Advances in Remote Sensing, 2013, 2, 258-268
http://dx.doi.org/10.4236/ars.2013.23028 Published Online September 2013 (http://www.scirp.org/journal/ars)
Irrigation Scheduling Using Remote Sensing Data
Assimilation Approach
Baburao Kamble1, Ayse Irmak1, Kenneth Hubbard1, Prasanna Gowda2
1University of Nebraska-Lincoln, Lincoln, USA
2USDA-ARS, Bushland, USA
Email: bkamble3@unl.edu
Received February 28, 2013; revised March 28, 2013; accepted April 28, 2013
Copyright © 2013 Baburao Kamble et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ABSTRACT
Remote sensing and crop growth models have enhanced our ability to understand soil water balance in irrigated agri-
culture. However, limited efforts have been made to adopt data assimilation methodologies in these linked models that
use stochastic parameter estimation with genetic algorithm (GA) to improve irrigation scheduling. In this study, an in-
novative irrigation scheduling technique, based on soil moisture and crop water productivity, was evaluated with data
from Sirsa Irrigation Circle of Haryana State, India. This was done by integrating SEBAL (Surface Energy Balance
Algorithm for Land)-based evapotranspiration (ET) rates with the SWAP (Soil-Water-Atmosphere-Plant), a process-
based crop growth model, using a GA. Remotely sensed ET and ground measurements from an experiment field were
combined to estimate SWAP model parameters such as sowing and harvesting dates, irrigation scheduling, and
groundwater levels to estimate soil moisture. Modeling results showed that estimated sowing, harvesting, and irrigation
application dates were within ±10 days of observations and produced good estimates of ET and soil moisture fluxes.
The SWAP-GA model driven by the remotely sensed ET moderately improved surface soil moisture estimates sug-
gesting that it has the potential to serve as an operational tool for irrigation scheduling purposes.
Keywords: Artificial Neural Network; Genetic Algorithms; SEBAL; Remote Sensing; Groundwater; Crop Growth
Modeling
1. Introduction
Water scarcity is causing more pressure on utilization of
fresh water resources in irrigated agriculture [1]. There-
fore, a paradigm shift is necessary from a supply driven
into a more demand driven water management. It is rec-
ognized that appropriate irrigation scheduling should
lead to improvements in water management performance,
especially at a farm level [2]. Evapotranspiration (ET) is
one of the components of water use efficiency and re-
sulting crop productivity. Periodic information of ET
based on remote sensing would be very useful to reduce
uncertainty in the crop model parameters and subsequent
accurate estimation of the water balance. Several algo-
rithms have been developed to utilize remote sensing
data for quantifying ET [3-9]. Researchers have also re-
viewed different ET algorithms [10] and used remotely
sensed data in conjunction with crop or hydrological
models via data assimilation for improving soil moisture
estimation [11-14]. Researchers also used the Ensemble
Kalman Filter (EnKF) with daily microwave observa-
tions over an eight-day period, as well as through a crop-
ping season for estimating soil moisture fluxes [11].
Studies were showed the concept of optimal downscaling
for a case where soil moisture estimates were required at
scales smaller than that of the microwave observations
[14]. An extensive review was conducted on soil water
simulation model that uses remotely sensed data to pre-
dict moisture in soil profiles [12,15-17]. Ines and Honda
developed an assimilation methodology [12] for the Soil,
Water, Atmosphere, and Plant (SWAP) simulation model
[18] with remote sensing data using Genetic Algorithm
(GA) [19]. Similar work was done at spatial scale [20]
with an objective to fuse remote sensing data.
Given the above background, the emphasis of this
study was to develop a comprehensive data assimilation
approach for scheduling on-demand irrigation using
SWAP model predictions and SEBAL (Surface Energy
Balance Algorithm for Land) based ET [5]. This meth-
odology was implemented using a GA to estimate values
for SWAP’s sensitive input parameters and by updating
C
opyright © 2013 SciRes. ARS
B. KAMBLE ET AL. 259
SWAP-ET predictions with SEBAL-ET. This was achi-
eved by (a) estimating ET from MODIS (Moderate Re-
solution Imaging Spectroradiometer) data using SEBAL,
(b) developing an ET data assimilation scheme with a
GA to optimize SWAP input parameters for scheduling
irrigation, (c) validating optimized parameters with the
observations made at the experimental site, (d) evaluat-
ing the potential use of optimized parameters for irriga-
tion scheduling with two separate runs of SWAP model
with and without optimizer, and (e) simulating and com-
paring yield and water use efficiency under different ir-
rigation scenarios for an irrigated cotton field in the Sirsa
Irrigation Circle (SIC) located in Haryana, India (Figure
1). The irrigation circle is an administrative irrigation
unit managed by the Haryana Irrigation Department
[21,22].
2. Materials and Methods
2.1. Study Area
The proposed approach was tested using a dataset on
irrigated cotton field in the SIC (Figure 1). Data used in
this study was collected as part of another study con-
ducted by the Wageningen Agricultural University, The
Netherlands during 2002 for calibrating the SWAP
model. The hourly meteorological measurements (Fig-
Figure 1. The location of the study area (green rectangle)
within Sirsa Irrigation Circle (CIS), Haryana, India.
ure 2) included air temperature, wind speed, solar radia-
tion, and precipitation from a weather station installed at
the Indian Council of Agricultural Research-Cotton Re-
search Institute (ICAR-CRS) (latitude 29˚35' North; lon-
gitude 75˚08' East) located within the SIC (Figure 1).
The SIC is located in the extreme western part of Hary-
ana between latitudes 29.1˚ and 30.0˚ North and longi-
tudes 74.2˚ and 75.3˚ East. The climate of the SIC area is
characterized by semi-arid and short, mild, variably wet
monsoons. The climate of this SIC is characterized by its
dryness and extremes temperatures and scanty rainfall.
Based on long-term records, January is coldest month
with daily minimum temperature 5˚C and May/June is
hottest month with temperature rises up-to 45˚C. The
average annual precipitation in Sirsa Irrigation Circle
varies from 100 to 400 mm, which is only 10% - 25% of
the potential evapotranspiration of common crop rota-
tions. The precipitation mainly occurs during monsoon
months of July to September [21]. Ground surface eleva-
tions vary from 192 to 207 m above mean sea level.
Rice-wheat cropping system is the major cropping sys-
tem in the SIC. Use of comparative short-duration (100 -
120 days after transplanting) of rice and wheat (135 to
150 days) wheat varieties has offered a unique opportu-
nity for extension of area under a two -crops-a-year. [17,
21-23].
The total area of the SIC is 44,200 km2 with about
82% of the area under cultivation. At present, only 40%
of the total cultivated area is under surface (canal) water
irrigation. Water management in the SIC, like any other
arid or semi-arid regions in the developing world, is very
complex in nature. Key characteristics of the SIC are: a)
scarce and erratic precipitation with no perennial rivers
in and around the area, b) high evaporative demand, c)
marginal to poor quality groundwater in most parts, d)
rising groundwater levels with occasional flooding, and e)
low water-holding capacity of soils.
Other factors affecting water use efficiency and crop
production include fluctuations in canal water supply,
Figure 2. Measured daily values of minimum and maximum temperature, humidity and precipitation, in Sirsa district during
he agricultural year 2002. t
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL.
260
low irrigation application efficiency due to light textured
soils, and conveyance losses from the irrigation system
[21,23]. A summary of the physical properties of soil at
the time of sowing of cotton is presented in the Table 1.
The soil at the experimental site was a typical sandy
loam with 8% - 17% clay, 10% - 16% silt and 60% -
80% sand with 0.5% organic matter, electrical conduc-
tivity rate of 0.15 dS/m, and pH of 8.5 with bulk density
1.65 gm/cc with low water-holding characteristics. Table
1 shows soil physical properties measured in the experi-
mental field. Soil water content measurements were
Table 1. Physical properties of soil of field at the time of
sowing of cotton.
Depth (cm)
Parameter Unit
0 - 1515 - 30 30 - 6060 - 90
Textural class Name Sandy
loam
Loamy
sand
Loamy
sand
Sandy
loam
Clay (%) 11.159.82 8.94 10.31
Silt (%) 11.7710.74 10.0713.04
Sand (%) 77.0879.44 80.9976.29
Soil moisture at
saturation (%) 31.3 31.2 31.5 35.7
Saturated hydraulic
conductivity cm/hr 4.9324.012 5.0375.011
Bulk density g/cc 1.65 1.69 1.65 1.63
made in the top two soil layers (0 - 15 cm and 15 - 30 cm)
using a gravimetric sampling method for seven times
during 2002.
There were about 15 - 20 soil samples takenfor gravi-
metric analysis on each measurement day and soil mois-
ture was estimated based on volumetric basis (% mois-
ture on dry weight basis × bulk density). The study field
was irrigated from a tube-well with a discharge of 64.76
m3/hr.
2.2. Satellite Data Processing
The MODIS Level 1B (L1B) images (radiometrically
corrected) of the Indo-gangatic area covering the SIC
were downloaded from the Earth Observing System Data
Gateway of NASA (National Aeronautics and Space
Administration). Although MODIS images were avail-
able for every 1 - 2 days, there were only 13 cloud-free
MODIS images from eight day composites available to
estimate seasonal ET for the Kharif (summer) growing
season (5/2002-10/2002). Table 2 shows the details of
MODIS products used in the study. A subset image for
the study area was extracted for better visualization and
computationally efficient analysis of satellite data. The
MOD11 L2 data comprised of two thermal bands with a
1 km resolution and was used to estimate surface tem-
perature and emissivity. Extraction of the binary files
was performed for two visible (bands 1 and 2), five
short-wave infrared (bands 3-7) and two thermal (31 and
Table 2. MODIS data products used in the analysis.
Data set Data Type Fill Value Valid range Scale Factor
MOD09: MODIS Terra Surface Reflectance (500 m)
Surface Reflectance Band 1 (620 - 670 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 2 (841 - 876 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 3 (459 - 479 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 4 (545 - 565 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 5 (1230 - 1250 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 6 (1628 - 1652 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Surface Reflectance Band 7 (2105 - 2155 nm) 16-bit signed integer 28672 100 - 18,000 0.0001
Solar Zenith Angle 16-bit signed integer 0 0 - 18000 0.01
Granule Time 16-bit signed integer 0 0 - 2355 1
MOD11: MODIS Land Surface Temperature and Emissivity (1000 m)
Land Surface Temperature 16-bit signed integer 0 7500 – 65,535 0.02
Band 31 emissivity 8-bit unsigned integer 0 1 - 255 0.002
Band 32 emissivity 8-bit unsigned integer 0 1 - 255 0.002
Local solar time of Land-surface Temperature observation 8-bit unsigned integer 0 0 - 240 0.1
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL. 261
32) bands. The original MODIS data was provided in
HDF (Hierarchical Data Format). A HEG conversion
tool (http://gcmd.nasa.gov/ records/HEG.html) was used
to convert the HDF files into Geo TIFF images. Individ-
ual images of each band were created for each day by
converting their corresponding HDF files. Our experi-
ment field size was about 4386 m2, which was less than
one pixel (1 × 1 km) in MODIS thermal image. MODIS
LST (Land Surface Temperature) product was down-
scaled to 250 m using a cubic convolution technique to
be consistent with the spatial resolution of MODIS visi-
ble and near infrared (MOD09) data. For geo-rectifica-
tion, we have changed the projection from sinusoidal to
UTM with a WGS84 datum. This ended up with gridded
data for which both the geographic coordinate system
and the projected coordinate system are defined in terms
of the WGS84 ellipsoid.
2.3. Evapotranspiration Mapping with SEBAL
SEBAL is a remote sensing based algorithm that com-
putes a complete surface energy balance along with re-
sistances for momentum, heat and water vapor transport
for each pixel [5]. Land surface parameters such as sur-
face albedo, vegetation index, emissivity, and surface
temperature were derived from MODIS data using the
SEBAL. The key input data for SEBAL consists of spec-
tral radiance in the visible, near-infrared and thermal
infrared part of the electromagnetic spectrum. In addition
to MODIS data, the SEBAL requires routine weather
data parameters (wind speed, humidity, solar radiation
and air temperature). Under the absence of advection, the
energy balance in SEBAL is calculated at an instant time
t for each satellite overpass by the following equation:
n
REHG
 (1)
where Rn is the net radiation (W/m2), G is the soil heat
flux (W/m2), H is the sensible heat flux (W/m2), and λE
is the latent heat flux which is the energy necessary to
vaporize water (W/m2). The λE under given atmospheric
conditions can be calculated as a residual of the energy
balance components in Equation (1). The instantaneous
evaporative fraction (, dimensionless) is an expression
to obtain the actual ET when the atmospheric moisture
conditions are in equilibrium with the soil moisture con-
ditions. The is used to calculate the daily value based
on the assumption that the evaporative fraction is con-
stant during daytime hours under non-advective condi-
tions [5]:
n
E
EH R

 E
G
(2)
where daily actual evapotranspiration (ET24) is calculated
from the , and the Rn integrated over the 24-h period
(Rn24). According to assumptions made in the SEBAL
model, net available energy (RnG) reduces to Rn at
daily timescales. ET24 is computed as:
3
24 24
86400 10
n
w
X
ET R

(3)
where Rn24 is the 24-h averaged net radiation (W/m2), is
the latent heat of vaporization (J/kg), w is the density of
water (kg/m3) and ET24 is daily actual ET (mm/day).
2.4. Soil-Water-Atmosphere–Plant (SWAP)
Model
An intermediate version of the SWAP model (SWAP-
GA) [12] was used in this study. The SWAP is a physic-
cally based one-dimensional model that simulates verti-
cal transport of water flow, solute transport, heat flow
and crop growth at the field scale level [16]. It requires
inputs including management practices and environ-
mental conditions to compute a daily soil water balance
and crop growth. The major processes taken into account
are phenological development, assimilation, respiration
and ET. The SWAP model uses Richard’s equation [24]
to simulate vertical soil water movement in variably
saturated soils as follows:

1K
tz z
 



(4)
where K is the hydraulic conductivity (cm·d1), ψ is the
pressure head (cm), z is the elevation above a vertical
datum (cm), θ is the water content (cm3·cm3), and t is
time (d). The soil hydraulic functions in the model are
defined by the Mualem-Van Genuchten (MVG) equa-
tions [25] which describe the capacity of the soil to store,
release and transmit water under different environmental
and boundary conditions. Darcy’s law is used to deter-
mine potential soil evaporation in wet soil conditions.
Root water extraction at various depths in the root zone
is calculated from potential transpiration, root length
density and possible reductions due to wet, dry, or saline
conditions. The SWAP also integrates the basic WO
FOST (World Foods Tudies) crop growth model and was
frequently used to study the effect of the climate change
on crop production [12,17,20-22]. Water requirements of
a crop depend mainly on crop growth stage and envi-
ronmental conditions. Root water uptake estimated by
model does not depend on the rooting density but only on
the actual rooting depth and available soil water. Dif-
ferent crops have different water-use requirements under
the same weather conditions. SWAP model simulation
gives the balance of water inputs from precipitation and
from addition of water to root zone by root growth and
water losses computed by crop transpiration, soil evapo-
ration, and percolation to deep soil layers, which gives a
complete picture of the water availability and water con-
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL.
262
sumption in particular cropping system [18].
2.5. Optimization Scheme
The SWAP-GA model relies heavily on assimilation of
land surface data, which has shown significant potential
to improve the realistic representation of the land surface
condition. The objective of data assimilation is to obtain
the best estimate of the state of the system by combining
observations with the forecast model’s first guess. Ge-
netic algorithms (GA) technique is a function of optimi-
zation derived from the principles of evolutionary theory.
It is designed to search, discover, and emphasize opti-
mum solutions by applying selection and crossover tech-
niques, inspired by nature, to supply solutions [19,26].
GA operates on pieces of information as nature does on
genes in the course of evolution. It has good global
search characteristics. Three operators are designed to
modify individuals: selection, mutation and crossover
[27]. The evolution usually starts from a population of
randomly generated individuals and happens in genera-
tions. In each generation, the fitness of every individual
in the population is evaluated; multiple individuals are
stochastically selected from the current population based
on their fitness, and recombined and possibly randomly
mutated to form a new population. The new population is
then used in the next iteration of the algorithm. The
strength of GA with respect to other local search algo-
rithms (lookup table method, ant colony etc,) is to derive
more strategies which can be adopted together to find
individuals to add to the mating pool, both in the initial
population phase and in the dynamic generation phase.
Thus, a more variable search space can be explored at
each step. Based on the above biological evolution idea,
a so-called “SWAP-GA” has been developed by resear-
chers [12] to estimate input parameters of SWAP from
remote sensing data.
Based on the above biological evolution idea, a so-
called “SWAP-GA” [12] to estimate input parameters of
SWAP from remote-sensing data. The model was
adopted and recoded according to the objectives of this
research. Cotton is grown in Kharif (April-October) sea-
son in the Haryana State of India. Time of sowing spread
over a period of April to first fortnight of June. The op-
timized parameters were planting date, crop growth
period, starting date of irrigation scheduling, and the
groundwater depth at the start and end of the simulation
(Table 3). The proposed parameters were fed to SWAP
by GA according to the objective function. The GA sear-
ches for an optimum crop parameter set, while SWAP
tests the proposed parameters simultaneously by using
them in forward simulations. We compared the results
from GA for different populations and different genera-
tions. Best results were obtained by applying the algo-
rithm that was configured for 100 populations and 100
generations with up to five optimized crop growth para-
meters (emergence day, time extent of crop, start of irri-
gation scheduling, groundwater at start of season, ground-
water at end of season). We also optimized two parame-
ters that represent depths to ground water at start of sea-
son, groundwater at end of season. There was no reliable
field information available to check the validity of these
parameters.
Optimizing groundwater at start of season allowed us
to initialize water table at the beginning of the simulation.
In general, the introduction of a priori information im-
proves the convergence and accuracy of the derived pa-
rameters, even in cases where the a priori information is
slightly erroneous.
Consider C as the cost function having (x, y, d) pa-
rameters. The x and y define coordinates of a pixel loca-
tion, with x being the longitude [0-180/E-W], y being the
latitude [0-90/N-S] and d is the satellite overpass date
[i,...,j].

2
SEBAL SWAP
xyd
ET ET
Cn
(5)
where ETSEBAL is estimated ET via the SEBAL model
using remotely sensed data (cm), as the “observed” data
for the experimental field in the SIC. ETSWAP is estimated
actual ET from SWAP-GA and based on optimized
model parameters, n is the time domain as number of
satellite images (sum of i to j = 13) and Cxyd is the object-
tive function (root mean square error: RMSE) for the
pixel at x, y location (cm) and i - j are satellite image dates.
When a minimum-difference defined threshold was
reached, SWAP parameters were stored for reconstruct-
Table 3. Definition, unit, minimum, and maximum values of optimized parameters in SWAP-GA.
Optimized parameters Definition Unit Minimum value Maximum value
DEC Emergence day Ordinal day 140 160
TC Time extent of crop Ordinal day 100 200
STS Start of irrigation scheduling Ordinal day 140 160
GWjan Groundwater at start of season cm 140 160
GWdec Groundwater at end of season cm 140 160
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL. 263
tion of ET for any required day in the cropping season.
We tested the procedure assuming that some degree of
error in remote sensing observations (ETSEBAL). The fit-
ness of an individual having x, y pixel location charac-
teristics is the inverse of the cost function times the con-
straints aimed at minimizing the RMSE between SWAP
ET and target SEBAL ET. Each water balance parameter
after optimization is estimated for kharif growing period
to calculate water use efficiency based on Yield/Transpi-
ration, Yield/Evapotranspiration, and Yield/Irrigation. This
will evaluate the phenomena with respect to the yield of
new irrigation scheme.
We used regression analysis, and root mean square er-
ror to evaluate the simulation results. Regression analysis
gives information on the relationship between the ob-
served ET variable and the predicted ET variable to the
extent that information is contained in the data. To eva-
luate the performance of the soil moisture simulation, co-
efficient of determination was used as a relative index of
model performance, and root mean square error (RMSE)
was used to compare the observed soil moisture and pre-
dicted soil moisture. This gave an indication of both bias
and variance from the 1:1 line. The RMSE provides a
good measure of how closely two independent data sets
match.
3. Results and Discussion
3.1. Estimation of Evapotranspiration with
SEBAL Model
During 2002 Kharif season, the actual ET via SEBAL
model has been quantified for a cotton field in the SIC.
The selection of study area is from homogenous cropping
practice region which is suitable for applying low spatial
resolution remote sensing [16,17,20]. Therefore, the sig-
nal in the specific pixel of study area represents the ac-
tual electromagnetic characteristics of the cotton. Figure
3 shows the temporal distribution of normalized differ-
ence vegetation index (NDVI) and remote sensing esti-
mated ET over the experimental field for Kharif 2002
growing season. Multidate satellite data provide the in-
formation of the different stages of the crop. Figure 3
shows the NDVI varied from 0.1 around 1 May to 0.3 at
flowering stage (early June) where photosynthetic capac-
ity of a cotton leaf depends on its age. Leaf area index
(LAI) and NDVI curves demonstrated gradual increase
or decrease in their values with changes in precipitation
in early and mid-season. The irrigation demand of cotton
increases with increase in NDVI or with increasing pho-
tosynthetic rate and vice-versa. NDVI gives important
information on the amount of area exposed to the at-
mosphere for photosynthesis. Soil water availability has
direct relation with stomatal behavior and is ultimately
related to the photosynthetic process of that crop [14].
Changing progression of ET over cotton crop followed
the trend in NDVI during the growing season except for
satellite overpasses in mid-September, 2002 (Figure 3)
which shows the low NDVI and high ET values. This
might be because of occurrence of precipitation, clima-
tological conditions or changes in land use within the
MODIS pixel covering the study field. The ET for the
study field was low early in the season and varied from
0.01 to 0.2 cm/day for months when the soil was bare
and open. During crop development and mid-season
stages, ET varied from 0.3 cm/day to 0.46 cm/day. This
is due to available soil water via irrigation and precipita-
tion events occurred during that period. After mid-season,
ET varied between 0.2 mm/day and 0.4 mm/day, ET de-
clined to 0.3 mm/day at the time of harvesting. However,
ET continued to increase in the experimental field even
after the NDVI reached its maxima. This may be due to
the saturation of NDVI after reaching a certain leaf area
index (LAI). Figure 3 shows NDVI is declining from 0.5
in mid-September to 0.1 in mid-October and during this
period, LAI increased from about 4 to 6. The differences
in NDVI may partly be due to change in landuse or dif-
Figure 3. Temporal distribution of LAI, NDVI and evapotranspiration (ET) during the Cotton growing season in the study
area.
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL.
264
ferences in timing and amount of irrigation in the sur-
rounding fields that fall within the study pixel.
3.2. Remotely Sensed Evapotranspiration Data
Assimilation in Hydrological Model
Figure 4 compares temporal distribution of SEBAL
based ET and SWAP-GA based ET, two curves shows
similar trends of under and over estimations of actual ET.
The SWAP-GA marginally overestimated than the SE-
BAL based ET in early season when the soil surface was
dry and underestimated late in the season when the soil
surface was wet and covered by the crop, which influ-
ences efficiency of water use, high water productivity
and efficient farming activities. The larger ET differ-
ences between SWAP-GA and SEBAL were found dur-
ing May 2002 and June 2002 with a mean absolute dif-
ference of 4 mm/day. However, mean absolute difference
was increased to 5 mm/day when simulations were made
without data assimilation. This difference has huge im-
pact on estimating irrigation demand and scheduling.
During early in the growing season, the bias far exceeded
the actual values. The main reasons for this bias are over
estimation of SEBAL-ET and model considers no tran-
spiration till plant emergence.
The SWAP-GA system tries to minimize the differ-
ence between SWAP model and SEBAL-ET and the dif-
ference between the SWAP-GA and SEBAL ET mini-
mized to 0.25 mm. On August 13, 2002 (Figure 4), the
difference between SEBAL-ET and SWAP-GA-ET was
about 8 mm and it was because the SWAP-GA model
usually overestimates ET right after irrigation application
or a precipitation event (any citation). The data assimila-
tion results in Figure 4 are promising but further refine-
ment is necessary to improve the propagation of the cor-
rection to the domain outside the assimilation points
caused by mixed pixels and to get better bias estimates.
The bias due to the comparison of pixel observations
with the model needs to be explicitly taken into account
to prevent unnecessary forcing of the model towards bi-
ased observations. In our case, there is a bias due to the
comparison of SEBAL pixel observations with the
SWAP-GA model.
3.3. Optimization of Crop Growth Parameter
Using SWAP-GA Model
SWAT-GA model parameters were optimized by mini-
mizing the RMSE between SWAP-GA-ET and the target
SEBAL-ET values and resulting parameter values were
used as input for simulating irrigation scheduling. Gen-
erally, remote sensing based ET values contains errors
due to errors associated atmospheric correction of the re-
flectance data and due to errors associated with ET algo-
rithms.
Furthermore, coarser resolution, multispectral images
such as MODIS have mixed pixel problems which makes
it more complicated if the selected pixel exhibits some
heterogeneity on the high spatial resolution satellite im-
age. Table 4 shows the values of optimized parameters
as well as data from the experimental field. Optimized
parameter values with SWAP-GA were closely matched
with field measurements. The simulated cropping period
from planting to harvest was 169 days against the actual
period of 179 days. The depth of groundwater (water
table) about 141 cm to 151 cm is not uncommon in
irrigated cotton cropping areas in the SIC [21,22], espe-
scially considering an inundated condition at the start of
the period of study.
3.4. Soil Moisture Based Irrigation Scheduling
Scheme
Figure 5 shows time series observed and simulated soil
water contents (cm3/cm3) at 0 - 15 cm and 15 - 30 cm soil
depths and at 30 - 60 cm and 60 - 90 cm soil depths in
Figure 6. About 50 - 60 percent of the total water uptake
Figure 4. Actual evapotranspiration (cm/day) for the 2002 cotton growing seasons. Observed ET is based on SEBAL algo-
rithms (SEBAL ET) on satellite overpass dates. ET predictions are with original SWAP and SWAP-GA models.
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL. 265
Figure 5. Simulated and observed soil water content (cm3/cm3) at 0 - 15 cm and 15 - 30 cm soil depths by SWAP-GA with
optimized parameters, rainfall and on-demand irrigation amounts are also shown.
Table 4. The simulated and observed optimized parameters based on 10 generations & 10 populations.
Emergence End of Crop Start of irrigation scheduling
Simulated Parameter 04-June* 23-October 02-June
Observed Parameter 20-May (Actual sowing date) 15-November 18-May
*Consider germination period 14 days i.e. Emergence = sowing date + germination period.
by the crop occurred within the top 90 cm depth, where
more than 90 percent of the total root mass found. Fig-
ures 5 and 6 shows the soil water depletion till 60 per-
cent as the season progresses. SWAP-GA predicted a
total of eleven irrigation applications with irrigation de-
mands varying from 7.8 cm to 9.9 cm per application
during wheat growth and development period. The major
portion of the irrigation demand usually occurs in May,
June, August and September months to avoid water stress
during critical crop development stages i.e. flowering and
fruiting. For cotton, the irrigation demand vary from 7.8
cm to 9.9 cm with respect to the irrigation timing, the
growth stage of the crop, climate and length of the total
growing period. Early development stages show more
difference in the actual and potential ET, while mid and
late season shows very less difference because of the
increased irrigation demand. Cotton crops received regu-
lar precipitation during the growing season, and most of
it occurred during mid-season. Further, presence of ex-
cess water in the root zone early in the growing is ex-
pected to restrict root and crop development. Figures 5
shows that, two consecutive irrigations (8.7 cm and 9.8
cm) during the cotton emergence in early June addition
to two precipitation events(1.8 cm and 6.9 cm).The pre-
cipitation contribution (17.67 cm) to crop ET mainly
during kharif (cotton) which is very low as compared to
seasonal irrigation supplies (102.68 cm) to the fields.
Irrigation frequencies are high in mid-season during the
flowering stage when the leaf area is at its maximum
level. Time series of moisture data indicated that the soil
water content at top and bottom layers were quite similar
from germination until the date of first precipitation. The
top soil layers have slightly lower water contents than
lower layers. The model predictions closely matched ob-
servations indicating model’s ability in simulating soil
water content. Root Mean Square Error (RMSE) for
simulated and observed soil water content (cm3/cm3) at 0
- 15 cm is 0.08, 15 - 30 cm is 0.01, 30 - 60 cm is 0.001
and 60 - 90 cm is 0.01. The RMSE of simulated and ob-
served soil moisture for four depths provides the mini-
mum possible error. Overall, our results showed that the
rainfall contribution to crop ET was very minimal as
compared to irrigation supplies to the fields. Although
the crest of the soil moisture curve and rainfall matched
at some locations, soil moisture tended to rise even
though there was no rainfall event.
Figures 5 and 6 shows the large ratio of evaporation to
precipitation in July/August and has insignificant impact
on the soil moisture due to the relatively small precipita-
tion events and less irrigation combined with low tem-
peratures and the soil moisture can be maintained to a
constant level. The top soil layers have slightly lower
water contents than lower layers. It is because the top
layer forms the sphere of life, which receives moisture in
pulses of precipitation and irrigation. From Figures 5
and 6, it also reveals that the top 30 cm of the soil ex-
perienced greater soil moisture fluctuations than in soil
layer below. It is because the top layer forms the sphere
of life that receives moisture in pulses of precipitation
and irrigation while a major portion of that water is ex-
tracted through evaporation and transpiration by plants.
he simulated and observed soil moisture levels showed T
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL.
266
Figure 6. Simulated and observed soil water content (cm3/cm3) at 30 - 60 cm and 60 - 90 cm soil depths by SWAP-GA with
ers for
Tab WAP-GA pre-
and observed water balance parameters
optimized parameters, rainfall and on- demand irrigation amounts are also shown.
creasing trend from June to September, and then de- Table 5. Estimated
in
creased onwards, which coincided with occurrence of
irrigation and precipitation. Lowest level of moisture in
the soil profile was simulated for August when plants are
transpiring at a maximum rate (Figure 3). However, this
phenomenon did not occur in the observed data. In SIC,
it is very common to have very dry conditions late in the
growing season and it can reduce crop yields if soil water
content is is not available. As the water table depth in-
creases, the soil layer tends to hold more water with no
fluctuations throughout the season. The Sirsa Irrigation
Circle has two water tables, the first one at a depth of 5
m from the surface and the second at a depth of 15 m.
The effect of capillary rise is expected because of the soil
type and deep percolation is the main phenomenon for
the water flow into various soil layers. Trends in soil
moisture predictions follow that in precipitation although
the crest of soil moisture curve and precipitation match at
some locations. Cotton plant consumes more water and it
has high sensitivity to moisture increase. As the crop
matures, soil moisture depletion allowances can be
greater. The Sirsa soil have storage reserves of 25 to 100
cm of water which mainly depends on rooting depth of
crops grown in this area which makes use of more soil
moisture to minimize risk of leaching.
3.5. Evaluation of Optimized Paramet
Yield Estimation under on-Demand
Irrigation Scheduling Scheme
le 5 presents a comparison between S
dicted and observed crop yield. A general progressive
yield was observed with respect to the simulation criteria
(with or without On-Demand irrigation). Current cotton
yields in the Haryana state is approximately 3500 kg/ha
under irrigated conditions [21] while SWAP-GA simu-
lated with optimized parameters for on-demand irrigation
case showed cotton yield 3686 kg/ha and SWAP simula-
tions with observed irrigation and yield dataset at
(cm) for two irrigation cases.
Parameters of interest Case 11 Case 22
Transpiration (cm) 44.2 79.5
E
1 1
H
Total pr+Lint)
Ge
vapotranspiration (cm)52.2 89.2
Irrigation (cm) 34.1 102.7
Crop dry mass (kg/ha) 7,170 8,701
arvesting Index 21.47 21.47
oduction of cotton (Seed3687 4015.1
inning percentag37.1 37.11
Lint weight (kg/ha) 1368 1490
1SWAP-Gized paramr on-deiga-
tion. 2SWserved irrigand yieldt at
farmer’s fie
lied by on-demand irrigation increased lint
reater irrigation ca-
A simulations with optim
AP simulations with ob
eters fo
tion a
mand irr
datase
ld.
farmer’s field showed 4015 kg/ha.
The SWAP-GA simulations show that crop water
eeds suppn
yield response to the progressively g
pacity treatments. Table 6 shows estimated and observed
water balance parameters (cm) for two irrigation cases.
The models indicate that the water use efficiency of cot-
ton increased from 0.15 to 0.4 kg/m3 based on irrigation
while 0.17 to 0.26 kg/m3 by ET (Table 6) which indicat-
ing considerable variation and scope exists for improve-
ments in WUE based on calibrated parameters. In Hary-
ana, successful crop production is not possible without
supplemental irrigation because of erratic precipitation
events. Irrigation application by the calibrated model and
on-demand irrigation, the water use efficiency obtained
from the on-demand is increased considerably without
water deficit. Factors responsible for the low WUE-val-
ues include both the relatively high fractions of soil
evaporation in the ET term and of water percolation from
the irrigation water applied.
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL. 267
Table 6. Water use efficiency (WUE, kg·m3) for two irriga-
tion case studies.
Water Use Efficiency3 Case 11 Case 22
WUET (kg·m3) 0.31 0.19
WUEET (kg·m3) 0.26 0.17
WUEIR (kg·m3) 0.40 0.15
1SWAP-GAptimized parafor on-de
tion. 2SWA observed irrigad yield dat far-
mer’s field. nd WUEIR are we efficiencid on
transpiratio and irrigation,ctively (kg
his
search to schedule irrigation based on the on-dem
AP crop growth model with a genetic
ptimizer. We used remote sensing ET
and parameters could be predictable reasonably
w
va,
and R. Barker, “World Water Demand and Supply, 1990
to 2025: Scenal Water
Management bo, Research Re-
. Tasumi and R. Trezza. “Satellite-Based
simulations with ometers mand irriga-
port No. 19.
[2] B. Kamble and A. Irmak “Combining Remote Sensing
Measurements and Model Estimates through Data As-
similation,” IEEE International, Vol. 3, 2008, p. 1036.
[3] R. G. Allen, M
P simulations with
3WUET, WUEET, a
tion an
ater us
taset a
es base
n, evapotranspiration, respe·m3).
4. Summary
conceptual modeling methodology was tested in tA
re
Ene
and
2007, pp. 380-394.
strategy using SW
algorithm as an o
data to characterize our model via a stochastic data as-
similation approach, and then the optimized crop growth
parameters were used as inputs to agro-hydrological
model. The strength of an integrated data assimilation
approach was shown explicitly in the scenario analysis. It
was shown that there is a strong relationship between
irrigation scheduling, ET, soil water availability, and
groundwater table. The effects of ET on the water bal-
ance have been demonstrated. It does show that there is a
lot of scope for reducing errors in estimated ET to im-
prove water balance estimates. Parameter estimations are
successful and the ability of the SWAP-GA to produce
ET and soil moisture values accurately in relations with
precipitation and irrigation were promising, although the
general performance of the model can be described as
reasonable. In summary, this study has explored the po-
tential of genetic algorithm to estimate the crop parame-
ters for improving characterization of water management
options for predicting soil water content to schedule irri-
gation.
In this study, the numerical case of 100 generations
and 100 populations showed that the GA was able to
characterize the terms included in the fitness function
very well
ell. It has also demonstrated the potential of the data
assimilation approach as used in this study is a powerful
tool in crop and water parameter estimation for irrigation
scheduling. Our results also indicate that the soil mois-
ture profile estimates obtained from this particular syn-
thetic experiment are as good as realistic data. In practice,
the feasibility of retrieving subsurface moisture profiles
from surface measurements depends on the accuracy and
the physical realism of the land surface model and the
associated error statistics. Since the subsurface states
cannot be remotely sensed at the pixel scale, they can
only be estimated by using the hydrologic model to
propagate information downward from the surface.
REFERENCES
[1] D. W. Seckler, U. Amarasinghe, D. Molden, R. De Sil
arios and Issues,” Internation
Institute (IWMI), Colom
rgy Balance for Mapping Evapotranspiration with In-
ternalized Calibration (METRIC)-Model,” Journal of Ir-
rigation and Drainage Engineering, Vol. 133, No. 4,
doi:10.1061/(ASCE)0733-9437(2007)133:4(380)
[4] M. Anderson, J. Norman, G. Diak, W. Kustas and J. R.
Mecikalski. “A Two-Source Time Integrated Model for
Estimating Surface Fluxes Using Thermal Infrared Re-
mote Sensing,” Remote sensing of Environment, Vol. 60,
1997, pp. 195-216. doi:10.1016/S0034-4257(96)00215-5
[5] W. G. M. Bastiaanssen, M. Menenti, R. A. Feddes and A.
A. M. Holtslag. “A Remote Sensing Surface Energy Bal-
ance Algorithm for Land (SEBAL): 1. Formulation,”
Journal of Hydrology, Vol. 212-213, 1998, pp. 198-213.
doi:10.1016/S0022-1694(98)00253-4
[6] W. P. Kustas and J. M. Norman, “Use of Remote Sensing
for Evapotranspiration Monitoring over Land Surfaces,”
Hydrological Sciences Journal, Vol. 41, No. 4, 1996, pp.
495-516. doi:10.1080/02626669609491522
[7] S. P. Loheide and S. M. Gorelick, “A High-Resolution
Evapotranspiration Mapping Algorithm (ETMA) with
Hydroecological Applications at Riparian Restoration
Sites,” Remote Sensing of Environment, Vol. 98, 1998, pp.
182-200.
[8] Z. Su, “The Surface Energy Balance System (SEBS) for
Estimation of Turbulent Heat Fluxes,” Hydrology and
Earth System Sciences, Vol. 6, No. 1, 2002, pp. 85-89.
doi:10.5194/hess-6-85-2002
[9] B. Seguin, E. Assad, J. P. Freaud, J. P. Imbernon, Y. Kerr,
and J. P. Lagouarde, “Use of Meteorological Satellite for
Precipitation and Evaporation Monitoring,” International
Journal of Remote Sensing, Vol. 10, 1989, pp. 1001-1017.
[10] P. H. Gowda, J. L. Chavez, P. D. Colaizzi, S. R. Evett, T.
A. Howell and J. A. Tolk, “ET Mapping for Agricultural
Water Management: Present Status and Challenges,” Ir-
rigation Science, Vol. 26, No. 3, 2008, pp. 223-237.
doi:10.1007/s00271-007-0088-6
[11] D. Entekhabi, H. Nakamura and E. G. Njoku, “Solving
the Inverse-Problem for Soil Moisture and Temperature
Profiles by Sequential Assimilation of Multifrequency
Remotely Sensed Observations,” IEEE Transactions on
Geoscience and Remote Sensing, Vol. 32, No. 2, 1994, pp.
438-448. doi:10.1109/36.295058
[12] A. V. M. Ines, “Improved Crop Production Integration
GIS and Genetic Algorithms,” Doctoral Thesis, Asian In-
Copyright © 2013 SciRes. ARS
B. KAMBLE ET AL.
Copyright © 2013 SciRes. ARS
268
Data Assimilation with
. B. McLau
Water Resources Research
stitute of Technology, Bangkok, 2002.
[13] B. Kamble, “Evapotranspiration
Genetic Algorithms and SWAP Model for On-Demand
Irrigation”, AIT Thesis, 2007.
[14] R. H. Reichle, D. Entekhabi and Dghlin,
,
“Downscaling of Radiobrightness Measurements for Soil
Moisture Estimation: A Four-Dimensional Variational
Data Assimilation Approach,”
Vol. 37, 2001, pp. 2353-2364.
doi:10.1029/2001WR000475
[15] M. R. Smith and R. W. Newton, “The Prediction of Root
Zone Soil Moisture with a Water Balance—Microwave
Emission Model,” Master’s The
sity, Texas, 1983.
sis, Texas A&M Unive
. doi:10.1007/s00271-009-0193-9
r-
[16] A. Irmak and B. Kamble, “Evapotranspiration Data As-
similation with Genetic Algorithms and SWAP Model for
On-Demand Irrigation,” Irrigation Science, Vol. 28, No.
1, 2009, pp. 101-112
[17] B. Kamble and A. Irmak, “Remotely Sensed Evapotran-
spiration Data Assimilation for Crop Growth Modeling,”
In: L. Labedzki, Ed., Evapotranspiration, InTech, 2011.
doi:10.5772/13990
[18] J. C. Van Dam, J. Huygen, J. G. Wesseling, R. A. Feddes,
P. Kabat, P. E. V. Van Waslum, P. Groenendjik and C. A.
Van Diepen, “Theory of SWAP Version 2.0: Simulation
of Water Flow and Plant Growth in the Soil-Water-At-
mosphere-Plant Environment,” Wageningen Agricultural
University and DLO Win and Staring Centre, Wagenin-
gen, Technical Document 45, 1997.
[19] D. E. Goldberg, “Genetic Algorithms in Search, Optimi-
zation, and Machine Learning,” Machine Learning, Vol.
3, No. 2-3, 1988, pp. 95-99.
doi:10.1023/A:1022602019183
[20] Y. Chemin, K. Honda, “Spatiotemporal Fusion of Rice
Actual Evapotranspiration with Genetic Algorithms and
an Agrohydrological Model,” IEEE Transactions on
Geosciences and Remote Sensing, Vol. 44, No. 11, 2006,
pp. 3462-3469. doi:10.1109/TGRS.2006.879111
[21] R. S. Malik, R. Kumar, D. S. Dabas, A. S. Dhindwal, S.
Singh, U. Singh, D. Singh, J. Mal, R. Singh and J. J. E.
rigated Crops in Sirsa District, In-
Bessembinder, “Measurement program and description
database,” In: J. C. Van Dam and R. S. Malik, Eds., Wa-
ter Productivity of Irrigated Crops in Sirsa District, India.
Integration of Remote Sensing, Crop and Soil Models and
Geographical Information Systems, WATPRO Final Re-
port, 2003, pp. 29-39.
[22] R. Singh, J. C. van Dam and R. A. Feddes, “Water Pro-
ductivity Analysis of Ir
dia,” Agricultural Water Management, Vol. 82, No. 3,
2006, pp. 253-278. doi:10.1016/j.agwat.2005.07.027
[23] K. B. Singh, P. R. Gajri and V. K. Arora, “Modelling the
Effect of Soil and Water Management Practices on the
Water Balance and Performance of Rice,” Agricultural
Water Management, Vol. 49, No. 2, 2001, pp. 77-95.
doi:10.1016/S0378-3774(00)00144-X
[24] L. A. Richards, “Capillary Conduction of Liquids throu
Porous Mediums,” Journal of Applied Physics
gh
, Vol. 1, No.
5, 1931, pp. 318-333. doi:10.1063/1.1745010
[25] M. Th Van Genuchten, “A Closed-Form Equation for
Predicting the Hydraulic Conductivity of Unsaturated
Soils,” Soil Science Society of America Journal, Vol. 44,
No. 5, 1980, pp. 892-898.
doi:10.2136/sssaj1980.03615995004400050002x
[26] J. H. Holland, “Adaptation in Natural and Artific
tems: An Introductory Analysis with Application
ial Sys-
s to Bi-
, Berlin, 1996,
ology, Control, and Artificial Intelligence,” The Univer-
sity of Michigan Press, Ann Arbor, 1975.
[27] S. Schulze-Kremer, “Molecular Bioinformatics-Algo-
rithms and Applications,” Walter de Gruyter
pp. 13-108.