Evaluation of 13 Empirical Reference Potential

Evapotranspiration Equations on the

Island of Crete in Southern Greece
Fotios Xystrakis, Ph.D.1; and Andreas Matzarakis2

Abstract: Knowledge of reference potential evapotranspiration (PETref ) conditions is important for a number of vegetation- and hydro-
logical-related applications. Direct estimations of PETref are difficult and require sophisticated instrumentation. The Food and Agriculture
Organization (FAO), therefore, proposed a method for the estimation of PETref using only meteorological data. This equation has been widely
accepted as the standard method for the estimation of PETref because of its good fit with measured values. Nevertheless, it requires several
meteorological variables (e.g., wind speed), which are rarely available. Where such data are not available, its application is hampered. To
overcome this problem, a number of simpler, empirical equations requiring only a fraction of the meteorological input variables required by
the FAO PETref have been developed. Before using these equations, it is important to evaluate their performance and choose the equation that
will have the lowest possible bias in the estimation of PETref . Using daily meteorological observations obtained from seven meteorological
stations on the island of Crete (southern Greece), the performance of 13 empirical equations (radiation- and temperature-based) for the
estimation of PETref has been evaluated against the estimations of PETref using the FAO equation. Performance was evaluated on a daily
and a monthly basis, and five different measures of goodness of fit were used. The results showed that when the use of the FAO equation is not
possible because of the unavailability of data, some empirical methods can serve as appropriate alternatives. The radiation-based equations
generally performed better than those that included only temperature-related input variables. The equations proposed by Hansen and
Turc were the most useful because they had an average monthly absolute error ranging from 5.7 to 17.7 mm and 5.5 to 19.2 mm, respectively.
DOI: 10.1061/(ASCE)IR.1943-4774.0000283. © 2011 American Society of Civil Engineers.
CE Database subject headings: Evapotranspiration; Evaluation; Performance characteristics; Validation; Greece.
Author keywords: Evapotranspiration; Evaluation; Performance characteristics; Validation.

Introduction Methodologies of Crop Water Requirements accepted the follow-

ing definition of the reference surface: “A hypothetical reference
Potential evapotranspiration (PET) is defined as the amount of crop with an assumed crop height of 0.12 m, a fixed surface resis-
water that can potentially evaporate and transpire from a vegetated tance of 70 s=m, and an albedo of 0.23” (Allen et al. 1998). Con-
surface with no restrictions other than the atmospheric demand (Lu sequently, PETref is defined as the amount of water that can
et al. 2005). PET provides a good representation of the maximum potentially evaporate and transpire from a reference crop with
possible water loss to the atmosphere. Knowledge of PET rates is no restrictions other than the atmospheric demand.
essential for a variety of applications, including hydrological mod- PETref can be obtained using specific devices, through measure-
eling, irrigation planning, geobotanical studies, and estimation of ments of various physical parameters, or on the basis of the soil
sensitive-to-climatic change aridity indexes. Although the PET water balance using lysimeters. These methods are expensive, de-
concept is applied in a wide spectrum of applications, the term manding in terms of accuracy of measurements, and require well-
is considered a source of confusion because of the vague definition trained research personnel (Allen et al. 1998). Perhaps the easiest
of “vegetated surface.” To overcome this problem, the term PET way to calculate PETref values is with the FAO Penman-Monteith
has gradually been replaced by “reference potential evapotranspi- method, which according to an expert consultation held in May
ration” (PETref ), for which the characteristics of the vegetated 1990, is now the recommended standard method for the definition
surface have been standardized. The Food and Agriculture and computation of PETref . The form of this equation, revised from
Organization (FAO) Expert Consultation of Revision of FAO Allen et al. (1998), is

1
Institute of Silviculture, Albert-Ludwigs Univ. of Freiburg, Tennenba- 0:408·Δ·ðRn
GÞ þ γ·ð900=T mean þ 273Þ·U 2 ·ðes
ea Þ
cherstr 4, D-79106, Freiburg. I Br., Germany (corresponding author). PETFAO ¼
Δ þ γ·ð91 þ 0:34·U 2 Þ
E-mail: xystrakis.fotis@waldbau.uni-freiburg.de
2
Professor, Meteorological Institute, Albert-Ludwigs Univ. of Freiburg, ð1Þ
Werthmannstr 10, D-79085, Freiburg I Br., Germany. E-mail: andreas
.matzarakis@meteo.uni-freiburg.de
where Δ = slope of saturation vapor pressure curve at air temper-
Note. This manuscript was submitted on January 24, 2010; approved on
August 3, 2010; published online on August 20, 2010. Discussion period ature T (kPa · °C
1 ); Rn = net radiation (MJ · m
2 ); G = soil heat
open until September 1, 2011; separate discussions must be submitted flux (MJ · m
2 ) (0 for daily periods); γ = psychrometric constant
for individual papers. This paper is part of the Journal of Irrigation (kPa · °C
1 ); T mean = daily mean air temperature at 2 m height (°C);
and Drainage Engineering, Vol. 137, No. 4, April 1, 2011. ©ASCE, ISSN U 2 = wind speed at 2 m height (m · s
1 ); es = saturation vapor pres-
0733-9437/2011/4-211–222/$25.00. sure (kPa); and ea = actual vapor pressure (kPa).

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 211

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
 This physically based equation, involving all the meteorological Table 1. Meteorological Variables Collected by Weather Stations in Study
variables controlling evapotranspiration, is deemed to provide the Area; Values Include Daily Observations
best estimates of PETref values over a wide range of climates and is Variable Description
widely used as a standard for the evaluation of other empirical
PETref formulas (Droogers and Allen 2002; Gavilán et al. 2006; T mean Daily mean air temperature at 2 m above ground (°C)
Nandagiri and Kovoor 2006; Popova et al. 2006; Trajkovic 2005; T min Minimum air temperature (°C)
Xu and Singh 2002). T max Maximum air temperature (°C)
However, the equation requires a variety of meteorological input Td Dew-point temperature (°C)
variables that the vast majority of meteorological stations do not RH Relative humidity (%)
record. For this reason, in most operational applications PETref U2 Wind velocity at 2 m above ground (m · s
1 )
is estimated by means of various simpler, empirical equations that Cl Cloud cover (octas)
are less demanding in terms of input variables. Depending on the
necessary input data, these equations, can be classified as either
temperature-based, radiation-based, or combined methods. The
Table 2. Location and Observation Period of Weather Stations;
number of empirical equations is rather high and the output may
Geographic Coordinates Are in Decimal Degrees; Projection System:
be inconsistent and prone to bias because of the different underly- WGS84
ing assumptions and the corresponding input data requirements or
because they were developed for specific climatic regions (Allen Station Latitude Longitude Elevation Observation period
et al. 1998; Grismer et al. 2002). To ensure accuracy, it is essential Ierapetra 35.02 25.73 10 1956–2001
that the performance of these empirical equations for the estimation Iraklio 35.33 25.18 39.3 1955–2001
of PETref be evaluated prior to their use, either by using existing Kastelli 35.2 25.33 335 1976–2001
observations of evapo(transpi)ration or PETref values calculated Rethimno 35.35 24.51 5.1 1957–2001
with one of the more sophisticated methods (energy balance, lysim-
Sitia 35.2 26.1 115.6 1960–2001
eters, etc.) in the region of interest.
Souda 35.55 24.11 151.6 1958–2001
To this end, daily values of PETref , estimated by using 13 em-
pirical equations obtained from a review of the existing literature, Tymbaki 35 24.75 6.7 1959–2001
were compared with the respective output of the PETFAO equation
using data obtained from seven meteorological stations in Crete, the meteorological stations. To estimate the extraterrestrial radia-
which has a typical semiarid, Mediterranean climate. The risk of tion (Ra ) and net radiation (Rn ) the method described by Allen
desertification is high (Croke et al. 2000; Grove and Rackham et al. (1998) was followed. An albedo value of 0.23 was used (Allen
1993; Vardavas et al. 1997) and sustainable irrigation planning for et al. 1998) in the calculation of Rn for the reference crop.
drought mitigation and adaptation is essential (Lambrakis and
Kallergis 2001; Tsakiris and Tigkas 2007; Tsakiris et al. 2007; Empirical PETref Equations
Vardavas et al. 1997). The evaluation of the empirical equations
The empirical equations evaluated include temperature- and radi-
in this region is of major interest. PETref estimates exist for Crete
ation-based methods, all of which are presented in the following
(for example, Naoum and Tsanis 2003; Tsanis and Naoum 2003),
sections. Because these equations include units from nonuniform
but a current evaluation of empirical equations is lacking.
systems, unit conversion was necessary for their application. A
Various measures of goodness of fit were applied to assess the
summary of the definitions and descriptions of the symbols is
accuracy of the empirical PETref equations. The results of this study
shown in Table 3. The equations where the input variables should
provide insight into the choice of the most suitable empirical PETref
be measured in different units than these reported in Table 3, are
equation when the unavailability of data prevents the application
noted in the text. The equations include a wide variety of input
of the PETFAO equation in Crete and in other regions with similar
parameters thus their applicability depends on the data availability.
climatic conditions.
A summary of the different potential evapotranspiration methods
included in this paper is presented in Table 4.
Methods Temperature-Based Methods
• Hargreave’s equation (Allen et al. 1998; Oudin et al. 2005;
Climate Data and Location of Meteorological Stations Xu and Singh 2002)
Daily values of meteorological data obtained from seven weather
stations in Crete were used in the study. The meteorological data pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PETHar ¼ 0:0023·ðRa =λÞ· ðT max
T min Þ·ðT mean þ 17:8Þ
included the variables shown in Table 1. The observation period
and other variables of the weather stations are presented in Table 2. ð2Þ
The locations of the meteorological stations are depicted in Fig. 1.
These meteorological variables build the input data used for the
estimation of the daily values of PETref in all of the equations. Days • McGuinness-Bordne’s equation (Oudin et al. 2005)
with at least one meteorological variable missing were excluded
from the analyses. The long observation period allowed for the PETMcG ¼ ðRa =λ·ρÞ·½ðT mean þ 5Þ=68 ð3Þ
extraction of safe conclusions over the performance of the
PETref equations for both daily and monthly analyses. The global
(total) solar radiation data, not measured by any of the meteorologi- • Romanenko’s equation (Oudin et al. 2005)
cal stations in Crete, were estimated using the RayMan model
(Matzarakis and Rutz 2007; Matzarakis et al. 2007), taking into
consideration the cloud cover, the latitude, and the elevation of PETRom ¼ 4:5·½1 þ ðT mean =25Þ2 ·ð1
ea =es Þ ð4Þ

212 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
 Fig. 1. Location of meteorological stations in Crete, Greece

Table 3. Definitions of and Procedures for Calculation of Units Used in PETref Equations
Symbol Definition Unit Calculation Reference
Ra Extraterrestrial radiation MJ=m2 Depends on latitude and Julian day Allen et al. (1998)
Rs Global (total) radiation MJ=m2 RayMan Matzarakis and Rutz (2007)
Rn Net radiation MJ=m2 Based on Rs , albedo and longwave radiation. Allen et al. (1998)
Computational details in Allen et al. (1998)
T mean Mean air temperature °C — —
T max Maximum air temperature °C — —
T min Minimum air temperature °C — —
DL Day length hours Function of latitude, RayMan Matzarakis and Rutz (2007)
eo Generalized equation of saturation vapor pressure kPa eo ¼ 0:6108 · exp½ð17:27 · TÞ=ðT þ 237:3Þ Allen et al. (1998)
es a Mean saturation vapor pressure kPa es ¼ ðeT max þ eT min Þ=2
ea Actual vapor pressure kPa ea ¼ 0:6108 · exp½ð17:27·T dew Þ=ðT dew þ 237:3Þ
SVDa Saturated vapor density at mean air temperature g=m3 SVD ¼ 216:7·½es =ðT mean þ 273:3Þ Lu et al. (2005);
Rosenberry et al. (2004)
RH Relative air humidity % h — i —
Δ Slope of saturation vapor pressure curve kPa=°C Δ ¼ 4098 · 0:6108 · exp T17:27·T mean
mean þ237:3
=ðT mean þ 237:3Þ2 Allen et al. (1998)

γb Psychrometric constant kPa=°C γ ¼ ðCp ·PÞ=ðε·λÞ ¼ 0:665·10
3 ·P Allen et al. (1998)

ρ Water density 1000ðkg=m3 Þ Allen et al. (1998)
λ Latent heat of vaporization MJ=kg λ ¼ 2:501
ð2:361 · 10
3 Þ · T mean Allen et al. (1998)
Note: Because of nonlinearity, the mean saturation vapor pressure is estimated as the average of the saturation vapor pressure (eo ) at T max and the saturation
vapor pressure at T min . For computation details, refer to Allen et al. (1998).
a
es is measured in mb.
b
For unit descriptions and calculation procedures, refer to Allen et al. (1998).

Table 4. Summary of the Input Data Requirements of PETref Equations

T mean T min T max T dew RH Wind speed DL Rs Ra Rn
PETFAO ✓ ✓ ✓ ✓ ✓ ✓
PETCap ✓ ✓
PETdeB ✓ ✓
PETHam1 ✓ ✓ ✓ ✓
PETHam2 ✓ ✓
PETHam3 ✓ ✓ ✓ ✓
PETHan ✓ ✓
PETHar ✓ ✓ ✓ ✓
PETJen ✓ ✓
PETMak ✓ ✓
PETMcC ✓
PETMcG ✓ ✓
PETRom ✓ ✓ ✓ ✓
PETTur ✓ ✓ ✓

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 213

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
• Hammon’s equation (Lu et al. 2005) (StatSoft 2007). The measures applied to evaluate the performance
of the empirical equations were:
PETHam1 ¼ 0:1651·ðDL=12Þ·½216:7·es =ðT mean þ 273:3Þ·a • Mean square error (MSE):
ð5Þ PN
i¼1 ðE i

O i Þ2
where a ¼ constant ¼ 1:2; and es in mb. ð15Þ
N
1
• Hammon’s equation (Oudin et al. 2005)
• Mean absolute error (MAE):
PETHam2 ¼ ðDL=12Þ2 · expðT mean =16Þ ð6Þ
PN
• Hammon’s equation (Rosenberry et al. 2004) jEi
Oi j
i¼1
ð16Þ
N
1
PETHam3 ¼ 0:55·ðDL=12Þ2 ·ðSVD=100Þ·25:4 ð7Þ
• Relative mean square error (RMSE):
• McCloud’s equation (Irmak et al. 2003; Jacobs and Satti 2001)
PN E i
Oi 2
PETMcC ¼ a · bðT mean
32Þ ð8Þ i¼1 ð E i Þ
ð17Þ
N
1
where PET (is in in.); a ¼ constant ¼ 1:07; and b ¼
constant ¼ 0:01. • Relative mean absolute error (RMAE):

Radiation-Based and Combination Methods PN jEi
Oi j

i¼1 Ei
• Hansen’s equation (Hansen 1984; Xu and Singh 2002) ð18Þ
N
1
PETHan ¼ 0:7·½Δ=ðΔ þ γÞ·ðRs =λÞ ð9Þ
• Mean bias error (MBE):
where Rs is in cal · cm
2 ; Δ and γ are in mb · °C
1 ; and λ is PN
in cal · g
1 . ðE i
Oi Þ
i
ð19Þ
• Caprio’s equation (Caprio 1974) (source: http://www.wsl.ch/ N
1
staff/niklaus.zimmermann/programs/aml3_2.html)
• Pearson’s correlation coefficient (CC), described in
PETCap ¼ ð6:1=106 Þ·Rs ·ð1:8·T mean þ 1:0Þ ð10Þ StatSoft (2007).
where N = number of observations; E i = values from empirical
where Rs is in kJ · m
2 . equations; and Oi = respective values from PETFAO equation.
• Jensen-Haise’s equation [exact form is from Rosenberg et al. These (or similar) statistical procedures have been widely dis-
(1983), but forms presented in Jensen and Haise (1963) and cussed in the relevant literature (Jacovides 1998; Jacovides and
Rosenberry et al. (200)4 were also tested] Kontoyiannis 1995; Knotters and Voshaar 1998; Willmott 1984;
PETJen ¼ ðRs =λÞ·ð0:025·T mean þ 0:08Þ ð11Þ Willmott 1982; Willmott et al. 1985) and have proven to be impor-
tant for the evaluation of the performance of empirical PETref
• Turc’s equation (Jacobs and Satti 2001; Lu et al. 2005). equations.
For RH < 50%: All but the MBE measure of goodness of fit were used for the
evaluation of the performance of the empirical equations on a daily
PETTurc ¼ 0:013·½T mean =ðT mean þ 15Þ·ðRs · 23:8846 and a monthly basis. The daily PETref estimates were averaged in
þ 50Þ·½1 þ ð50
RHÞ=70 ð12a Þ monthly totals because it has been demonstrated that any bias in
calculation can be reduced when integrated over longer periods
For RH > 50%: and, therefore, the errors were expected to be lower (Hupet and
Vanclooster 2001). Additionally, the monthly analysis allows for
PETTurc ¼ 0:013·½T mean =ðT mean þ 15Þ·ðRs · 23:8846 þ 50Þ the suggestion over the use of specific empirical formulas during
ð12b Þ the different months (different underlying climate conditions). The
MBE was used to evaluate the averaged monthly totals of PETref
• Makkink’s equation [from Rosenberry et al. (2004); forms cited estimates.
in Xu and Singh (2000, 2002) were also tested] Prior to the analyses of goodness-of-fit measures, the meteoro-
logical stations were ordinated by the weight of each variable in
PETMak ¼ ½0:61·ðΔ=Δ þ γÞ·ðRs =λÞ
0:012·10 ð13Þ
influencing PETref , allowing for the identification of groups with
where Rs is in cal · cm
2 ; Δ and γ are in mb · °C
1 ; and λ is climatic homogeneity and the detection of possible regional
in cal · g
1 . patterns or climatic regions (Comrie and Glenn 1998; Giles et al.
• De Bruin’s equation (de Bruin and Keijman 1979; Rosenberry 1989; Morin et al. 1979) across the study area. These variables
et al. 2004) included the mean daily values of T mean , T max , T d , Rs and wind
speed over the observation period recorded at each station. The
PETdeB ¼ f½Δ=ð0:85·Δ þ 0:63·γÞ·ðRn =λÞg · 10 ð14Þ ordination method used was an s-mode principal component analy-
sis (PCA). The type of cross-products matrix was the correlation
where Rn is in cal · cm
2 ; Δ and γ are in mb · °C
1 ; and λ is matrix (data were centered and standardized), which was used
in cal · g
1 . to eliminate differences in measurement units between variables.
Outlier analysis did not reveal any extreme value at three standard
Evaluation of PETref Equations
deviations cut-off level. The scores of the variables were calculated
The performance of the empirical equations was assessed by using the distance-based biplot method. The analysis was per-
using the calculations from several goodness-of-fit statistics formed with the PCORD software (McCune and Mefford 1999).

214 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Results which resulted in negative values of PETref . For this reason, the
evaluation of PETRom was based only on the MAE, RMAE and
Climatic Regions the Pearson’s CC.

The ordination diagram shown in Fig. 2 helped identify spatial pat- Temperature-Based Equations
terns in the climate data recorded by the meteorological stations. The temperature-based equations that exhibited the best fit were
From the PCA diagram, the stations Tymbaki, Ierapetra, and PETMcG , PETHam2 , and PETHam3 . The minimum values obtained
Sitia form one group, the stations Iraklio and Rethimno form for the MSE, MAE, RMSE, and RMAE (indicating best perfor-
another. While the two remaining stations, Souda and Kastelli, mance) were for PETHam2 in Kastelli, PETHam2 in Kastelli,
are both located on the left-hand side of the diagram they do PETMcG in Sitia, and PETHam2 in Sitia, respectively. The maximum
not form a distinct group. The first three stations, located in the CC value was for PETMcG in Kastelli. The PETMcG revealed the
east and southeast, represent the driest part of Crete (see Fig. 2). best fit for the stations Iraklio and Rethimno, where the wind speed
The stations Iraklio and Rethimno, situated in northcentral Crete values were highest (and PETref ). For the other stations, the MSE
(see Fig. 2), stand out owing to the high wind speed values at these and MAE values were much higher. The worst fits were observed
locations. The high wind speed values are positively correlated with for these two stations. PETHam3 was identified as the empirical
high PETref values, especially in hot and dry climates (Allen et al. PETref equation with the highest Pearson’s CC for three stations,
1998). Souda, a lowland station in western Crete, lies between the but given the results of the remaining goodness-of-fit measures it
aforementioned groups and the Kastelli meteorological station, cannot be considered a suitable equation for the estimation
which is situated to the very left of the diagram. Its position is ex- of PETref .
plained by the low values of T mean , T max , T dew , and Rs because its Radiation-Based Equations
altitude (300 m a.s.l.) is higher than the other stations (see Table 2). The pattern is clearer for the radiation-based equations. PETTurc
The northwest to southeast climatic gradient observed in Crete stood out clearly from the other empirical equations, followed
(Pennas 1977) is also apparent in Fig. 2. by PETHan and PETdeB . PETMak also demonstrated good perfor-
mances because its goodness-of-fit measures were comparable with
Evaluation of the Empirical PETref Equations on a Daily
PETHan and PETdeB . As with the temperature-based methods, the
Basis
worst fits were observed for the stations Iraklio and Rethimno.
The results of the analysis of the daily PETref values are shown in PETJen had the best Pearson’s CC at the stations Ierapetra and
Table 5. Some equations perform relatively well, whereas others Kastelli. PETdeB2 was also identified as the best method in terms
reveal a great bias in the estimation of PETref . The radiation-based of RMSE for the stations Souda and Kastelli, whereas PETdeB had
formulas perform better for all meteorological stations except the best fit in terms of CC for Souda. The minimum values of MSE,
Rethimno and Iraklio. There is great variation with respect to MAE, RMSE, and RMAE were for PETTur in Sitia,; PETTur in
the best temperature-based equations, which contrasts with the Kastelli; PETTur in Tymbaki; and PETTur in Tymbaki, respectively.
clearer pattern observed for the radiation-based equations. Table 5 The maximum value of CC was for PETJen in Kastelli. As with the
shows that the RMSE and RMAE of the PETRom , could not be temperature-based-equations, the best fits were observed for the
evaluated because of the calculation procedures. Since the satura- stations Kastelli and Sitia, but good fits were also observed for
tion vapor pressure (es ) is here calculated as ðeTmax þ eTmin Þ=2 as Tymbaki.
proposed from Allen et al. (1998), for some humid and warm days Evaluation of the Empirical PETref Equations on a
(e.g., for 167 days in Ierapetra), the actual vapor pressure (ea ) Monthly Basis
is calculated to be lower than the saturation vapor pressure (es ),
The results of the integrated monthly PETref values are presented in
Table 6. It can be seen that some equations clearly stood out for
their consistently good fit with the PETFAO equation, whereas
others clearly showed a very poor fit. For the daily analysis (see
the “Evaluation of the Empirical PETref Equations on a Daily
Basis” section), the radiation-based equations provided better esti-
mates than the temperature-based formulas for all the stations other
than Iraklio and Rethimno. In the case of the latter stations, the best
temperature-based equations outperformed the best radiation-based
equations. Additionally, the relative errors (RMSE and RMAE) for
the integrated monthly PETref values were smaller than the corre-
sponding errors for the daily values.
Temperature-Based Equations
The pattern observed from the study of the integrated monthly val-
ues was clearer than that of the daily values (see the “Evaluation of
the Empirical PETref Equations on a Daily Basis” section). PETMcG
exhibited the best fit for the stations Iraklio and Rethimno, but
the performance of the empirical equations was still worse for
these stations than elsewhere. For the remaining stations, PETHam2
stood out from all of the other temperature-based equations. The
minimum values of MSE, MAE, RMSE, and RMAE were for
Fig. 2. Axes 1 and 3 of the PCA diagram of meteorological variables
PETHam in Sitia; PETHam2 in Sitia; PETHam2 , in Sitia and Souda;
recorded by the seven weather stations; cumulative of total variance
and PETHam2 in Sitia, respectively. The maximum value of CC
explained from axes 1 and 2 equals 91.9%
was for PETMcG in Souda.

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 215

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Table 5. Goodness-of-Fit Measures of the Empirical PET Equations when Compared with the PETFAO ; Bold Values Indicate Radiation- and Temperature-
Based Equations with Best Fits for Each Meteorological Station and for Each Goodness-of-Fit Measure
Temperature-based Radiation-based
PETHar PETMcG PETRom PETHam1 PETHam2 PETHam3 PETMcC PETMak PETHan PETCap PETJen PETTur PETdeB
Ierapetra
MSE 3.619 1.506 — 1.776 1.150 3.635 2.421 1.704 0.959 2.176 2.493 0.630 1.412
MAE 1.339 0.996 1.658 1.029 0.825 1.474 1.188 0.864 0.703 1.103 1.195 0.602 0.867
MRSE 0.396 0.088 — 0.117 0.103 0.424 0.500 0.155 0.083 0.158 0.114 0.067 0.148
MRAE 0.448 0.242 16.778 0.282 0.240 0.527 0.478 0.271 0.202 0.263 0.252 0.179 0.260
CC 0.818 0.901 0.871 0.913 0.910 0.923 0.848 0.932 0.933 0.944 0.944 0.953 0.893
Sitia
MSE 2.762 0.952 — 0.896 0.560 2.141 1.792 0.512 0.400 2.530 2.974 0.365 0.673
MAE 1.253 0.779 1.472 0.781 0.560 1.208 1.035 0.525 0.497 1.211 1.330 0.466 0.657
MRSE 0.507 0.061 — 0.081 0.061 0.319 0.507 0.149 0.089 0.215 0.150 0.070 0.088
MRAE 0.523 0.199 0.317 0.236 0.180 0.471 0.484 0.230 0.185 0.302 0.287 0.166 0.201
CC 0.797 0.944 0.731 0.916 0.935 0.938 0.839 0.956 0.957 0.961 0.962 0.960 0.960
Tymbaki
MSE 1.342 1.165 — 1.158 0.815 2.522 2.197 0.721 0.484 2.766 3.174 0.444 0.741
MAE 0.831 0.866 1.958 0.863 0.671 1.271 1.089 0.571 0.549 1.226 1.338 0.522 0.671
MRSE 0.166 0.073 — 0.099 0.079 0.362 0.557 0.110 0.066 0.149 0.111 0.055 0.080
MRAE 0.278 0.219 4.801 0.259 0.207 0.493 0.498 0.208 0.174 0.272 0.263 0.162 0.195
CC 0.884 0.931 0.790 0.908 0.927 0.926 0.849 0.950 0.951 0.958 0.959 0.959 0.939
Iraklio
MSE 4.289 1.263 — 1.974 1.367 4.256 2.925 2.527 1.804 2.945 3.137 1.499 1.895
MAE 1.570 0.848 1.284 1.079 0.852 1.725 1.351 1.129 0.920 1.337 1.380 0.852 0.987
MRSE 0.970 0.098 — 0.160 0.185 0.780 0.916 0.914 0.570 0.881 0.606 0.378 0.522
MRAE 0.684 0.224 0.255 0.317 0.291 0.733 0.664 0.579 0.429 0.533 0.464 0.360 0.426
CC 0.756 0.869 0.831 0.884 0.875 0.888 0.825 0.847 0.849 0.868 0.866 0.873 0.842
Rethimno
MSE 3.424 1.616 — 1.910 1.635 4.054 2.931 2.765 2.105 3.639 3.845 1.757 2.275
MAE 1.368 0.971 1.122 0.998 0.896 1.633 1.314 1.112 0.922 1.466 1.510 0.845 1.030
MRSE 0.864 0.113 — 0.164 0.191 0.783 0.743 0.982 0.623 0.795 0.583 0.393 0.607
MRAE 0.586 0.243 0.268 0.298 0.283 0.694 0.584 0.573 0.427 0.515 0.458 0.351 0.429
CC 0.789 0.838 0.840 0.845 0.837 0.847 0.781 0.808 0.809 0.825 0.823 0.836 0.802
Souda
MSE 1.275 1.048 — 0.873 0.674 1.955 1.947 0.644 0.448 2.555 2.914 0.481 0.535
MAE 0.843 0.819 1.912 0.750 0.596 1.115 1.018 0.571 0.502 1.167 1.253 0.500 0.562
MRSE 0.178 0.069 — 0.080 0.062 0.274 0.550 0.200 0.116 0.360 0.212 0.100 0.092
MRAE 0.305 0.213 0.535 0.232 0.186 0.432 0.500 0.268 0.203 0.348 0.306 0.186 0.191
CC 0.908 0.944 0.850 0.922 0.935 0.939 0.847 0.954 0.955 0.957 0.958 0.956 0.957
Kastelli
MSE 1.141 0.845 — 0.920 0.545 2.117 2.045 0.494 0.409 1.953 2.303 0.421 0.564
MAE 0.781 0.719 1.400 0.759 0.544 1.148 1.102 0.476 0.475 1.004 1.098 0.455 0.569
MRSE 0.163 0.067 — 0.091 0.063 0.320 0.803 0.133 0.084 0.791 0.177 0.305 0.077
MRAE 0.290 0.207 4.587 0.251 0.189 0.473 0.636 0.210 0.180 0.345 0.281 0.180 0.185
CC 0.884 0.950 0.850 0.930 0.939 0.943 0.858 0.957 0.958 0.963 0.964 0.963 0.955

Radiation-Based Equations Seasonal Patterns

Of the radiation-based equations, the PETHan and PETTur exhibited
the most consistent pattern of best fit, both had comparable good- Additional information was extracted from the analysis of the MBE
of the averaged monthly totals of PETref , presented in Fig. 3. This
ness-of-fit values. PETdeB followed closely, also with comparable
provided information concerning the over- or underestimation of
results. In spite of the different climatic conditions in Rethimno and
PETref by the empirical equations, and can be used as a tool for
Iraklio, no other equation was observed to perform better than these the choice of the best-fitting empirical formulas during the different
three, although the error measures were 2–3 times higher than for seasons and their underlying different climate conditions in the ab-
the other stations. The minimum values of MSE, MAE, RMSE, and sence of a more thorough analysis of the physical conditions that
RMAE were for PETHan in Sitia for all of the aforementioned stat- lead to over- or underestimations on a daily basis.
istical measures. The highest value of Pearson’s CC was for PETHan The graphs reveal that for some equations there is a clear pattern
in Kastelli. of over- or underestimation for all weather stations, but for some of

216 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Table 6. Goodness-of-Fit Measures of Results of Empirical PET Equations Compared with those of the PETFAO Equation; Basis for Comparison Was the
Integrated Monthly Values of PET Not Grouped in Monthly Subsets; Bold Values Indicate Radiation- and Temperature-Based Equations with Best Fits for
Each Meteorological Station and for Each Goodness-of-Fit Measure
Temperature-based Radiation-based
PETHar PETMcG PETRom PETHam1 PETHam2 PETHam3 PETMcC PETMak PETHan PETCap PETJen PETTur PETdeB
Ierapetra
MSE 2154.9 526.1 2017.16 877.06 209.14 2586.06 923.56 962.20 268.03 1302.58 1581.63 129.16 481.94
MAE 32.962 18.87 37.260 23.188 11.228 42.278 24.596 22.791 11.065 26.489 29.632 8.256 15.661
MRSE 0.145 0.035 0.084 0.048 0.018 0.262 0.205 0.052 0.012 0.034 0.033 0.008 0.051
MRAE 0.310 0.155 0.251 0.192 0.106 0.480 0.329 0.197 0.087 0.159 0.161 0.071 0.163
CC 0.938 0.971 0.948 0.977 0.980 0.984 0.939 0.985 0.986 0.990 0.990 0.991 0.958
Sitia
MSE 1905.7 546.63 1614.62 475.41 131.37 1647.69 989.23 199.76 47.02 1850.76 2238.92 51.43 311.66
MAE 33.722 20.780 34.490 18.560 8.94 35.512 24.650 12.215 5.564 31.616 34.804 5.749 14.458
MRSE 0.219 0.032 0.103 0.039 0.014 0.222 0.274 0.023 0.005 0.049 0.046 0.006 0.025
MRAE 0.396 0.163 0.267 0.174 0.091 0.442 0.374 0.131 0.056 0.193 0.187 0.058 0.127
CC 0.924 0.989 0.915 0.969 0.984 0.986 0.915 0.996 0.996 0.991 0.992 0.995 0.995
Tymbaki
MSE 546.1 595.30 3262.3 555.25 199.23 1852.99 908.99 334.06 79.69 2039.51 2404.09 83.99 256.19
MAE 16.127 20.927 48.488 19.35 10.70 36.608 24.038 13.62 7.19 32.76 36.21 7.41 12.54
MRSE 0.030 0.036 0.130 0.042 0.016 0.229 0.268 0.022 0.008 0.048 0.047 0.009 0.023
MRAE 0.138 0.167 0.317 0.178 0.101 0.446 0.369 0.125 0.070 0.191 0.193 0.073 0.117
CC 0.977 0.983 0.899 0.973 0.984 0.985 0.935 0.991 0.991 0.991 0.992 0.994 0.982
Iraklio
MSE 2722.0 287.69 1271.58 1031.88 389.54 3097.88 1682.9 1215.77 473.14 1446.50 1572.81 402.43 510.95
MAE 45.209 13.214 31.054 26.55 16.63 51.835 35.419 33.095 18.822 32.838 32.646 17.354 18.888
MRSE 0.404 0.021 0.065 0.076 0.080 0.587 0.547 0.309 0.144 0.242 0.156 0.105 0.230
MRAE 0.573 0.112 0.217 0.241 0.212 0.726 0.572 0.464 0.275 0.370 0.306 0.238 0.318
CC 0.922 0.973 0.945 0.968 0.976 0.978 0.928 0.982 0.983 0.981 0.982 0.984 0.978
Rethimno
MSE 1888.8 488.57 980.73 816.92 518.99 2693.14 1574.97 1238.14 568.87 1815.53 1953.25 481.53 670.99
MAE 39.016 18.265 26.146 23.88 18.42 48.427 32.622 32.340 19.268 35.310 35.281 17.700 21.413
MRSE 0.347 0.033 0.054 0.08 0.087 0.583 0.461 0.347 0.172 0.237 0.163 0.120 0.265
MRAE 0.505 0.147 0.191 0.231 0.213 0.694 0.501 0.468 0.285 0.364 0.310 0.241 0.334
CC 0.946 0.962 0.923 0.948 0.95 0.959 0.882 0.966 0.966 0.960 0.962 0.969 0.961
Souda
MSE 677.7 581.95 3804.69 406.09 164.92 1421.13 903.58 288.28 57.07 1809.1 2121.29 87.64 162.33
MAE 19.25 21.579 55.73 16.88 9.731 32.215 23.290 14.354 5.904 30.454 32.526 7.067 10.305
MRSE 0.052 0.037 0.142 0.037 0.014 0.182 0.293 0.038 0.009 0.068 0.049 0.009 0.023
MRAE 0.190 0.176 0.352 0.165 0.094 0.397 0.383 0.166 0.067 0.219 0.192 0.073 0.112
CC 0.979 0.990 0.956 0.973 0.98 0.987 0.926 0.993 0.993 0.987 0.989 0.992 0.993
Kastelli
MSE 495.7 464.43 1919.36 517.14 137.96 1579.19 1179.33 189.09 75.39 1311.21 1610.48 94.96 224.57
MAE 14.91 18.327 35.626 18.25 8.979 33.173 27.653 10.440 6.855 26.179 28.245 7.348 10.879
MRSE 0.038 0.036 3086.6 0.050 0.019 0.239 0.489 0.020 0.010 0.070 0.044 0.011 0.018
MRAE 0.149 0.168 3.473 0.194 0.104 0.453 0.523 0.113 0.079 0.219 0.183 0.081 0.104
CC 0.964 0.988 0.932 0.970 0.982 0.983 0.936 0.993 0.993 0.985 0.987 0.993 0.988

the others such a clear pattern is not apparent. An increase in the PETHam3 correlated with one another because of the similarity
MBE values during the summer period is normal due to the in- of the equations. Finally, PETMcG and PETRom revealed no constant
creased values of PETref . pattern. At some stations, it performed better during the winter and
More specifically, of the temperature-based equations, PETHar at others during the summer. The success of this equation in Iraklio
and PETHam3 constantly underestimated PETref for all stations and Rethimno must be attributed to the considerably lower bias
and all seasons with a bias that increased during the summer. during the winter period, when all of the other equations severely
PETHam2 , which exhibited similar bias to this of the radiation-based underestimated PETref .
equations, slightly underestimated PETref during winter and Of the radiation-based equations, PETHan and PETTur revealed
overestimated during summer (except at the stations Iraklio and the most consistent pattern of good fit (for all stations except
Rethimno), whereas PETHam1 tended to overestimate it during win- Rethimno and Iraklio). Their estimates were always close to
ter and underestimate it during summer. PETHam1 , PETHam2 , and PETFAO , and they either slightly under- or overestimated PETref .

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 217

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
 Fig. 3. Mean bias error between the averaged monthly totals of PETref estimated using empirical equations and the PETFAO equation

218 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
The underestimation generally occurred in winter and the overesti- study will be based on a comparison with the results of studies
mation during summer. PETCap and PETJen , two very similar equa- carried out in regions with similar climate conditions.
tions, revealed great variations in their MBE, with low bias during The results suggested that PETHar can be used as an alternative
winter and a slight underestimation of PETref , and with a very large to the PETref equation in situations where there is a lack of meteoro-
overestimation during the summer. PETMcC exhibited a distinct logical data (Allen et al. 1998). PETHar estimates for southern Spain
pattern, with a high underestimation of PETref during the spring. (Gavilán et al. 2008) produced an average daily RMSE of 0.8 mm
PETMak constantly underestimated PETref , while PETdeB tended when compared with the ASCE-Penman-Monteith equation
to overestimate PETref in summer and underestimate in winter (Gavilán et al. 2008). In Crete, which is at comparable latitude,
for most stations. This tendency was more pronounced in Iraklio the result of the comparison with the PETFAO was almost double
and Rethimno. (1.55 mm). Quantitative differences between this study and Gavilán
et al. (2008) should also be attributed to the differences in the
method for the estimation of PETref (integrated daily values from
Discussion hourly time steps, measure of radiation fluxes, etc.). The highest
error measures associated with PETHar were obtained for the sta-
Radiation- and Temperature-Based Equations tions Iraklio and Rethimno, followed by Ierapetra, Sitia, Tymbaki,
Souda, and Kastelli. This pattern indicates that the bias associated
The results revealed that the radiation-based equations generally with PETHar is positively correlated with the PETref values, which
overperformed the respective temperature-based equations. This agrees with Gavilán et al. (2008). The latest can be attributed to the
is in accordance with a number of similar studies (Lu et al. 2005; fact that PETHar is not able to account for the effect of high advec-
Xu and Singh 1998). This pattern is observed for both the daily and tion phenomena of semiarid environments (Berengena and Gávilan
monthly based analysis, for all goodness-of-fit measures, and at all 2005). The fact that the error associated with PETHar increases in
weather stations except Iraklio and Rethimno, where the equations areas with high wind speeds (Iraklio and Rethimno) does not cor-
of PETMcG and PETHam1 are found to be the best alternatives to the respond to the findings of Martínez-Cob and Tejero-Juste (2004).
PETFAO . The fact that the equations PETMcG and PETHam1 provided The reason for this is that in the case of the latter, high wind speeds
better estimates for the stations Iraklio and Rethimno should not be were related to a lower daily air temperature range [see p. 259 of
attributed solely to the better performance of these equations. The Martínez-Cob and Tejero-Juste (2004)], thus resulting in reduced
measures of their goodness of fit were relatively poor, indicating PETref rates, which was not true for this study. Nevertheless, other
that although they provided better results, they did not really per- studies have concluded that PETHar does not perform well or that it
form all that well. That is, all of the other equations performed requires calibration, in semiarid environments (Alexandris et al.
rather badly. This was obvious from both the very low values of 2006; Droogers and Allen 2002; Gavilán et al. 2006; Jabloun
Pearson’s CC and the high values of the relative errors. The RMSE and Sahli 2008; Stöckle et al. 2004). Similarly, this equation
and RMAE of, for example, 0.076 and 0.241 associated with was found to underperform in the warm humid climate of the
PETHam1 for the station Iraklio ascertained for the monthly analyses southeastern United States (Lu et al. 2005) and in the humid climate
(see Table 6) were very high when compared with the values of of the western Balkans (Trajkovic and Kolakovic 2009). Alexandris
0.008 and 0.071 for the corresponding measures of the PETTur et al. (2008) also indicated a severe underperformance of PETHar in
in the station Ierapetra. Their success should be attributed to the the Balkan region. By contrast, in the semiarid environment of
observed tendency to greatly overestimate PETref ; moreover, of California (Temesgen et al. 2005) and of Iran (DehghaniSanij et al.
the empirical equations take into account, the high wind speed val- 2004) the PETHar equation performed well relative to other empiri-
ues actually increase PETref rates. Their overestimation, observed cal and physical equations, despite the simplicity of the formula
for all of the other stations, was counterbalanced by the severe
which only requires air temperature data. The accuracy of this
underestimation of the PETref for these two stations by all of the
equation depends greatly on local climatic conditions (Alexandris
other equations.
et al. 2006; Wang et al. 2009) and, therefore, does not have the
The seasonal analysis of PETref estimates, except for some
performance stability required for it to be considered as a globally
PETref equations, did not show a constant under- or overestimation,
accepted method.
suggesting that for the different regions, the final choice of the em-
The proposed use of PETHar for the estimation of PETref
pirical PETref , beside the consideration of data availability, should
for periods of 10 days or more (Droogers and Allen 2002;
be also based on the patterns of performance. For different seasons,
Martínez-Cob 1996) is also underlined by the results of this study,
different equations should be chosen and this choice should be
reflected in the large decrease in the error associated with the
accompanied with a discussion over the level of under- or overesti-
integrated monthly values table 6.
mation of PETref . The identified differences in the performance of
PETMcG was found to perform well when compared with 27
the equations among seasons also allows for their season-based
other PET equations tested in various rainfall-runoff models (Oudin
calibration to obtain more accurate estimates of PETref . Further
et al. 2005) applied in 308 catchments in Australia, France, and the
research should be carried out to identify the physical conditions
United States. This corresponds with the findings of this study.
that contribute to the over- or underestimation of the empirical
However, whereas Oudin et al. (2005) found that the results of
PETref equations, but that is beyond the scope of this paper.
the PETMcG equation were comparable to those produced by the
Comparison to Similar Studies radiation-based equations, in Crete, this equation, like all temper-
ature-based equations, was outperformed by the radiation-based
Unfortunately, an extensive evaluation of empirical equations is equations at all stations except for Iraklio and Rethimno.
lacking for Greece. Only a small number of scientific publications A slightly modified version of the PETRom equation provided
dealing with the topic exist and the comparisons made are limited to good results in Switzerland (Xu and Singh 1998), as a method
a small number of empirical equations (Alexandris and Kerkides to estimate pan evaporation. However, it did not perform well in
2003; Michalopoulou and Papaioannou 1991; Papadopoulou et al. the regions studied by Oudin et al. (2005). The results of this study
2003) or to the estimation of free-water surface evaporation demonstrated that, in this semiarid region, this equation should not
(Valiantzas 2006). Therefore, the discussion of the results of this be chosen for the estimation of PETref .

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 219

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
 PETHam1 was found to produce reasonable results in the warm can be expected. Based on the results of this study, the large over-
and humid climates of the northeastern United States (Lu et al. estimates of PETref by these equations should limit their use.
2005), but it did not perform well in the warm and semiarid climate PETHan , a modification of the PETMak equation (Hansen 1984),
of Crete. The modification of this equation, herein referred to as was one of the best methods for the estimation of PETref values in
PETHam2 , performed quite well not only in Crete but also in the Switzerland (Xu and Singh 2000). In the semiarid environment of
regions studied by Oudin et al. (2005). On the other hand, PETHam3 Crete, PETHan performed very well, and it is surprising that such
underestimated PET rates in a wetland prairie in North Dakota good estimates of PETref were obtained using an equation origi-
(Rosenberry et al. 2004) and was among the temperature-based nally developed and tested in Nordic environments.
equations with the largest bias. The same results were obtained
for Crete; this equation resulted in even greater underestimations Pitfalls and Concerns
than PETHam1 . Therefore, its use in Crete is not recommended.
Although PETMcC is the simplest formula, it did not perform There were a number of drawbacks in relation to the methods
well in evaluations conducted in various climates (Oudin et al. adopted in this research. The most important was possibly that
2005), in Florida (Jacobs and Satti 2001), or in the semiarid the radiation-related inputs were not measured but calculated.
climate of Crete. Therefore, its use is not recommended. The estimation of input variables may result in errors in the final
PETMak has been widely used as an empirical equation for estimate of PETref (Llasat and Snyder 1998; Mahmood and
estimating PETref . However, in the southwestern United States Hubbard 2005) and results can be altered both qualitatively and
(Lu et al. 2005), it did not perform as well as other empirical equa- quantitatively if actual radiation data are used. However, as there
tions. In Switzerland (Xu and Singh 2000), PETMak did not perform is no meteorological station in Crete measuring any form of radi-
well for estimating of pan evaporation. PETMak was only moder- ation (global, net, etc.) the only approach available was to estimate
ately successful in a prairie wetland (Rosenberry et al. 2004), its fit incoming solar radiation using other meteorological observations
improving with some modifications to its coefficients. Similar as auxiliary variables.
results were obtained in a study of the PETMak equation in a small Additionally, there is a concern that some parameters of the
lake in the northcentral United States (Winter and Rosenberry PETFAO equation should be modified prior to its use in semiarid
1995). Alexandris et al. (2008) reported that PETMak produced environments (Berengena and Gávilan 2005; Gavilán et al. 2008).
the greatest underestimates of PETref of the six empirical equations These changes refer mainly to the use of different parameters
tested in the Balkan Peninsula. Finally, Oudin et al. (2005) found (coefficients) during the day and at night, as proposed by ASCE-
that PETMak performed better than the PETTur equation, yet ranked EWRI (ASCE-EWRI 2005). These modifications require hourly
it as being average overall. meteorological observations, which the meteorological stations
Although PETTur performed well in Crete, the results obtained in Crete do not provide. Nevertheless, for overcoming the lack
in other studies were mixed. Jacobs and Satti (2001) found that this of daily observations, ASCE-EWRI (2005) also proposed the
equation underestimates PETref in Florida throughout the year ex- use of the coefficients provided by Allen et al. (1998), as used
cept in summer. Xu and Singh (1998) also found that this equation in this study. The absence of hourly meteorological observations
significantly underestimated evaporation values in Switzerland. (low sampling frequency) also contributes to the relatively high
Underestimates, albeit small, were also reported by Alexandris bias in the final estimate of several parameters, including solar
et al. (2008) when this equation was applied in the Balkan radiation and PET (Hupet and Vanclooster 2001).
Peninsula. Oudin et al. (2005) did not recommend the use of this Other pitfalls related mostly to the quality and integrity of
equation for rainfall-runoff models because it failed to perform meteorological data were summarized by Hargreaves (2003).
well. By contrast, Trajkovic and Kolakovic (2009) found that Several authors proposed the adjustment of meteorological obser-
PETTur performed well in the humid western Balkans. This corre- vations in the case of nonreference conditions (Allen 1996;
sponds with the PETTur equation that was initially proposed for
Dinpashoh 2006; Jensen et al. 1997; Temesgen et al. 1999). These
the estimation of PETref in the humid climate of western Europe
procedures require values of relative wetness (precipitation/PET
(Turc 1961). Similarly, Douglas et al. (2009) also found that
ratio) (Dinpashoh 2006) or other factors (Temesgen et al. 1999)
PETTur provided good results. In arid and semiarid climates in
for all stations. Another source of uncertainty concerning the
India, PETTur performed better than other equations (Nandagiri
PETref values was the location of the meteorological stations. Most
and Kovoor 2006). In the same study, PETHar , which was also com-
of them were situated in coastal areas and were, therefore, subject
pared there, did not perform equally well. Lu et al. (2005) proposed
the use of PETTur for regional use in the southwestern United to large variations, especially in humidity-related values, depending
States. That this equation also performed well in Crete suggests on the wind direction. However, through the use of integrated
that it is adjustable to the local climatic conditions, most likely monthly values, these variations are smoothed out. The identifica-
because it requires RH data as an input. Although ASCE ranks tion of the sources of bias in the estimation of PETref requires more
PETTur below PETHar for application in semiarid and arid climates detailed information and sophisticated analyses exceeding the
(George et al. 2002), PETTur clearly outperforms PETHar under the aims and the scope of this study, that is, the evaluation of several
climatic conditions prevailing in Crete. empirical PETref equations within an operational framework. The
The PETJen equation performed well for the estimation of inte- meteorological stations may not reflect optimally the conditions of
grated monthly evaporation values in the northcentral United States the reference areas, yet the data they provide are sufficient for a
(Winter and Rosenberry 1995) and for evapotranspiration in comparative analysis of PET equations. Nevertheless, adjustments
the prairie wetlands of North Dakota (Rosenberry et al. 2004). may introduce further unpredictable bias into calculations (Howell
Xu and Singh (2000) reported that the uncalibrated PETJen resulted 2000; Temesgen et al. 1999), thus minimizing possible positive
in a large bias in the estimation of pan evaporation values in effects.
Switzerland. However, none of the latter three publications directly Unfortunately, all of these pitfalls could not be addressed in this
corresponds to the estimation of PETref . According to Oudin et al. study, owing mainly to the unavailability of the necessary data.
(2005), PETJen was one of the equations that performed best. Given Given the restrictions, and in spite of the expected bias, the authors
the close relationship between PETJen and PETCap , similar results are convinced that the optimum method was followed.

220 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Conclusions central Serbia, calculated by six empirical methods against the Penman-
Monteith formula.” J. Eur. Water Resour. Assoc., 21/22, 17–28.
There is no golden rule concerning the optimal equations to esti- Allen, R. G. (1996). “Assessing integrity of weather data for reference
mate PETref under various climates because even in the same evapotranspiration estimation.” J. Irrig. Drain Eng., 122(2), 97–106.
climatic type, different studies have produced mixed results in Allen, R., Pereira, L., Raes, D., and Smith, M. (1998). “Crop evapotran-
relation to the performance of the empirical PETref equations. spiration—Guidelines for computing crop water requirements.”
FAO Irrigation and Drainage, Paper No. 56, FAO, Rome, 300.
Additionally, because modifications to the parameters of some
ASCE Environmental and Water Resources Institute (ASCE-EWRI)
equations [see the example of PETHan and PETMak in Xu and Singh
(2005). “ASCE standardized reference evapotranspiration equation.”
(2002)] result in radically different output, suggests that a regional Standardization of Reference Evapotranspiration Task Committee
evaluation and calibration of the empirical PETref equations prior to Final Rep., Reston, VA, 59.
their use is needed. Berengena, J., and Gávilan, P. (2005). “Reference evapotranspiration
The radiation-based equations generally exhibited better fits estimation in a highly advective semiarid environment.” J. Irrig. Drain
than the temperature-based equations. PETHan and PETTur were Eng., 131(2), 147–163.
the most successful of the radiation-based equations, the latter Caprio, J. (1974). “The solar thermal unit concept in problems related
having a more complicated formula requiring RH data. This addi- to plant development ant potential evapotranspiration.” Phenology
tional input variable renders the PETHan a more favorable equation. Seasonality Model., 353–364.
On the other hand, PETJen , PETCap , PETMak , and PETdeB clearly Comrie, A. C., and Glenn, E. C. (1998). “Principal components-based
revealed a great bias and, therefore, their use for the estimation regionalization of precipitation regimes across the southwest United
States and northern Mexico, with an application to monsoon precipita-
of PETref in the semiarid climate of Crete is not recommended.
tion variability.” Clim. Resour., 10(3), 201–215.
With respect to the temperature-based equations, attention Croke, B., Cleridou, N., Kolovos, A., Vardavas, I., and Papamastorakis,
should be focused on the PETMcG and PETHam2 equations. They J. (2000). “Water resources in the desertification-threatened Messara
proved to be the best of these equations, although they were inferior Valley of Crete: Estimation of the annual water budget using a
to the best radiation-based equations for all stations except Iraklio rainfall-runoff model.” Environ. Model. Software, 15(4), 387–402.
and Rethimno. The latest resulted in a bias comparable with that of de Bruin, H. A. R., and Keijman, J. Q. (1979). “Priestley-Taylor evapora-
the radiation-based equations, rendering it an attractive option tion model applied to a large, shallow lake in the Netherlands.” J. Appl.
owing to its limited requirements in terms of input data. Although Meteorol., 18(7), 898–903.
the radiation-based equations generally performed better, one DehghaniSanij, H., Yamamoto, T., and Rasiah, V. (2004). “Assessment of
should not discount the fact that the temperature-based equations evapotranspiration estimation models for use in semi-arid environ-
require a lower number of input variables. This is important in ments.” Agric. Water Manage., 64(2), 91–106.
Dinpashoh, Y. (2006). “Study of reference crop evapotranspiration in IR of
situations where meteorological data are missing.
Iran.” Agric. Water Manage., 84(1-2), 123–129.
It is also important that the climatic parameters of the meteoro-
Douglas, E. M., Jacobs, J. M., Sumner, D. M., and Ray, R. L. (2009).
logical stations should be carefully taken into consideration to “A comparison of models for estimating potential evapotranspiration
choose the best empirical equation. To exemplify, two stations for Florida land cover types.” J. Hydrol., 373(3-4), 366–376.
(Rethimno and Iraklio) exhibited high PETref values as a result Droogers, P., and Allen, R. G. (2002). “Estimating reference evapotranspira-
of high wind speeds. At these stations, the temperature-based tion under inaccurate data conditions.” Irrig. Drain. Syst., 16(1), 33–45.
PETMcG equation outperformed the other equations yet the bias Gavilán, P., Estevez, J., and Berengena, J. (2008). “Comparison of stand-
was high. ardized reference evapotranspiration equations in Southern Spain.”
Although there were some factors that could possibly result in a J. Irrig. Drain Eng., 134(1), 1–12.
bias in the estimation of PETref , the method followed was robust Gavilán, P., Lorite, I. J., Tornero, S., and Berengena, J. (2006). “Regional
and efficient and the results of some of the empirical equations, as calibration of Hargreaves equation for estimating reference ET in a
described in the aforementioned sections, were found to compare semiarid environment.” Agric. Water Manage., 81(3), 257–281.
George, B. A., Reddy, B. R. S., Raghuwanshi, N. S., and Wallender, W. W.
reasonably well with those of the PETFAO equation. These formulas
(2002). “Decision support system for estimating reference evapotran-
can be applied where a lack of data limits the use of PETFAO or spiration.” J. Irrig. Drain Eng., 128(1), 1–10.
where there are no current measurements of PETref . Giles, B. D., Balafoutis, C. J., and Maheras, P. (1989). “Regionalization of
potential and actual evapotranspiration in Greece as revealed by prin-
cipal component analysis.” Z. Meteorol., 39(3), 150–156.
Acknowledgments
Grismer, M. E., Orang, M., Snyder, R., and Matyac, R. (2002). “Pan evapo-
The authors would like to thank Dr. David Butler-Manning for ration to reference evapotranspiration conversion methods.” J. Irrig.
Drain Eng., 128(3), 180–184.
proofreading the manuscript. We thank also the Hellenic National
Grove, A. T., and Rackham, O. (1993). “Threatened landscapes in the
Weather Service for providing us the climate data of Crete. To this Mediterranean—Examples from Crete.” Landscape Urban Plann.,
end we would also like to thank the two anonymous reviewers and 24(1-4), 279–292.
the editor of the Journal of Irrigation and Drainage Engineering— Hansen, S. (1984). “Estimation of potential and actual evapotranspiration.”
ASCE for their valuable comments which significantly contributed Nordic Hydrol., 15, 205–212.
to the improvement of this publication. Hargreaves, G. H. (2003). “Discussion of ‘Decision support system for
estimating reference evapotranspiration’ by B. A. George, B. R. S.
References Reddy, N. S. Raghuwanshi, and W. W. Wallender.” J. Irrig. Drain
Eng., 129(5), 381–382.
Alexandris, S., and Kerkides, P. (2003). “New empirical formula for hourly Howell, T. A. (2000). “Adjusting temperature parameters to reflect well-
estimations of reference evapotranspiration.” Agric. Water Manage., watered conditions—Discussion.” J. Irrig. Drain Eng., 126(5), 340–342.
60(3), 157–180. Hupet, F., and Vanclooster, M. (2001). “Effect of the sampling frequency of
Alexandris, S., Kerkides, P., and Liakatas, A. (2006). “Daily reference meteorological variables on the estimation of the reference evapotran-
evapotranspiration estimates by the Copais approach.” Agric. Water spiration.” J. Hydrol., 243(3–4), 192–204.
Manage., 82(3), 371–386. Irmak, S., Allen, R. G., and Whitty, E. B. (2003). “Daily grass and alfalfa-
Alexandris, S., Stricevic, R., and Petkovic, S. (2008). “Comparative analy- reference evapotranspiration estimates and alfalfa-to-grass evapotran-
sis of reference evapotranspiration from the surface of rain-fed grass in spiration ratios in Florida.” J. Irrig. Drain Eng., 129(5), 360–370.

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011 / 221

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org
Jabloun, M., and Sahli, A. (2008). “Evaluation of FAO-56 methodology Environmental Science and Technology, Lemnos Island, Greece,
for estimating reference evapotranspiration using limited climatic data Poster Presentations, 650–658.
Application to Tunisia.” Agric. Water Manage., 95(6), 707–715. Pennas, P. I. (1977). “The climate of Crete [in Greek].” Ph.D., Aristotle
Jacobs, J. M., and Satti, S. R. (2001). “Evaluation of reference evapotran- Univ. of Thessaloniki, Thessaloniki, Greece.
spiration methodologies and AFSIRS crop water use simulation Popova, Z., Kercheva, M., and Pereira, L. S. (2006). “Validation of the FAO
model.” Final rep., St. Johns River Water Management District, methodology for computing ETo with limited data. Application to south
Palatka, FL. Bulgaria.” Irrig. Drain. Syst., 55(2), 201–215.
Jacovides, C. P. (1998). “Reply to comment on statistical procedures for the Rosenberg, N., Blad, B., and Verma, S. (1983). Microclimate: The
evaluation of evapotranspiration computing models.” Agric. Water biological environment, Wiley, New York.
Manage., 37(1), 95–97. Rosenberry, D. O., Stannard, D. I., Winter, T. C., and Martinez, M. L.
Jacovides, C. P., and Kontoyiannis, H. (1995). “Statistical procedures for (2004). “Comparison of 13 equations for determining evapotranspira-
the evaluation of evapotranspiration computing models.” Agric. Water tion from a prairie wetland, Cottonwood Lake Area, North Dakota.”
Manage., 27(3-4), 365–371. J. Soc. Wetlands Sci., 24(3), 483–497.
Jensen, D. T., Hargreaves, G. H., Temesgen, B., and Allen, R. G. (1997). StatSoft, I. (2007). STATISTICA data analysis software system.
“Computation of ETo under non-ideal conditions.” J. Irrig. Drain Eng., [Computer software], Tulsa, OK.
123(5), 394–400. Stöckle, C. O., Kjelgaard, J., and Bellocchi, G. (2004). “Evaluation of esti-
Jensen, M. E., and Haise, H. R. (1963). “Estimating evapotranspiration mated weather data for calculating Penman-Monteith reference crop
from solar radiation.” J. Irrig. Drain Eng., 89, 15–41. evapotranspiration.” Irrig. Sci., 23(1), 39–46.
Knotters, M., and Voshaar, J. H. O. (1998). “Comment on ‘Statistical Temesgen, B., Allen, R. G., and Jensen, D. T. (1999). “Adjusting temper-
procedures for the evaluation of evapotranspiration computing models’ ature parameters to reflect well-watered conditions.” J. Irrig. Drain
by C. P. Jacovides, and H. Kontoyiannis.” Agric. Water Manage., Eng., 125(1), 26–33.
37(1), 93–94. Temesgen, B., Eching, S., Davidoff, B., and Frame, K. (2005). “Compari-
Lambrakis, N., and Kallergis, G. (2001). “Reaction of subsurface coastal son of some reference evapotranspiration equations for California.”
aquifers to climate and land use changes in Greece: Modelling of J. Irrig. Drain Eng., 131(1), 73–84.
groundwater refreshening patterns under natural recharge conditions.” Trajkovic, S. (2005). “Temperature-based approaches for estimating refer-
J. Hydrol. (Amsterdam), 245(1-4), 19–31. ence evapotranspiration.” J. Irrig. Drain Eng., 131(4), 316–323.
Llasat, M. C., and Snyder, R. L. (1998). “Data error effects on net radiation Trajkovic, S., and Kolakovic, S. (2009). “Evaluation of reference
and evapotranspiration estimation.” Agricul. Forest Meteorol., 91(3-4), evapotranspiration equations under humid conditions.” Water Resour.
209–221. Manage., 23(14), 3057–3067.
Lu, J. B., Sun, G., McNulty, S. G., and Amatya, D. M. (2005). “A com- Tsakiris, G., and Tigkas, D. (2007). “Drought risk in agriculture in
parison of six potential evapotranspiration methods for regional use in Mediterranean regions. Case study: Eastern Crete.” Methods and
the southeastern United States.” J. Am. Water Resour. Assoc., 41(3), tools for drought analysis and management, G. Rossi, T. Vega, and
621–633. B. Bonaccorso, eds., Springer, New York, 399–414.
Mahmood, R., and Hubbard, K. G. (2005). “Assessing bias in evapotran- Tsakiris, G., Tigkas, D., Vangelis, H., and Pangalou, D. (2007). “Regional
spiration and soil moisture estimates due to the use of modeled solar drought identification and assessment. Case study in Crete.” Methods
radiation and dew point temperature data.” Agricul. Forest Meteorol., and tools for drought analysis and management, G. Rossi, T. Vega, and
130(1-2), 71–84. B. Bonaccorso, eds., Springer, New York, 169–191.
Martínez-Cob, A. (1996). “Multivariate geostatistical analysis of evapo- Tsanis, I. K., and Naoum, S. (2003). “The effect of spatially distributed
transpiration and precipitation in mountainous terrain.” J. Hydrol. meteorological parameters on irrigation water demand assessment.”
(Amsterdam), 174(1-2), 19–35. Adv. Water Resour., 26(3), 311–324.
Martínez-Cob, A., and Tejero-Juste, M. (2004). “A wind-based qualitative Turc, L. (1961). “Evaluation des besoins en eau d’irrigation.” Évapotran-
calibration of the Hargreaves ET0 estimation equation in semiarid spiration Ann. Agron., 12, 13–49.
regions.” Agric. Water Manage., 64(3), 251–264. Valiantzas, J. D. (2006). “Simplified versions for the Penman evaporation
Matzarakis, A., and Rutz, F. (2007). “RayMan—Modeling of mean radiant equation using routine weather data.” J. Hydrol. (Amsterdam),
temperature in urban structures—Calculation of thermal indices.” 331(3–4), 690–702.
Freiburg, Germany. Vardavas, I. M., Papamastorakis, J., Fountoulakis, A., and Manousakis, M.
Matzarakis, A., Rutz, F., and Mayer, H. (2007). “Modeling radiation fluxes (1997). “Water resources in the desertification-threatened Messara
in simple and complex environments—Application of the RayMan Valley of Crete: Estimation of potential lake evaporation.” Ecol. Model.,
model.” Int. J. Biometeorol., 51(4), 323–334. 102(2-3), 363–374.
McCune, B., and Mefford, M. J. (1999). PC-ORD. Multivariate analysis Wang, Y. M., Traore, S., and Kerh, T. (2009). “Computational performance
of ecological data [Computer software]. MjM Software, Gleneden of reference evapotranspiration in semiarid zone of Africa.” Sci. Res.
Beach, OR. Essays, 4(6), 577–583.
Michalopoulou, H., and Papaioannou, G. (1991). “Reference crop evapo- Willmott, C. J. (1982). “Some comments on the evaluation of model
transpiration over Greece.” Agric. Water Manage., 20(3), 209–221. performance.” Bull. Am. Meteorol. Soc., 63(11), 1309–1313.
Morin, G., Fortin, J. P., Sochanska, W., and Lardeau, J. P. (1979). “Use Willmott, C. J. (1984). “On the evaluation of model performance in physi-
of principal component analysis to identify homogeneous precipita- cal geography.” Spatial statistics and models, G. L. Gaile and
tion stations for optimal interpolation.” Water Resour. Res., 15(6), C. J. Willmott, eds., Reidel, Boston, 443–460.
1841–1850. Willmott, C. J., et al. (1985). “Statistics for the evaluation and comparison
Nandagiri, L., and Kovoor, G. M. (2006). “Performance evaluation of refer- of models.” J. Geophys. Res. Oceans, 90(C12), 8995–9005.
ence evapotranspiration equations across a range of Indian climates.” Winter, T. C., and Rosenberry, D. O. (1995). “Evaluation of 11 equations
J. Irrig. Drain Eng., 132(3), 238–249. for determining evaporation for a small lake in the northcentral
Naoum, S., and Tsanis, I. K. (2003). “Hydroinformatics in evapotranspi- United-States.” Water Resour. Res., 31(4), 983–993.
ration estimation.” Environ. Model. Software, 18(3), 261–271. Xu, C., and Singh, V. (2000). “Evaluation and generalization of radiation-based
Oudin, L., et al. (2005). “Which potential evapotranspiration input for a methods for calculating evaporation.” Hydrol. Processes, 14(2), 339–349.
lumped rainfall-runoff model?: Part 2–Towards a simple and efficient Xu, C. Y., and Singh, V. P. (1998). “Dependence of evaporation on
potential evapotranspiration model for rainfall-runoff modeling.” meteorological variables at different time-scales and intercomparison
J. Hydrol. (Amsterdam), 303(1/4), 290–306. of estimation methods.” Hydrol. Processes, 12(3), 429–442.
Papadopoulou, E., Varanou, E., Baltas, E., Dassaklis, A., and Mimikou, Xu, C. Y., and Singh, V. P. (2002). “Cross comparison of empirical
M. (2003). “Estimating potential evapotranspiration and its spatial dis- equations for calculating potential evapotranspiration with data from
tribution in Greece using empirical methods.” Proc,. 8th Int. Conf. on Switzerland.” Water Resour. Manage., 16(3), 197–219.

222 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / APRIL 2011

Downloaded 10 Apr 2011 to 140.116.210.6. Redistribution subject to ASCE license or copyright. Visithttp://www.ascelibrary.org