CHAPTER 13 Reference evapotranspiration in the River Azul Basin, Argentina Rau ´ l Rivas* Instituto de Hidrologı ´a de Llanuras, Comisio ´n de Investigaciones Cientı ´ficas de la Provincia de Buenos Aires, Universidad Nacional del Centro de la Provincia de Buenos Aires Vicente Caselles Departamento de Termodina ´mica, Facultad de Fı ´sica, Universidad de Valencia, Espaþa Eduardo UsunoffInstituto de Hidrologı ´a de Llanuras, Comisiœn de Investigaciones Cientı ´ficas de la Provincia de Buenos Aires, Universidad Nacional del Centro de la Provincia de Buenos Aires ABS TRACT: Re mot e sensin g is among the lat est tech niq ues to ass ess the spati al and time- depe ndent variati ons of evap otra nspiration (ET). The newe st mode ls base d on energ y bala nce approaches are able to come up with precise estimates of ET at any point of the time-space domain. However, those models have to be fed by information that is not always available. This paper presents a simpler approach that allows the estimation of ET by combining satellite information and data from a weather station. The model needs to rely on images of surface temperature (T s ) and to estimate ET from regular meteorological data and methods. Its application to the River Azul Basin 600 0 km 2 in the centre of Buenos Aires Province, Argentina led to a precision of 0.4 mm/ day ( 11 W/m 2 ) with a correlation coefficient (r 2 ) of 0.88. The model was applied to an image of T s and the results compared with what was obtained from data of a reference weather stati on. The model gave an estimate of 4.2 mm/d ay (119 W/m 2 ), whereas the conventional method ren der ed a val ue of 4.5 mm/da y (12 8 W/m 2 ). It is concluded that the proposed approach is able to deliver reliable data on the spatial and temporal variations of ET in areas where the availability ofdata is scarce. INTRODUCTION Input data for hydrological regional models consist of a large number of variables for several time periods. In order to describe properly the features of large basins, such hydrological models should be distributed, that is, able to take into account the spatial distribution of variables (Sandholt et al., 2002). Evapot rans pi rati on (ET) is one of the most important output vari ables fr om hydrological models as far as its distribution in time and space is concerned. Classic methodologies for ET estimation do not handle properly such time-space variations. ET *[email protected]
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Reference evapotranspiration in the River Azul Basin,
Argentina
Raul Rivas* Instituto de Hidrologı a de Llanuras, Comisio n de Investigaciones Cientı ficas de la
Provincia de Buenos Aires, Universidad Nacional del Centro de la Provincia de Buenos Aires
Vicente Caselles Departamento de Termodina mica, Facultad de Fı sica, Universidad de Valencia, Espaþa
Eduardo Usunoff Instituto de Hidrologı a de Llanuras, Comisiœn de Investigaciones Cientı ficas de la
Provincia de Buenos Aires, Universidad Nacional del Centro de la Provincia de Buenos Aires
ABSTRACT: Remote sensing is among the latest techniques to assess the spatial and time-
dependent variations of evapotranspiration (ET). The newest models based on energy balanceapproaches are able to come up with precise estimates of ET at any point of the time-space domain.
However, those models have to be fed by information that is not always available. This paper
presents a simpler approach that allows the estimation of ET by combining satellite information
and data from a weather station. The model needs to rely on images of surface temperature (Ts) and
to estimate ET from regular meteorological data and methods. Its application to the River Azul
Basin 6000 km2 in the centre of Buenos Aires Province, Argentina led to a precision of
0.4 mm/day (11 W/m2) with a correlation coefficient (r2) of 0.88. The model was applied to an
image of Ts and the results compared with what was obtained from data of a reference weather
station. The model gave an estimate of 4.2 mm/day (119 W/m2), whereas the conventional method
rendered a value of 4.5 mm/day (128 W/m2
). It is concluded that the proposed approach is able todeliver reliable data on the spatial and temporal variations of ET in areas where the availability of
data is scarce.
INTRODUCTION
Input data for hydrological regional models consist of a large number of variables for
several time periods. In order to describe properly the features of large basins, such
hydrological models should be distributed, that is, able to take into account the spatialdistribution of variables (Sandholt et al., 2002).
Evapotranspiration (ET) is one of the most important output variables from
hydrological models as far as its distribution in time and space is concerned. Classic
methodologies for ET estimation do not handle properly such time-space variations. ET
estimates are obtained at discrete sites and their regionalization is made on the basis of
interpolation, which is not representative of what actually occurs.
More recently, models for ET estimation that solve the energy balance equation from
remote sensing data have been developed (Boegh et al., 2002). They make it possible to
obtain the space and time variations of ET by carrying out energy balances at soil level(Schmugge et al., 2001). The remote sensing technique becomes a useful tool for retrieving
information on the energy and water balance, and is able to detect changes at the regional
or even the parcel levels. The unit area for which the balance is carried out depends on the
spatial resolution of the sensor used. For instance, for the Advance Very High Resolution
Radiometer (AVHRR) and Moderate Resolution Imaging Spectroradiometer (MODIS),
the working unit is one km2; for the LANDSAT Thematic Mapper (TM) 5 the area is
14,400 m2 and 3600 m2 for the LANDSAT-ETMþ7. Thus, remote sensing thoughtfully
employed promotes an adequate knowledge of water losses from the system, which, in
turn, can be used to improve the management of regional water resources.Energy balance-driven models need to count on net radiation data, which are a type of
information that are not always readily available. This paper aims at assessing the spatial
variation of ET in an area of scarce data. To do so, a linear model for determining the
reference ET from surface temperature data (Ts) at the pixel level was applied. The
model is based on the split window approach (Coll & Caselles, 1997).
The model requires the availability of enough meteorological information to compute
ET with the Penman-Monteith equation, as well as series of Ts images to adjust its
parameters.
A detailed description of the steps followed to adjust the model parameters ispresented and its application to the River Azul Basin is shown. The water balance in such
a basin, located at the centre of Buenos Aires Province, Argentina, indicates that ET is
the single largest component inasmuch as it accounts for about 90 per cent of the water
that enters the system as annual rainfall (Usunoff et al., 1999).
METHODOLOGY
A linear model that combines remote sensing data with the Penman-Monteith evapo-
transpiration equation was used to estimate the reference ET (ETo mm/day) at the pixel
level. The model is generated by splitting the Penman-Monteith equation into two terms:
the radiation and the aerodynamic components. Rearranging the equation leads to the
following linear model:
ET o ¼ a T s þ b ð1Þ
where Ts is the surface temperature (C), and a (mm/day/ C) and b (mm/day) are
parameters specific to the study area.The a parameter represents the radiation effects on the surface and is quite sensitive to
changes in the wind speed. The b parameter takes into account the effect of meteoro-
logical conditions on the surface.
The model assumes a flat surface covered by vegetation 0.12 m high and with enough
water availability. Besides, the meteorological variables are supposed to show spatial
where ETo is the reference evapotranspiration (mm/day), Rn is the net radiation (MJ/m2 /
day), G is the soil heat flux density (MJ/m2 /day) (disregarded because of the time scale
considered), g is the psychrometric constant (kPa/ C), Ta is the air temperature at 2 m
height (C), U2 is the wind speed at 2 m height (m/s), es is the saturation vapour pressure
(kPa), ea
is the actual vapour pressure (kPa), and
is slope of the water vapour pressurecurve (kPa/ C).
Before calculating ETo, a consistency analysis of the meteorological variables recorded
at the weather station was made. To do this, data from the reference weather station were
compared (homogeneity check) with data from three other weather stations in the area.
The variables analysed and compared were Ta max, Ta min, U2, n, HRmax and HRmin at daily
time intervals and for the same time period. The mean, standard deviation, covariance,
regression equations, correlation coefficients and residuals were calculated and evaluated.
Those values that were deemed not consistent were eliminated from the records.
ETo was calculated using meteorological data from the reference weather station(Azul) for the period 19962000. Mean weekly estimates of ETo were obtained by
processing the satellite images.
Calculation of Ts
Ts was calculated using the split window approach, which accounts for the atmospheric
and emissivity correction based on the difference in atmospheric absorption recorded in
the radiances of bands 4 and 5 of the AVHRR sensor (Coll & Caselles, 1997). Radiation
measured in such bands were converted into brightness temperature (noted as T4 and T5)
by using the calibration equation provided for the sensor.
Estimates of Ts used the quadratic split window equation proposed by Coll and
Caselles (1997):
T s ¼ T 4 þ 1:34 þ 0:39 T 4 T 5ð Þ½ T 4 T 5ð Þ þ 0:56 þ 1 "ð Þ " ð3Þ
where Ts is the surface temperature (K) in the pixel, T4 and T5 are the brightness
temperatures from bands 4 and 5 of NOAA-AVHRR (K), a and b are coefficients that
depend on the water vapour content in the atmosphere (K), e is the surface emissivity inthe range 10.512.5 mm, and e is the difference in emissivity between bands 4 and 5.
The algorithm is simple in that it requires knowledge of the water vapour content in
the atmosphere to determine the a and b coefficients. The standard error for Ts is
1.01.5 K for atmospheres with low to moderate humidity.
The a and b parameters in Equation 3 correspond to atmospheres of medium latitude
in winter and summer. Inasmuch as a does not vary with the water vapour content in the
atmosphere, a value of 50 K was employed. The b coefficient decreases linearly with the
water vapour content of the atmosphere and values of b¼ 75 K for summer and
b¼ 150 K for winter were adopted.e was assessed from the equation proposed by Valor and Caselles (1996) for the
spectral range of 10.512.5mm:
" ¼ "v Pv þ "s ð1 PvÞ þ 4 d" Pv ð1 PvÞ ð4Þ
where e is the effective emissivity of a surface made up of soil and vegetation, ev is the
vegetation emissivity, Pv is the percentage of vegetation cover, es is the soil emissivity
and de is the cavity term that takes into account the internal reflections between the
vegetation and the soil.
The percentage vegetation cover (Pv) was calculated from the NDVI using the equation
obtained from a linear model of reflectivity with two components (soil and vegetation) and
two bands (red and near infrared), according to Valor and Caselles (1996):
Pv ¼ð1 NDVI
NDVIsÞ
ð1 NDVINDVIs
Þ K ð1 NDVINDVIv
Þ ð5Þ
being K ¼ IRCv Rv
IRCs Rs
where IRCv is the vegetation reflectivity in the near infrared, IRCs is the soil reflectivity
in the near infrared, Rv is the vegetation reflectivity in the red, Rs is the soil reflectivity inthe red, NDVIv corresponds to pure vegetation, NDVIs corresponds to pure soil and
NDVI is the vegetation index of any considered mixed pixel.
For the study area, Equation 4 reduces to the first two terms because no internal
reflections are expected due to the type of vegetation cover. Indeed, the vegetation cover
consists mainly of natural pastures and agricultural crops (wheat, corn, soybean) whose
cavity effects are minimal.
For the determination of e the following values were used in Equation 4: 0.985 for
vegetation emissivity and 0.96 for bare soil emissivity. The ev value matches field data
from Rubio et al. (1997), whereas the es value corresponds to data reported by Salisburyand D’Aria (1992, 1994).
Ts estimates used 21 images captured by the AVHRR sensor during the period
19922000, of 1.1 km2 pixel size for different seasons throughout the year. The selection
of weekly images took into account the following constraints: that the maximum daily air
temperature at the reference weather station had to differ at least 2C from the mean of
the weekly maximum temperatures, it had to be a cloudless day (n ffi N) and there had to
be no rainfall in the previous days. Images of dry and humid periods were used in order
to cover the maximum and minimum values for Ts. Each image selected and processed
represented the mean weekly Ts.Using weekly images increases the probability of making use of high-quality
radiometric data, which is essential for the Ts estimation.
The Ts value was taken as the average of an area of 3 by 3 pixels, where the central
pixel corresponded to the location of the reference station. Such an average comes from
the geometric correction of the image, whose error is 1 pixel.
Determination of parameters a
and b
To determine a and b, it is imperative to know Ts and ETo for the area where the
reference weather station is located. The expression for a and b comes from the following
equations, which assume that the Penman-Monteith equation can be written as the sum of
two terms: aerodynamic (ETaero) and radiation (ETradio):
ET o ¼ ET aero þ ET radio ¼ a T s þ b ð6Þ
Reference evapotranspiration in the River Azul Basin, Argentina 163
By rearranging Equation 6 it is possible to obtain explicit equations for a and b:
a ¼ 0:408 oradio "s s A ð7Þ
b ¼ ET aero þ 0:77 Rs þ 0:408 ½"s s ð B "a ðT aÞ4Þ oradio ð8Þ
where oradio (defined below) is in mm m2 /MJ, es is the surface emissivity, s is the StefanBoltzmann constant (4.9 10
9 MJ/m2 /K 4 /day), A is equal to 1.1 108 K 3, 0.408 is the
conversion factor to mm/day, ETaero is the aerodynamic term of Penman-Monteith
equation (mm/day), RS is the solar radiation (MJ/m2 /day), 0.77 (1 a) is a value that
comes from assuming a surface albedo a of 0.23, B equals 2.46 1010 K 4, ea is the air
emissivity and Ta is the air temperature (K). oradio has the following form:
oradio ¼
þ g 1 þ 0:34 U 2ð Þ ð9Þ
where is the slope vapour pressure curve, (kPa/ C), g is the psychrometric constant(kPa/ C), 0.34 is a conversion factor for g and U2 is the wind speed at 2 m height (m/s).
The a parameter represents the radiation effects on the surface and is quite sensitive to
changes in the wind speed. For a given radiation, as the wind increases, a decreases
linearly. The b parameter represents the effects of the meteorological conditions, being
sensitive to air temperature and relative humidity, to wind speed and to sunshine. In
general, b decreases linearly as the air temperature and relative humidity increase.
Likewise, b shows a logarithmic behaviour as the wind speed increases and increases
linearly as sunshine increases. The sunshine effect is minimum since counting on good
images implies cloudless days.Figure 2 depicts a simplified scheme used to obtain the parameters at the reference
weather station.
Values for the a and b parameters were obtained from the TsETo data at the reference
weather station, given all the images available and a linear fit using least squares.
Having fitted the a and b parameters, Equation 1 was applied to a Ts image corres-
ponding to the second week of January 1996. Finally, the ETo calculated with data from
the weather station was compared to that given by the model (average of a 3 by 3 pixels).
The same procedure was applied for each Ts image used.
Study area
The study area corresponded to the River Azul Basin, located at the centre of Buenos
Aires Province, Argentina (Figure 3). It covers about 6000 km2 of Argentina’s humid
pampas, essentially a flat landscape. The mean regional slope is less than 1 per cent, even
reaching values as low as 0.2 per cent at the northern portion (Varni & Usunoff, 1999).
The climate is humid, with low to zero water deficit, a mean annual precipitation of
950 mm and a mean annual ET of 915 mm (Usunoff et al., 1999; Varni et al., 1999).
RESULTS
Figure 4 displays the resulting Pv by applying the linear model of reflectivity with two
components from the image of the second week of January 1996. It can be observed that
the basin shows a high percentage of vegetated areas.
The model presented allows the estimation of weekly ETo from the Ts information
retrievable from a AVHRR-NOAA sensor. Images of Ts should be available, as well as
meteorological data to calculate ETo.
Nowadays, series of satellite images to calculate Ts are readily available and the
meteorological information required by the model is rather modest, so its application can
be easily implemented.
The procedure to determine the model’s parameters is simple and from them it is
relatively easy to calculate ETo at the regional level.
The methodology proposed here leads to the assessment of the spatial and time-relatedvariations of ETo, which can be picked up later on as input for distributed hydrological
models. As a matter of fact, that was the main objective of this project.
As an example, the model has been applied to the River Azul Basin, Buenos Aires
Province, Argentina. The parameters were fitted by a function with an r2 of 0.88 and an
estimation error of 0.4 mm/day (11 W/m2).
By using a Ts image of the study area, the mean weekly ETo turned out to be 4.5 mm/
day (128 W/m2), whereas the corresponding value at the reference weather station was
4.2 mm/day (119 W/m2
), values that are close enough to indicate a good performance of
the model.
ACKNOWLEDGEMENTS
The work was carried with financial support from the Comision de Investigaciones
Cientıficas de Buenos Aires, the Universidad Nacional del Centro de la Provincia de
60ºW
Azulstation 1
59ºW
N
60ºW
2 7 mm/day4.5
59ºW0 50
km
3 7 º S
3 7 º S
Figure 9. Mean daily ETo (mm/day) obtained from the model.
Reference evapotranspiration in the River Azul Basin, Argentina 169
Buenos Aires, the Ministerio de Ciencia y Tecnologıa de Espana (Contract REN-2001-
3116/CLI), and the European Commission (FEDER). We would also like to thank the
Azul weather station staff (Servicio Meteorologico Nacional).
REFERENCES
Allen RG, Pereira LS, Raes D and Smith M. 1998. Crop evapotranspiration Guidelines forcomputing crop water requirements-FAO. Irrigation and Drainage Paper 56. Water Resources,Development and Management Service, Rome.
Boegh E, Soegaard H and Thomsen A. 2002. Evaluating evapotranspiration rates and surfaceconditions using Landsat TM to estimate atmospheric resistance and surface resistance. RemoteSensing of Environment 79, 32943.
Coll C and Caselles V. 1997. A split window algorithm for land surface temperature from advancedvery high resolution radiometer data: Validation and algorithm comparison. Journal of Geophysical Research, 102 (14), 16697713.
Rubio E, Caselles V and Badenas C. 1997. Emissivity measurements of several soils and vegetationtypes in the 8–14 mm waveband. Analysis of two field methods. Remote Sensing of Environment
59, 490521.Salisbury J and D’Aria D. 1992. Emissivity of terrestrial materials in the 814 mm atmospheric
window. Remote Sensing of Environment 42, 83106.Salisbury J and D’Aria D. 1994. Emissivity of terrestrial materials in the 35 mm atmospheric
window. Remote Sensing of Environment 47, 34561.Sandholt I, Rasmussen K and Andersen J. 2002. A simple interpretation of the surface temperature/
vegetation index space for assessment of surface moisture status. Remote Sensing of
Environment 79, 21324.Schmugge T, French A and Kustas W. 2001. Evapotranspiration estimates using ASTER thermal
infrared imagery. In: Proceedings of SPIE EurOpto 8th International Symposium on RemoteSensing, Toulouse, France.
Usunoff E, Varni M, Weinzettel P and Rivas R. 1999. Hidrogeologıa de grandes llanuras: Lapampa humeda argentina [Hydrogeology in large plains: Argentina’s humid pampas]. Boletı nGeolo gico y Minero 110 (4), 391406.
Valor E and Caselles V. 1996. Mapping land surface emissivity from NDVI: Application toEuropean, African and South American areas. Remote Sensing of Environment 57, 16784.
Varni M and Usunoff E. 1999. Simulation of regional-scale groundwater flow in the River Azulbasin, Buenos Aires Province, Argentina. Hydrogeology Journal 7 (2), 1807.
Varni M, Usunoff E, Weinzettel P and Rivas R. 1999. The groundwater recharge in the Azulaquifer, central Buenos Aires Province, Argentina. Physics and Chemistry of the Earth 24 (4),34952.