significance of the expected errors in the overall predictions of the
model. This aim cannot yet be achieved but it is possible, using the techniques
of sensitivity analysis, to make an assessment of the relative importance of
the primary operational errors in input data and parameter values. This is the
aim of the present study.

MODEL SENSITIVITY ANALYSIS

Consider first the sensitivity of an estimate of actual evapotranspiration,


Ea, calculated using the Penman--Monteith equation to changes in a param-
eter or input data variable, Pi, where in general:
Ea = f(Pl,P2,P3,... ,PN)
where N is the number of parameters and input data variables. Then:
E a + A E a = f ( p l + A p l , p 2 + A p 2 , . . . , PN+APN) (4)
Expanding eq. 4 in Taylor series and ignoring second-order' terms and above
leads to:

AEa = ~~Ea Apl + aEa Ap2 + . . . + ~Ea


-- Ap N (5)
~Pl ~P2 aPN
where the differentials aEa/aPi define the sensitivity of the estimate to each
parameter or variable. These sensitivity coefficients are themselves sensitive
to the relative magnitudes of E a and the Pi and following McCuen (1974) a
non-dimensional relative sensitivity may be defined as:
~Ea Pi
Si - (6)
~Pi Ea
Si now represents t h a t fraction of the change in Pi that is transmitted to
change Ea, i.e. an Si value of 0.1 would suggest t h a t a 10% increase in Pi may
be expected to increase Ea by 1%. Negative coefficients would indicate t h a t
a reduction in Ea will result from an increase in Pi. Note t h a t both sensitivity
coefficients may vary from time step to time step being dependent on the
current value of all the Pi and the value of Ea.
Eq. 6 remains sensitive to the magnitudes of E a and Pi. In particular, the
relative sensitivity coefficients Si may n o t be a good indication of the signif-
icance of the Pi if either E a or Pi tend to zero independently, or if the range
of values taken by Pi is small in relation to its magnitude. Parameter sensitivity
180

analyses of other evaporation and evapotranspiration models have been pub-


lished by McCuen (1974), Saxton (1975), and Coleman and DeCoursey
(1976). The sensitivity coefficients for the Penman--Monteith equation used
in this study are as follows:
aEa RN 1
SRN aRN E a 1 + pcpSHD/AsRNr a (7)
aE a r a 1 1
Sra - - - (8)
ara Ea 1 + (As+Cp/X) raX 1 + RNAsra/pcpSHD
cpr c
bEa rc Cprc
Sr c -
arc Ea XAsra [1 + cp(l+rc/ra)~/ (9)
L 'J

= TO = TD [ raRN(aAs/aTD) + pCp(aSHD/aTD)
STD aTD Ea raAsR N + pcpSHD

aAs/aTD ]
- A s + Cp(1 +rc/ra) (10)

aE a D E P DEPpcp(aSHD/aDEP)
SDEP . . . . (11)
aDEP Ea {raAsR N + p c p S H D }
where the differentialsaAs/aTD, a S H D / a T D and a S H D / a D E P are derived
from the functional relationships of Appendix A and are given in full in
Appendix It D E P rather than T w was chosen for use in the sensitivityanalysis
because the value of T w includes an implicit dependence on TD. D E P m a y be
considered as independent of TD (although the consequent calculation of
atmospheric humidity is not).
The analysis of the sensitivityof individual parameters and variables can
be extended to an analysis of the error variance in the Pi values, but only
under restrictiveassumptions. Following Hahn and Shapiro (1967) for a
system:
Z = h(xl,x2,... ,XN)
in which the parameters and variables, x, are uncorrelated, the error variance
of mean system performance is given by:
N N
E[var(z)] = ~ (ah/axi) 2 var(xi) + (ah/axi) (a2h/axi2)p3(xi) (12)
i=l i=1
where p3(xi) is the third central m o m e n t of the distribution of xi, and the
181

differentials are evaluated at their expected values. This expression was used
by Coleman and DeCoursey (1976) to explore the error variance of several
models arising from instrument measurement errors alone, assuming that
those errors are independent, that the second term of the RHS of eq. 12 is
negligible relative to the first and using some average sensitivity differential
determined by simulation.
Hahn and Shapiro (1967) extended the analysis to variables that are cross-
correlated. However, in b o t h cases there are problems in application for the
present study, due to the lack of knowledge a b o u t the distribution of the
residuals, and the effects of possible autocorrelation structures on the error
variance of cumulative evapotranspiration rates. In the face of other sources
of error, variance due to instrumentation errors when the instruments are
working normally, may be expected to be relatively small.

RESULTS OF MODEL SENSITIVITY ANALYSIS

Sensitivity coefficients for a grass surface

Sensitivity coefficients, as defined b y eqs. 7--11 have been calculated on an


hourly basis using data from the three meteorological stations and parameters
representative of a grass surface as given in the section "Parametric and
meteorological input data" above. These sensitivity coefficients exhibited
considerable variation over time. In particular values during the dusk to dawn
periods, when calculated values of Ea are small, were erratic. This is reflected
in the mean hourly values of the sensitivity coefficients for the grass surface
over the full study period plotted in Fig. 6 together with actual evapotranspira
tion values predicted by the Penman--Monteith equation. The predicted
evapotranspiration increases in value from .the upland Plynlimon site in the
west to the Thetford site in the east.
The mean hourly sensitivity coefficients all exhibit a diurnal variation, b u t
not necessarily following the pattern of change in the calculated evapo-
transpiration rate. The differences in mean sensitivity coefficients between
sites are generally smaller than the differences in evapotranspiration rates
relative to the diurnal range, and the relationship between the sites is also
different with the Thetford (higher evapotranspiration) site generally lying
between the other t w o on the sensitivity plots.
In the case of Src, the values are uniformly negative as befits its operation
in the denominator of the Penman--Monteith equation. However, the pattern
of diurnal change in Src is not consistent between the sites becoming more
negative during the day at the Carreg Wen site and less at the Grendon
Underwood site. The magnitude of Sr c suggests that the influence of the
canopy resistance in reducing predicted evapotranspiration below some
"potential" value is restricted, as would be expected for grass surfaces where
the noon-tide values of rc/ra under the conditions of the analysis are close to
unity.
182
mmS"

aoOO ° t ' n Ea ST x Sr

15 0 ~ ~n~

!
~0°9 ' O;Oo~ | o o 2 Oooo
6 12 18 24Hour 0 6 12 18 24H~r 6 12 18 24 Hour
T0 1
o o o o'
o
o
o ~ S°[ p O
, haaaaaa
o o o oo 06
oOOOOoa~, o oa o •
05 I 80 °~x o ~ * ~ OOoo
oa
© o a ~, *£°OoooO~
O Sr~

xa
oo O Oo o
,o , i
6 12 1/8 6 24 Hour 6 12 18 24Hour -1 6 12 18 24HOUr

Fig. 6. Mean hourly values of predicted actual evapotranspiration and sensitivity coeffi-
cients for a grass surface. Mean of hours with positive net radiation, May--November 1977.
Site key as Fig. 3.

The changes in Sr a between positive and negative values are a result of the
occurrence of ra in both numerator and denominator of the Penman--
Monteith equation so that the bulk sensitivity to ra will depend on the relative
importance of the radiation and aerodynamic terms in the numerator. It is
clear from Fig. 6 that the aerodynamic term has greater significance during
early and late parts of the day when net radiation is low, and that it is relative-
ly more important at the Grendon Underwood site, there suppressing the mean
values of Sr a to below zero throughout the day.
For all the sensitivity coefficients, values for the mid-day hours, when
evapotranspiration is highest, are relatively stable. Cmsidering daytime values
alone, the evapotranspiration estimates are most sensitive to RN, with all the
other coefficients generally falling below a mean value of 0.5 close to mid-
day. This confirms that for grass surfaces under these conditions, the radiation
term is generally dominant over the aerodynamic term in the prediction
equation.
Given the diurnal variability of the sensitivity coefficients, mean mid-day
values ( 1 2 h 0 0 m - - 1 3 h 0 0 m) have been used to investigate the seasonal change
in the sensitivity of evapotranspiration rates to parameters and input data.
The mean values together with their standard deviations are plotted for the
three sites in Fig. 7 and show that seasonal variation in the sensitivity coeffi-
cients is small relative to the change in predicted actual evapotranspiration.
The sensitivity to RN remains the highest, while those to the resistance coeffi-
cients Sra and Src show the greatest difference between sites. In both cases,
the Thetford and Grendon sites plot closely together while the Plynlimon
site, where the sensitivity to rc is particularly high, differs.
183

m,nS
0002 05'

0 ST x
x

Ea x x St.
x
×
o o
0001 0"5 X X X a
0 o

x × x o o
x
x
x

o[
x

J j k ~ 6 o ~ j j k g 6 O.5
M J J A S O
~0

Sn~ I°I SDe~ S rc

x x x x

o
o o o o
0"5 o. - 0.5 X × X
O5 X
o- x X
O O
o o
o ~ ~
~ x ~
I × x x x

j j k. ~ 6 ~ ol a .~ ] k g 6 J ; k s b

Fig. 7. Mean m o n t h l y mid-day ( 1 2 h 0 0 m - - 1 3 h 0 0 m ) values of predicted actual evapo-


transpiration and sensitivity coefficients for a grass surface, M a y - - N o v e m b e r 1976.
Site key as Fig. 3.

Sensitivity coefficients for a forest surface

The sensitivity calculations were repeated using parameters representative


of a forest surface, as given in the Section "Parametric and meteorological
input data". The mean hourly estimated evapotranspiration and sensitivity
coefficients are shown in Fig. 8. Evapotranspiration rates are not generally

mrnS-'
,0001 1.C 1,0 !

S% Sra
aa~a
t,~o o°o
0 x

05 ~aao ~°° o
o o~ x x
o o~
o~ x x XXxxxx x Xx x
oo IoX~ x x x x
~××~xx × 8
X 4~6
Oom
Xx x
x o
x
x ~
XooooOooooo~a a
Xx

x x 00004 0~0000
I
6 112 1B 24 Hour 6 112 24 Hour 112 118 24 Hour

OO8o oooO°~ £2 soEp S%


SR N o Oooo ~a x
~ z ~ x x X
o x xxX x x x
x XxXXX
x
x o |
xx X
x x x o x
x xXxxxxx x
x xx x x
x o
x o
x x Xo~o° °°Ooo~ ~ Xx
{
°Sa~
,'~ ,'~ ~.oor - ,o1~
1
6
I
12
~IilI18 OOOOOi
24 Hour
6 12 1B 24 Hour

Fig. 8. Mean hourly values o f predicted actual evapotranspiration and sensitivity coeffi-
cients for a forest canopy. Mean of hours with positive net radiation, May--November,
1976. Site key as Fig. 3.
184

as high as those for the grassland. [Note that dry-canopy conditions are
assumed throughout. For a discussion of the significance of the evaporation
of intercepted water to total losses of forests see Stewart (1977)i] The
sensitivity coefficients show the same erratic nature during the dusk to dawn
period but during the day show more regular diurnal pattern in the mean. It
is interesting to note that SRN is n o w quite low with a diurnal distribution
having higher morning values, whereas STD and S D E P are n o w much higher
with higher values in the afternoon. Values of Sra remain small, though with
less scatter than for grass, but those for Src are uniformly high at all three
sites with mean daytime values greater than 0.9, rising further during the
afternoon.
Mean mid-day values have again been used to demonstrate the seasonal
drift in the sensitivity coefficients (Fig. 9). Only STD shows any marked
seasonal change in this case, but the relative significance of the value of Src
and SDE P is reinforced. These results for the forest canopy suggest that given
the low aerodynamic resistance of the forest, the aerodynamic term of the
combination equation dominates the radiation term but that the estimated
evapotranspiration rates are significantly controlled by the canopy resistance
under dry-canopy conditions.

mmS-'
0~2
I.O S r.
Ea $I
o
x

o
o ×
a ~ a o
o
x x x
o
o
~ 8 ~ o o

J j k ~ 6 j j k i 6 J J k b N
oI
o o
o ~
Sr~
SR N

-05
x

x
x Soe p

o o o o ~ ~ ~ ~
-I
6

Fig. 9. Mean monthly mid-day ( 1 2 h 0 0 m - - 1 3 h 0 0 m ) v a l u e s of predicted actual evapo-


transpiration and sensitivity coefficients for a forest canopy, May--November, 1976.
Site key as Fig. 3.

DISCUSSION

The results of the previous section show that within a broadly humid
temperate climate, .although the three sites differ considerably in the range
of meteorological conditions experienced, and consequently in the predicted
185

evapotranspiration rates, the sensitivity coefficients do n o t differ significantly


between sites. It would appear t h a t vegetation type, as introduced through the
resistance parameters of the P e n m a n - M o n t e i t h equation, has a much more sig-
nificant effect. This is n o t reflected in the sensitivity coefficients for the aero-
dynamic resistance since it appears in both numerator and denominator of
the P e n m a n - M o n t e i t h equation and the sensitivity values are consequently
generally low. However, the sensitivity to the specific h u m i d i t y deficit of the
aerodynamic term, which is closely reflected by SDEP,will also show the
influence of the value of r a.
However, it is the values of Src that show the greater difference between
grass and forest surfaces, being especially high in the case of the forest, when
a 10% error in rc will result in a 9% or more error in predicted Ea. This
assumes particular importance since values of rc may be difficult to estimate
accurately. It is accepted that values of rc may be complex functions of other
variables that have yet to be established satisfactorily (see, e.g., Jarvis 1976;
Tan and Black, 1976; Calder, 1977).
Further runs of the sensitivity analysis program with different values of ra
and rc were made to evaluate the importance of the two resistance parameters.
The data used were mean mid-day 12h00m--13h00 m August values of the
input variable at the central Grendon Underwood site. The results are shown
In Fig. 10 (note t h a t dry-canopy conditions were again assumed throughout).
On the basis of the mean mid-day wind speed for August 1 at the site and
assuming that (from Maki, 1975):
z = h + 2 (m); z = 0.1 h (m); and d = 0.7 h (m)
where h is mean vegetation height, an equivalent height scale has been added
to the r a scales of Fig.10.
Fig. 10 allows the regions of ra and rc for which the prediction of evapo-
transpiration is particularly sensitive to net radiation RN, dry-bulb tempera-
ture, TD, wet-bulb depression, DEP, or canopy resistance, rc, to be identified.
Although specific values of the sensitivity coefficients will change, the evidence
of this paper suggests that these regions, representing the changing balance be-
tween the aerodynamic and radiation terms of the combination equation,
will be representative of a wide range of humid temperate areas under the
summer daytime conditions when evapotranspiration rates are highest.
The results presented in this paper may be compared with those of other
sensitivity analyses of combination-type evapotranspiration equations, al-
though in all the published cases available, only the prediction of " p o t e n t i a l "
or water surface evapotranspiration rates has been considered. This is equiv-
alent to the case of high r a and zero rc in the Penman--Monteith equation.
Both McCuen (1974) and Saxton (1975) defined a relative sensitivity coeffi-
cient similar to that used here and analysed different forms of the Penman
equation. The results of Saxton, who used daily data from Iowa for surfaces
of corn and grass, were similar to those obtained for a grass surface in this
study. McCuen analysed the Penman equation for evaporation from open-
water surfaces for a number of stations over a wide range of climates in the
186

r~ ($m-') ro (st.")
? ;. . . . ,o ,,oo ,~o ,~.... ?
20
40
°°:t 4O
~ 6o
oos eo

,6C
/ ~ Ea mm h r -~ OOl
14o

rc ( s i n - ' ) ,~ (sr'.-')
2,o ,,o ~,o 8,o ,?o ,p ,~o ,6o 210 ,,0 ~,o ~.~,o ,?° ,~,0 ~I0 160
o.2 ios
i
2O o-6
o.4 os
40 40

E
6O C
~ so 80i
oo~ ~,ool

o
SR N

20 40
1 "
60
00~

rc ( s m -~)
OO 100
12C
14C

160

120 140 60
0-2
SDEP ool

I I I I I I i
20

40

'E eo 0'05

001

Fig. 1 0 . T h e c h a n g e in p r e d i c t e d e v a p o t r a n s p i r a t i o n a n d s e n s i t i v i t y c o e f f i c i e n t s w i t h r a a n d
r e for m e a n A u g u s t m i d - d a y c o n d i t i o n s at t h e G r e n d o n U n d e r w o o d site. ( R N -- 2 6 2 . 3 9 W
m - 2 ; T D -- 2 2 . 5 5 ° C ; D E P -- 6 . 9 6 ; v a l u e o f u = 2 . 5 2 m s -1 u s e d in c a l c u l a t i o n o f h - s c a l e . )

U.S.A. Mean summer meteorological conditions only were used. The results
suggested that radiation, humidity and temperature could all show very high
sensitivity coefficients.
Coleman and DeCoursey's (1976) definition of the sensitivity coefficient
was based on a different form of eq. 6, where the approximate differential is
evaluated using an (undefined} perturbation of the variable concerned. A long
period (16 yr.) of daily U.S. Weather Bureau data from Oklahoma was used.
The sensitivities defined in this way are not directly comparable but Coleman
and DeCoursey's results showed that the Penman equations for water surfaces
analysed were equally sensitive to air temperature, relative humidity and solar
radiation; the sensitivity balance between these variables changing over the
year. Air temperature and solar radiation were the most important in the sum-
mer. Thus comparison with the present results reinforces the view that the
introduction of the influence of vegetation directly into the prediction of
sensitivity, changes the sensitivity balance markedly.
Sensitivity to equation parameters was not analysed in any of these three
previous studies.
187

CONCLUSIONS

The most important conclusion of this w o r k is that the sensitivity of Pen-


man - M o n t e i t h estimates of actual evapotranspiration to different input
variables and parameters is more dependent on the values of the aerodynamic
and canopy resistance parameters that introduce the influence of vegetation
t y p e into the equation, than on climatic difference between sites in a broadly
humid temperate area. Dependent on these parameters, all the input variables
exhibit high daytime sensitivity coefficients during a summer period when
evapotranspiration rates are high. Thus accurate measurement of these
variables is essential to accurate predictions of evapotranspiration. This will
generally not be difficult with the temperature variables TD and DEP b u t the
energy balance term, which has here been assumed to be adequately char-
acterised by measured net radiation, R N , will in general involve other energy
fluxes (see eq. 1) that may be difficult to quantify.
However, such considerations may be reduced in significance by losses in
accuracy due to errors in the estimation of the equation parameters r a and rc.
Under conditions when the values of these parameters are markedly different
(e.g., for forest surfaces) evapotranspiration estimates will be particularly
sensitive to the canopy resistance rc. It has already been noted that this param-
eter may n o t be easy to estimate accurately.
However, for vegetation surfaces that allow significant amounts of inter-
ception of rainfall, evapotranspiration rates may be highest of all under wet-
canopy conditions that have been neglected in the present study. The use of
a model to predict evaporation from intercepted rain water [such as that of
Rutter et al. (1975)] which may take place at a "potential" rate (rc = 0) will
reduce the overall sensitivity to values of rc. This will be examined further in
a later paper analysing the joint interception- -evapotranspiration--unsaturated
zone c o m p o n e n t of the SHE model.

ACKNOWLEDGEMENTS

This study would n o t have been possible without the careful work of the
staff of the Institute of Hydrology concerned with the collection and pro-
cessing of the automatic weather station data. I should especially like to thank
Richard Harding and Dave Woolhiser who helped in the early stages of the
study, and John Stewart and Howard Oliver who provided valuable comments
on earlier drafts of this paper. Our colleagues on the SHE project at the
Danish Hydraulics Institute and SOGREAH also contributed useful discussions.
The paper is published with the permission of the Director of the Institute of
Hydrology.

APPENDIX A

Given input data values of dry-bulb temperature, TD, and wet-bulb depres-
188

sion DEP = (TD - TW) in °C the values of the slope of the specific h u m i d i t y /
temperature curve, AS, and the specific h u m i d i t y deficit SHD = {qw, T,., - q
required in the Penman--Monteith equation (eq. 1) are calculated as follows.
Let:

x = TD/s- 3 (A-l)
then the saturation vapour pressure at dry-bulb temperature, 8VPTD is cal-
culated from:
SVPTD = 0.003x 4 + 0.063x 3 + 0.776x: + 5.487x + 17.044 (A-2)
and

DSVPT D - dSVPTD - (0.012x 3 + 0.189x 2 + 1.552x + 5.487)/5 (A-3)


dT
Then:
AS = (0.622 D S V P T D P / 1 . 0 0 4 5 ) / ( P / 1 . 0 0 4 5 - 0.378SVPTD)2 (A-4)
where P is atmospheric pressure in mbar.
The saturated specific h u m i d i t y at dry-bulb temperature, qw,TD, is calcu-
lated from:
0.622
qw, T D = (A-5)
P/1.0045SVPTD - 0.378

and the actual specific h u m i d i t y from:

q = qw, Tw - K ( T D - TW)

where qw, Tw is the saturated specific h u m i d i t y at wet-bulb temperature


calculated by substituting Tw into eqs. A-l, A-2 and A-5, and K is the screen
psychrometric constant.

APPENDIX B

To complete the sensitivity analysis of the Penman- -Monteith equation, it


is necessary to calculate:
aAs/aTD, aSHD/aTD and aSHD/aDEP
given the relationships of Appendix A. Then:
aA S 1 2c
~TD B2 [0.622A ( d S V P T D / d T D ) ] - "-~ ( - 0 . 3 7 8 ( d S V P T D / d T D )

where
A = P/1.0045; B = A - 0.378SVPT D; and C = 0.622 DSVPTDA
d S V P T D / d T D = (0.036x 2 + 0.278x + 1.552)/25
189

and x, dSVPTD/dTD axe d e f i n e d in A p p e n d i x A, using TD a n d D E P as t h e


i n p u t variables o f i n t e r e s t s u c h t h a t Tw = TD - DEP, t h e n :

0SHD_ 0.622( A dSVPTD)/( A )2


0 TD SVP~,D dTD SV--~TD 0.378

SVP~. w dTD ]]\SVPTw

o "o = 0.622
(- " w dT, 11
I SVPr, 0.37s
), + K

REFERENCES

Calder, I.R., 1977. A model of transpiration and interception loss from a spruce forest in
Plynlimon, central Wales. J. Hydrol., 33: 247--265.
Coleman, G. and DeCoursey, D.G., 1976. Sensitivity and model variance analysis applied
to some evaporation and evapotranspiration models. Water Resour. Res., 12(5): 873--
879.
Freeze, R.A., 1978. Mathematical models of hillslope hydrology. In: M.J. Kirkby (Editor),
Hillslope Hydrology, Wiley, Chichester.
Hahn, G.J. and Shapiro, S.S., 1967. Statistical Models in Engineering. Wiley, New York,
N.Y.
Jarvis, P.G., 1976. The interpretation of the variations in leaf water potential and stomatal
conductance found in canopies in the field. Philos. Trans. R. Soc. London, Ser. B, 273:
593--610.
Maki, T., 1975. Wind profile parameters of various canopies as influenced by wind velocity
and stability. J. Agric. Meteorol. (Tokyo), 31(2): 61--70.
McCuen, R.H., 1974. A sensitivity and error analysis of procedures used for estimating
evaporation. Water Resour. Bull., 10(3): 486--498.
McNaughton, K.G., 1976. Evaporation and advection II Evaporation downwind of a
boundary. Q.J.R. Meteorol. Soc., 102: 193--202.
Monteith, J.L., 1965. Evaporation and environment. Symp. Soc. Exp. Biol., 19: 205--234.
Monteith, J.L. and Szeicz, G., 1962. Radiative temperature in the heat balance of natural
surfaces. Q.J.R. Meteorol. Soc., 88: 496--507.
Penman, H.L. and Long, I.F., 1960. Weather in wheat. Q.J.R. Meteorol. Soc., 86: 16.
Rutter, A.J., Morton, A.J. and Robins, P.C., 1975. A predictive model of rainfall inter-
ception in forests, II. Generalisation of the model and comparison with observations in
some coniferous and hardwood stands. J. Appl. Ecol., 12: 367--380.
Saxton, K.E., 1975. Sensitivity analysis of the combination evapotranspiration equation.
Agric. Meteorol., 15: 343--353.
Stewart, J.B., 1977. Evaporation from the wet canopy of a pine forest. Water Resour. Res.,
13(6): 915--921.
Stewart, J.B., 1979. Using the combination equation to estimate evaporation: a note on
the relationships between the principle forms. Agric. Meteorol. (submitted).
Stewart, J.B. and Thom, A.S., 1973. Energy budgets in pine forests. Q.J.R. Meteorol. Soc.,
99: 154--170.
190

Strangeways, I.C. and McCulloch, J.S.G., 1965. A low priced automatic hydrometeorological
station. Bull. Int. Assoc. Sci. Hydrol., 10(4): 57--62.
Szeicz, G. and Long, I.F., 1969. Surface resistance of crop canopies. Water Resour. Res.,
5(3): 622--633.
Tan, C.S. and Black, T.A., 1976. Factors affecting the canopy resistance of a Douglas-fir
forest. Boundary-Layer Meteorol., 10: 475--488.
Thorn, A.S. and Oliver, H.R., 1977. O n Penman's equation for estimating regional evapora-
tion. Q.J.R. Meteorol. Soc., 103: 345--357.
Van Bavel, C.H.M., 1966. Potential evaporation: the combination concept and its experi-
mental verification.Water Resour. Res., 2: 455--467.
Van Bavel, C.H.M., 1967. Changes in canopy resistance to water loss from alfalfainduced
by soil water depletion. Agric. Meteorol., 4: 165--176.

You might also like