Comparative Study of Reference Evapotranspiration Estimation Methods Including Artificial Neural Network For Dry Sub-Humid Agro-Ecological Region
Comparative Study of Reference Evapotranspiration Estimation Methods Including Artificial Neural Network For Dry Sub-Humid Agro-Ecological Region
Comparative Study of Reference Evapotranspiration Estimation Methods Including Artificial Neural Network For Dry Sub-Humid Agro-Ecological Region
233
ABSTRACT
In the present study, an attempt has been made to compare the reference evapotranspiration (ETo),
computed by eight different methods, namely, Penman-Monteith, Modified Penman-Monteith,
Hargreaves-Samani, Irmak, Hargreaves, Valiantzas, ANN and FAO(24) model for the dry subhumid agro-ecological region (Varanasi). An attempt was also made to find out utility of artificial
neural networks (ANN) for estimation of ET0 with minimum input. Feed forward network has
been used for prediction of ETo using resilient back-propagation method and the architecture 2-21(having parameters Tmean and solar radiation) was found to be the best one. The average annual
evapotranspiration (by PenmanMonteith method) for Varanasi was found to be 1447.4 mm. When
compared among the different methods for estimation of reference evapotranspiration with PenmanMonteith method, the FAO-24 and Hargraves-Samani (3) under estimate, however, Modified
Penman-Monteith, Hargreaves-Samani, Irmak, Hargreaves, Valiantzas over-estimate and ANN
closely estimates reference evapotranspiration.
INTRODUCTION
Reference evapotranspiration (ET o ) is an
important parameter of the hydrologic cycle and
its accurate estimation is important for many
studies, such as hydrologic water balance,
irrigation scheduling, water resources planningmanagement and for hydrologic and climatic
studies. Evapotranspiration (ET) is the combined
process of evaporation from the soil and plant
surfaces and transpiration through the stomata of
the plant surface. The evapotranspiration rate from
a reference surface having no shortage of water is
called the reference crop evapotranspiration (ETo).
The reference surface is a hypothetical grass with
specific characteristics. The concept of the reference
evapotranspiration was introduced to study the
evaporative demand independent from crop type,
crop development and management practices. The
only factors affecting ETo are climatic parameters
and can be computed from weather data. Many
equations are used to estimate ETo and can be
divided into two main groups, those that are
physically based like Penman, Penman-Monteith
1 Institute of Agricultural Sciences, Banaras Hindu University, Varanasi-221005 (U.P.); 2College of Agricultural Engineering &
Technology & Research (IJKVV) Mungali (C.G.); Email: anupamnema@ gmail.com
234
KAR et al.
where,
ETo = Reference evapotranspiration [mm day-1]
Rn = Net radiation at the crop surface [MJm-2 day-1]
G = Soil heat flux density [MJ m-2 day-1]
(3)
July-September 2016]
235
236
Performance Criteria
Mean Square Error
(MSE)
KAR et al.
Formula
Remarks
An MSE of zero, means that the Pi predicts
observation of the parameter Ei with perfect
accuracy.
Flemming (1975)
It always lies between -1 to +1, if r =1, the
correlation coefficient is perfect and positive
and if r = -1, then it is perfect and negative.
Coefficient of determination (r2) is a statistical
measure of how well the regression line
approximates the real data points. An r2 of 1
indicates that the line of regression is of best fit.
Coefficient of
determination (r2)
Table 1. Comparison of various neural network predicted ETo data with FAO-56 (PM) estimated ETo
ANN architecture
Inputs
No of neurons
Input
Hidden
Output
Nodes in hidden layer
MSE
RMSE
Coefficient of
correlation (r)
Coefficient of
determination (r2)
Relative Error
(Holdout)
4-3-1
Tmean,
Solar
radiation,
VPD, U2
3-3-1
Tmean,
Solar
radiation
(Rs), VPD
2-2-1
Tmean,
Solar
radiation
(Rs)
1-9-1
Tmean
3-5-1
Solar
radiation
(Rs),
( VPD, U2)
2-6-1
Tmean,
U2
2-3-1
Solar
radiation
(Rs), VPD
4
1
1
3
0.003
0.053
0.992
3
1
1
3
0.015
0.124
0.989
2
1
1
2
0.017
0.130
0.988
1
1
1
9
0.624
0.790
0.920
3
1
1
5
0.024
0.156
0.985
2
1
1
6
0.129
0.359
0.959
2
1
1
3
0.185
0.430
0.944
0.985
0.978
0.976
0.846
0.970
0.920
0.892
0.001
0.006
0.005
0.089
0.009
0.034
0.061
where, Tmean= Daily mean temperature (C), Solar radiation (Rs) on daily basis, Vapour Pressure Deficit (VPD) = es-ea,
U2= Wind speed daily basis
July-September 2016]
237
Table 2. Comparison of the results obtained from various ETo estimation methods
Month
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEPT
OCT
NOV
DEC
ETo(mm)
No. of days
I (10)
II(10)
III(11)
I(10)
II(10)
III(8)
I(10)
II(10)
III(11)
I(10)
II(10)
III(10)
I(10)
II(10)
III(11)
I(10)
II(10)
III(10)
I(10)
II(10)
III(11)
I(10)
II(10)
III(11)
I(10)
II(10)
III(10)
I(10)
II(10)
III(11)
I(10)
II(10)
III(10)
I(10)
II(10)
III(11)
Annual
avg.
Daily
average
Pan
1.04
1.26
1.43
1.62
1.90
2.22
2.56
3.03
3.86
4.59
5.13
5.59
5.42
5.64
5.48
5.80
5.35
4.23
3.73
3.21
3.19
3.08
2.68
2.93
3.02
2.73
2.61
2.49
2.25
2.02
1.92
1.84
1.54
1.37
1.24
1.16
1107.10
3.03
4.12
4.12
4.6
4.38
4.66
3.97
238
KAR et al.
Fig. 1. Architecture (2-2-1) of feed forward neural network with one hidden layer
Table 3. Statistical comparison of ETo estimated by different methods with ETo estimated by FAO-24 (Pan) model
ETo estimation
methods
PenmanMonteith
Hargreaves
Irmak
HargreavesSamani
Mod PM
Valiantzas
ANN
MSE
RMSE
0.984
Coefficient of
correlation (r)
0.977
Coefficient of
determination (r2)
0.955
C
(y intercept)
1.056
0.968
1.397
1.649
2.534
1.182
1.284
1.592
0.960
0.940
0.979
0.922
0.884
0.959
Y2 = 0.8051X + 1.683
Y3 = 0.5877 X + 2.334
Y4 = 0.973 X + 1.645
1.683
2.334
1.645
1.899
2.852
1.002
1.378
1.689
1.001
0.985
0.952
0.971
0.970
0.907
0.942
Y5 = 1.0309 X + 1.257
Y6 = 0.9663 X + 1.727
Y7 = 0.9457 X + 1.100
1.257
1.727
1.100
where, X = Estimated ETo by Pan method and Y1, Y2, Y3, Y4, Y5, Y6 and Y7 are estimated ETo by Penman- Monteith, Hargreaves,
Irmak, Hargreaves samani, Modified PM, Valiantzas and ANN methods, respectively.
July-September 2016]
239
240
KAR et al.
July-September 2016]
CONCLUSION
Estimation of the crop evapotranspiration is
very useful for crop planning in dry Sub-humid
Agro-ecological region. In case of limited weather
data, Artificial Neural Network (architecture: 2-21) with input parameters mean temperature and
solar radiation (R s ) estimate the reference
evapotranspiration very accurately. The Pan data
can be correlated with penman Monteith method
high degree of correlation.
REFERENCES
Allen, R.G., Pereira, L.S., Raes, D. and Smith, M. 1998.
Crop evapotranspiration- Guidelines for computing
crop water requirements. Journal of irrigation and
drainage engineering. Paper No. 56, FAO, Rome.
Doorenbos, J. and Pruitt, W. O. 1977. Guidelines for
computing crop water requirements. (FAO). Journal
of irrigation and drainage engineering. Paper No.
24 (rev). FAO, Rome.
Hargreaves, G. H. and Samani, Z. A. 1985. Reference
crop evapotranspiration from temperature. Applied
Engineering in Agriculture 1(2): 96-99.
Hargreaves, G. H. 1975. Moisture availability and crop
production. Trans. American Society of Agricultural
Engineering 18(5): 980-984.
Irmak, S., Payero, J. O., Martin, D. L., Irmak, A. and
Howell, T. A. 2006. Sensitivity analyses and
sensitivity coefficients of standardized daily ASCE
Penman-Monteith equation. Journal of Irrigation and
Drainage Engineering 132(6): 564-578.
Irmak, S., Allen, R. G. and Whitty, E. B. 2003. Daily grass
241