Remote Sens. 2013, 5, 1588-1602; doi:10.

3390/rs5041588
OPEN ACCESS

Remote Sensing
ISSN 2072-4292
www.mdpi.com/journal/remotesensing
Article

Estimating Crop Coefficients Using Remote Sensing-Based

Vegetation Index
Baburao Kamble 1,*, Ayse Irmak 2 and Kenneth Hubbard 2
1
Department of Civil Engineering, 3310 Holdrege Street, University of Nebraska-Lincoln,
Lincoln, NE 68583, USA
2
School of Natural Resources, 3310 Holdrege Street, University of Nebraska-Lincoln,
Lincoln, NE 68583, USA; E-Mails: [email protected] (A.I.); [email protected] (K.H.)

* Author to whom correspondence should be addressed; E-Mail: [email protected].

Received: 11 January 2013; in revised form: 13 March 2013 / Accepted: 15 March 2013 /
Published: 26 March 2013

Abstract: Crop coefficient (Kc)-based estimation of crop evapotranspiration is one of the

most commonly used methods for irrigation water management. However, uncertainties of
the generalized dual crop coefficient (Kc) method of the Food and Agricultural
Organization of the United Nations Irrigation and Drainage Paper No. 56 can contribute to
crop evapotranspiration estimates that are substantially different from actual crop
evapotranspiration. Similarities between the crop coefficient curve and a satellite-derived
vegetation index showed potential for modeling a crop coefficient as a function of the
vegetation index. Therefore, the possibility of directly estimating the crop coefficient from
satellite reflectance of a crop was investigated. The Kc data used in developing the
relationship with NDVI were derived from back-calculations of the FAO-56 dual crop
coefficients procedure using field data obtained during 2007 from representative US
cropping systems in the High Plains from AmeriFlux sites. A simple linear regression
model ( 1.457 0.1725) is developed to establish a general relationship
between a normalized difference vegetation index (NDVI) from a moderate resolution
satellite data (MODIS) and the crop coefficient (Kc) calculated from the flux data measured
for different crops and cropping practices using AmeriFlux towers. There was a strong
linear correlation between the NDVI-estimated Kc and the measured Kc with an r2 of 0.91
and 0.90, while the root-mean-square error (RMSE) for Kc in 2006 and 2007 were 0.16 and
0.19, respectively. The procedure for quantifying crop coefficients from NDVI data
presented in this paper should be useful in other regions of the globe to understand regional
irrigation water consumption.
Remote Sens. 2013, 5 1589

Keywords: crop coefficient; evapotranspiration; remote sensing; MODIS; NDVI; Nebraska

1. Introduction

The crop coefficient (Kc)-based estimation of crop evapotranspiration (ETc) is one of the most
commonly used methods for irrigation water management at the field scale. Crop evapotranspiration (ETc)
can be calculated using the Kc defined as the ratio of crop evapotranspiration to some reference
evapotranspiration (ETo) defined by weather data [1,2]. In this framework, Kc values are specific to
each crop (and to the method used for ETo) and have traditionally been derived from data sets, where
measured ETc for a well-watered crop is divided by a standard reference ETo (usually grass, but
sometimes alfalfa and, most recently, not crop-specific, but only a tall or short crop [3])). The crop
coefficients representative of well watered conditions are tabulated for the practical purpose of crop
water management among engineers, farmers and irrigation managers [3–5]. During the crop growing
season, the value of Kc for most agricultural crops increases from a minimum value at emergence, in
relation to changes in canopy development, until a maximum Kc is reached at about full canopy cover.
A crop coefficient curve is the seasonal distribution of Kc, often expressed as a smooth continuous
function in time or some other time-related index. The Kc tends to decline at a point after a full cover is
reached in the crop season. The declination extent primarily depends on the particular crop growth
characteristics [6,7] and the irrigation management during the late season [1]. Crop coefficients
primarily depend on the dynamics of canopies, light absorption by the canopy, canopy roughness,
which affects turbulence, crop physiology, leaf age and surface wetness [4]. As a crop canopy
develops, the ratio of transpiration to evapotranspiration increases, until most of the evapotranspiration
comes from transpiration, and soil evaporation is a minor component. This occurs because the interception
of radiant energy by the foliage increases until most light is intercepted before it reaches the soil. Moreover,
the relationships between spectral indices and crop coefficients are usually observed for the single crop
coefficient. The normalized difference vegetation index (NDVI) has been used extensively for vegetation
monitoring, crop yield assessment and drought detection [4,6–8]. Higher NDVI indicates a greater level
of photosynthetic activity [8,9]. Increase in crop coefficient caused by higher temperature results in a
decrease in soil water and a decline of NDVI, while dense vegetation induces more evapotranspiration
and lowers the land surface temperature [10]; or the transpiring canopy is cooler [11]. On the regional
scale, the current procedure to estimate Kc and resulting consumptive water use has limitations. The Kc
varies in space and in time, due to inherent variability in emergence date, land use pattern, antecedent
precipitation, emissivity, vegetation amount and atmospheric boundary conditions, such as air
temperature, wind speed and vapor pressure deficit.
A number of researchers have used multispectral vegetation indices derived from remote sensing to
estimate Kc values at the field scale for maize [11–15], wheat, cotton [11] and beans [13,14]. Bausch
and Neale demonstrated application of ground based physical remote sensing technique to relate
seasonal NDVI to Kc [15–17]. Several researchers [8,15–22] have shown that vegetation indices from
remote sensing could be used to predict “basal” crop coefficients for agricultural crops. Bastiaanssen
has discussed in detail the need to develop a methodology to determine aerial crop coefficient values
Remote Sens. 2013, 5 1590

for different agriculture land use patterns and practices [12]. Relationships between crop growth
parameters, meteorological variations and tillage practices encourage researchers to study the
experimental retrieval of crop coefficients at the field or regional scale. Leaf area index (LAI) has been
estimated by several researchers using NDVI [23,24], even though the crops they were monitoring
were both stressed and unstressed and, sometimes, were different crops. In the same manner we are
looking at rainfed and irrigated crops to determine if NDVI explains the variations between the crop
coefficients. Techniques are needed to map Kc weekly or biweekly for the purpose of consumptive
water use calculation at the field or regional scale. With the launch of NASA’s MODerate resolution
Imaging Spectroradiometer (MODIS) satellite, a new data set is available for utilization in daily crop
monitoring for water management. The Terra-MODIS satellite provides eight-day composite images
with a combination of spectral and thermal bands. The eight-day composite images are treated as
cloud-free products to monitor the Earth’s terrestrial activity [7,20].
The objective of this study was to investigate the applicability of time-series MODIS 250 m
normalized difference vegetation index (NDVI) data to the development of spatially representative
Kc. We will discuss below a simple regression approach for the retrieval of Kc and provide a brief
rationale for various components of the procedure. The specific objectives of this study were to
(a) analyze the seasonal dynamics of crop coefficients (Kc) and the vegetation index (NDVI),
(b) develop a regression model to establish the relationship between the NDVI and Kc values for
agricultural production systems and (c) evaluate the performance of the new model for estimating Kc
using an independent dataset for South-Central Nebraska. Implicit in our investigation is the
assumption that NDVI is specific to the crop located at each pixel.

2. Material and Methodology

2.1. Study Area and Crop Evapotranspiration Dataset

This study is conducted using data from three different locations for climate conditions in the High
Plains of the USA, located in the Midwest, between dense eastern forest and the western mountains
and deserts. This area is focused on agriculture production and is composed of ranches and farms,
where irrigated and rainfed farming systems present [25]. As of the last census of agriculture in 2007,
there were 2.2 million farms, covering an area of 373.12 million ha. Major crops grown in this area are
maize, sorghum, alfalfa, soybean, wheat and cotton. The first dataset is from agricultural production
fields and part of an ongoing carbon sequestration research program at the University of
Nebraska-Lincoln, Agricultural Research and Development Center (ARDC) near Mead, NE (Mead
irrigated (center-pivot) continuous maize production system (48.7 ha), Mead irrigated (center-pivot)
maize–soybean rotation (52.4 ha) and Mead rainfed maize–soybean rotation (65.4 ha)). All Mead
sites are within 1.6 km of each other and are found on silty clay loam soils. We obtained the Level 4
data product for AmeriFlux sites over the period of 2007 from the AmeriFlux website
(http://public.ornl.gov/ameriflux) (Table 1). The second dataset is from a rainfed grassland site at
Cottonwood, South Dakota. Both the Mead and Cottonwood sites are part of the AmeriFlux
network [26], with detailed measurement activities. The third site is an irrigated agricultural field at the
South Central Agricultural Laboratory (SCAL) (16 ha) located near Clay Center, NE, USA [26,27].
Remote Sens. 2013, 5 1591

Mean annual precipitation of Mead is 704 mm, and annual precipitation for 2007 is 927 mm; while
precipitation during the growing season (290–350 mm) received at the three sites during the growing
seasons of 2006 and 2007 (Table 1; Figures 1 and 2) was significantly large, the rainfed fields were not
stressed much and, therefore, were considered similar to the irrigated fields.
The field at South Central Agricultural Laboratory (SCAL) is an irrigated maize-soybean rotation
with a center-pivot irrigation system. Table 1 shows the details of each field. We obtained the hourly
temperature, latent heat flux data at 4 AmeriFlux [25,27] sites for the growing season of 2007 from the
data repository at Oak Ridge National Laboratory (Figure 1; Table 1).

Table 1. Vegetation type, geographical location and cropping patterns of the study sites
used to develop the normalized difference vegetation index (NDVI)-crop coefficient
(Kc) relationship.
Elevation Canopy
Year Name Latitude Longitude Vegetation Type Crop
(m) Height
Mead Irrigated Agriculture
2007 41.1649 −96.4701 362 1.83 m Maize
Rotation (maize-soybean rotation)
2007 Mead Rainfed 41.1797 −96.4396 363 1.71 m Agriculture (maize) Maize
Mead Irrigated
2007 41.1651 −96.4766 361 2.90 m Agriculture (continuous maize) Maize
Continuous
2007 Cottonwood 43.95 −101.8466 744 20–40 cm Grassland/range Grass
South-Central
Agriculture
2007 Agricultural Laboratory, 40.56667 −98.1333 552 N/A Maize
Soybean/maize
Clay Center
South-Central
Agriculture
2006 Agricultural Laboratory, 40.56667 −98.133333 552 N/A Soybean
(Soybean/maize)
Clay Center

Figure 1. Seasonal progression of weather (maximum (red) and minimum (gray)

temperature, precipitation (blue)) data at model calibration sites: (a) Mead Irrigated Rotation,
NE, USA (Year 2007); (b) Mead Irrigated, NE, USA (Year 2007); (c) Mead Rainfed, NE,
USA (Year 2007); (d)Cottonwood, SD, USA (Year 2007).
Remote Sens. 2013, 5 1592

Figure 1. Cont.

Figure 2. Seasonal progression of weather (maximum (red) and minimum (gray) temperature,
precipitation (blue)) data at model validation sites (a) South Central Agricultural Laboratory
(Year 2007), (b) South Central Agricultural Laboratory (Year 2006).

Figures 1 and 2 show the daily maximum (red) and minimum (gray) temperature and precipitation
(blue) data at model calibration and model validation sites, respectively. These sites are distributed
across the High Plains of the US and cover a wide range of agriculture vegetation types (Table 1). We
also used a published dataset of crop coefficients from SCAL [28], obtained from NEBFLUX
(Nebraska Water and Energy FLux Measurement, Modeling and Research Network) [13]. Both
networks have micro-metrological towers that are used to monitor fluxes over agriculture sites.

2.2. NDVI and Kc Model Development and Validation

The Hargreaves and Samani (1985) equation is a temperature-based empirical method developed
at Davis, California, using lysimeter data [29,30]. The Hargreaves and Samani (1985) equation is
originally developed for time periods of five days or longer to reduce the influence of the temperature
range, large variations in wind speed or cloud cover. Nevertheless, several studies have shown that
the Hargreaves equation provides reliable estimates of daily ETo [14,30]. Daily reference crop
evapotranspiration values were computed based on the equations presented by Hargreaves and Samani:
.
0.0135 1 . 17.8 (1)
Remote Sens. 2013, 5 1593

where is grass-reference evapotranspiration (mmd−1), Tmax is daily maximum air temperature (°C),
Tmin is daily minimum air temperature (°C), Tmean is daily mean air temperature (°C) and Ra is
extraterrestrial radiation (MJ·m−2·d−1).
KT is an empirical coefficient, which is a function of the temperature difference
0.00185 0.0433 0.423 (2)
was calculated for each day using the equation and methodology of Duffie and Beckman [31].
for the four sites was estimated using the above equation throughout the growing season of 2007.
The MODIS instrument offers new possibilities for large-area crop mapping by providing a daily
global coverage of high-quality, intermediate resolution (250 m) data since February 2000 at no cost to
the end user. In an analysis of a MODIS surface reflectance 8-day composite, 250 m (MOD09) images
were downloaded from the National Aeronautic and Space Administration (NASA) Land Process
Distributed Active Archive Center (LPDAAC) (https://lpdaac.usgs.gov/lpdaac/get data/data pool) for
the study area. The normalized difference vegetation index (NDVI) [32] was derived from downloaded
remote sensing data using near-infrared and red bands. The normalized difference vegetation index (NDVI)
utilizes reflectance of the canopy in the near-infrared (NIR) and red (R) bands of the spectrum [32]. This
vegetation index is given by:

(3)

where NIR is the reflectance in near infrared and R is the reflectance in the red region of the spectrum.
The NDVI values are reported to be well correlated with vegetation parameters, like leaf area index,
net primary productivity and gross primary productivity [23,33].
Crop coefficient (Kc) is used to calculate evapotranspiration. The ETc/ETo ratio is commonly
referred to as the crop coefficient [1]. To develop an NDVI-Kc relationship requires ground-truth data
for measured Kc values. We have used AmeriFlux data for modeling the NDVI-Kc relationship
(Table 1), and Kc values from SCAL for 2006 and 2007 were used for validation purposes [28]. The
daily Kc values from five locations (Table 1) for May, June, July and August were sampled for
particular Day of Year (DOY) when NDVI values were available. To avoid mixed pixel problems, we
used NDVI pixels located as close as possible to the center of the field. NDVI and satellite pass day for
respective station geolocations (see Table 1) in 2006 and 2007 were extracted from the analyzed
remote sensing dataset using ENVI software. For each satellite overpass date, we created a site and
land use-specific dataset. The NDVI-Kc linear regression model can be used to identify the relationship
between a single predictor variable NDVI and the response variable Kc when all the other predictor
variables in the model are “held fixed”. The goal of regression analysis is to express the response
variable as a function of the predictor variables. We used simple linear regression analysis to develop a
relationship between NDVI and Kc, which can be applied to a satellite image:

, (4)

where, for the ith case, is the response variable, , , , ,…, , are n repressors
and is a mean zero error term. The quantities and are unknown coefficients, whose values are
determined by least squares regression, where is the intercept and is slope for the jth term in the
Remote Sens. 2013, 5 1594

regression. The coefficients in the above equation were determined by the use of four different
agriculture landuse datasets and DOY throughout the season, so that the above equation is applicable
to a wide range of agriculture land use throughout the growing season. The coefficient of determination
(r2) represents the proportion of variability in that may be attributed to the linear combination of
the independent variables. The coefficient of determination is used to assess the degree of fit (r2 = 0,
no fit; r2 = 1, perfect fit) for the linear model shown in Equation (4):

∑
(5)
∑

To develop the above NDVI-Kc relationship, we did not conduct any site visits. Rather, we
quantified the , NDVI, their interactions and the actual evapotranspiration values for the natural
conditions of the fields, as reflected in the MODIS satellite images [7,20]. The NDVI-Kc model was
evaluated using data for another full season maize grown in the years 2006 and 2007 at the Clay
Center (Table 1). We used regression analysis and root-mean-square error to evaluate the model.
Regression analysis gives information on the relationship between the observed Kc variable and the
predicted Kc variable to the extent that information is contained in the data. For validation, we used a
plot of the observed Kc against the predicted Kc (using Equation (4)). Correlation between the actual
and predicted Kc values was then calculated to assess the goodness of fit. The coefficient of
determination (r2), which was used as a relative index of model performance, and root-mean-square
error (RMSE) (reference) was used to compare the observed Kc and predicted Kc crop coefficients.
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. Result and Discussion

3.1. Seasonal NDVI and Crop Coefficient Patterns for Selected Agricultural Land Use

We used the Ameriflux Mead sites and the Cottonwood site in 2007 as the model development
dataset, while model validation datasets included the South-Central Agriculture laboratory of UNL in
2006 and 2007. The results of the temporal progression of NDVI and crop coefficients for the four
selected sites for validation and modeling show that the phenology of the crops can be tracked by
MODIS-based NDVI (Table 1). Figures 3 and 4 shows the NDVI patterns of the different agriculture
land uses selected for modeling and validation. Irrigated maize from SCAL (Figure 4) shows smooth,
increasing greenness intensity with the tendency to go down at the end of the season.
An increase in vegetation activity is also observed before the cropping season, because of the
growth of vegetation at the onset of the spring rainy season. The respective fluctuation of crop
coefficients for agriculture and grass is seen throughout the season, which shows the strong correlation
between the NDVI and crop coefficients. The maximum NDVI of irrigated rotation, irrigated and
rainfed maize peaks at 0.84 by the end of July (DOY 215) and 0.77 (DOY 221), respectively. It is
obvious that the NDVI of irrigated maize is greater than that of rainfed maize throughout the growing
season (Figure 4). Maize is usually planted in late April to early May and harvested in late September
to early November.
Remote Sens. 2013, 5 1595

Figure 3. Seasonal progression of NDVI and Kc at model calibration sites: (a) Mead
Irrigated Rotation, NE, USA (Year 2007); (b) Mead Irrigated, NE, USA (Year 2007);
(c) Mead Rainfed, NE, USA (Year 2007); (d) Cottonwood, SD, USA (Year 2007).

Figure 4. Seasonal progression of NDVI and Kc at model validation sites: (a) South
Central Agricultural Laboratory (Year 2006); (b) South Central Agricultural Laboratory
(Year 2007).

It is interesting to note that the standard deviation for NDVI, as well as for Kc, of irrigated maize and
irrigated rotation are almost the same while for rainfed NDVI and Kc are lower at the Clay Center for
2006 and slightly higher in 2007, but the crop coefficient standard deviation is almost the same
(Table 2). During the growing season, the crop coefficient (Kc) for irrigated maize was approximately
0.3 ± 0.5 early in the season and 1.00 ± 0.05 during mid-season. From Mead Irrigated (a) and (b)
conditions (Figure 4), when NDVI is high and the vegetation is unstressed, the crop coefficient is on
average 1 and can reach 1.2, which shows that actual ET can be larger than the reference ET, which is
defined for well-irrigated condition. While the condition in Mead Rainfed, Cottonwood and SCAL
Remote Sens. 2013, 5 1596

(Figure 3) shows that lower NDVI are associated with the rainfed condition, owing to the under-developed
canopies and the lower capacity to absorb photosynthetically active radiation. Because of such a condition,
the crop coefficient is below 1.1, which shows that actual ET can be less than the reference ET.

Table 2. Statistics of NDVI and Kc for model calibration.

Mean Max Min Standard Deviation
Sites
NDVI Kc NDVI Kc NDVI Kc NDVI Kc
Cottonwood, SD, USA 0.318 0.279 0.490 0.693 0.257 0.103 0.083 0.208
Mead Irrigated Rotation, NE, USA 0.676 0.822 0.822 5.302 0.309 0.293 0.209 0.316
Mead Irrigated, NE, USA 0.706 0.905 0.894 1.212 0.277 0.323 0.218 0.323
Mead Rainfed, NE, USA 0.676 0.763 0.785 1.086 0.442 0.300 0.104 0.240
SCAL-2007 0.767 1.034 0.866 1.271 0.502 0.257 0.111 0.320
SCAL-2006 0.636 0.635 0.852 0.984 0.330 0.133 0.200 0.323

3.2. Development of NDVI and Crop Coefficient Relationship

In the development of the NDVI-Kc model to estimate the crop coefficient from remote sensing
NDVI, we simplify an agriculture landscape as a mixture of only irrigated agriculture, rainfed
agriculture and grass land. The ground measurement dataset of crop coefficients was collected from
Ameriflux for the growing season of 2007 and SCAL for the growing season of 2006 [28]. We used
the Mead sites and Cottonwood site in 2007 as the model development dataset, while model validation
datasets were used from SCAL in 2006 and 2007. The statistics (maximum, minimum, mean and CV)
for NDVI and crop coefficient values were given for each year for each site (Table 2). With the NDVI
from remote sensing and the corresponding crop coefficient from field measurement, a simple linear
regression equation was developed.
Figure 5 shows the relationship between NDVI and the crop coefficient for irrigated agriculture,
rainfed agriculture and the combined relationship for irrigated and rainfed crops. There was a strong
correlation between NDVI and the crop coefficient for all of the crops represented in the figure. The
combined relationship between NDVI and the crop coefficient for rainfed and irrigated agriculture
from all modeling locations is:
1.4571 0.1725 (6)
where 1.4571 and 0.1725 represent the slope and intercept coefficients. The correlation coefficient (r2)
is 0.8259, which is highly informative about the amount of variance that NDVI can explain in the Kc
data set, regardless of whether their joint distribution is normal. There was a strong correlation
between NDVI and the crop coefficient for all of the crops represented in the Figure 5. The NDVI-Kc
relationship for irrigated agriculture and rainfed agriculture is identical. There is no significant
difference in the regression, slope and intercept coefficients for irrigated and rainfed agriculture data.
Maize, both irrigated and rainfed, indicates a stronger relationship than for soybean. Furthermore, variation
in NDVI and was larger during mid- and late season in rainfed maize fields than for irrigated. This
could be attributed to variation in ETo, which increases evapotranspiration rates during crop development
and senescence, and also, the frequent irrigation condition in irrigated field makes fields more evaporative,
as compared to the rainfed condition, which was completely dependent on rainfall.
Remote Sens. 2013, 5 1597

Figure 5. Relationship between Terra-MODIS NDVI and AmeriFlux measured crop

coefficients under irrigated and rainfed crop condition.

3.3. Validation of NDVI and Crop Coefficient Relationship

In this section, we present the performance of the NDVI-Kc model (Equation (6)) for maize and
soybean fields at the SCAL Bowen-Ratio Energy Balance System (BREBS) site. We used the Mead
sites and Cottonwood site in 2007 as the model development dataset, while model validation datasets
were used from SCAL in 2006 and 2007. This validation site is independent from the calibration sites,
Mead and Cottonwood. The NDVI-Kc regression model (Equation (6)) developed in this study was
used to calculate Kc from NDVI values for the SCAL validation data sets for 2006 and 2007. The
progression of NDVI and crop coefficient shows the linear relationship. Figure 6 shows the linear
regression relationship between the measured crop coefficient and predicted crop coefficient values
from the model for year 2006 and 2007.
Obviously, the majority of crop coefficient estimates are accurate, with regression analysis near
90% for both fields, where estimates are from Equation (6). The coefficients of determination ( )
between measured and simulated crop coefficients values for 2006 and 2007 were, respectively, about
0.91 and 0.90. The RMSE between measured and predicted crop coefficients by the NDVI-Kc model
values were, respectively, about 0.16 and 0.19 for 2006 and 2007 for SCAL. The additional statistical
results presented in Table 3 confirm a reasonable performance of the NDVI-Kc model. Figure 7
illustrates the time course of measured and predicted crop coefficients by the NDVI-Kc model for
maize and soybean, respectively. However, some discrepancies between measured and predicted crop
coefficients by the NDVI-Kc model can still be seen (Figure 7).
Remote Sens. 2013, 5 1598

Figure 6. Validation of the NDVI-Kc model: (a) irrigated maize for growing season in 2006
and (b) soybean for 2007 in SCAL data. The graph depicts regression scatter plots of
estimated vs. observed crop coefficient.

Table 3. Performance of NDVI-Kc model for SCAL data: 2006 and 2007.
Statistical Index 2006 2007
Mean error 0.12 −0.09
Mean absolute error 0.14 0.17
Mean square error 0.02 0.03
Root mean square error 0.16 0.19
Ratio of standard deviations 0.9 0.51
Nash-Sutcliffe efficiency 0.75 0.62
Willmott index of agreement 0.93 0.84
Coefficient of persistence −0.01 0.62
Pearson product-moment correlation coefficient 0.95 0.95
Coefficient of determination 0.91 0.90

Figure 7. Seasonal progression of measured Kc and estimated Kc: (a) irrigated maize for
growing season in 2006 and (b) soybean for 2007 in SCAL data.
Remote Sens. 2013, 5 1599

3.4. Uncertainties, Errors and Accuracies for NDVI and Crop Coefficient Relationship

Comparison of the normalized difference vegetation index (NDVI) from moderate resolution
satellite data (MODIS) and the crop coefficient (Kc) indicate a useful relationship within the limits of
the estimated errors or uncertainties. The accuracy of the Kc via remote sensing estimates was ±9% at
the 90% confidence level. The Kc from remote sensing accuracy estimates are based on uncertainties
associated with input data, resolution and satellite data accuracy. The statistical procedure used to
calculate accuracy is explained in the previous section. Our results show that NDVI-Kc models estimate
Kc to ±0.2 (see RMSE in Table 3), with the low resolution NDVI estimation. In modeling Kc, it is
inevitable that both model and data input will present some uncertainty. Whatever model is used, the
errors in the input will propagate to the output of the calculated Kc. However, the NDVI and Kc used in
this model calibration and validation are often observed with sufficient accuracy to avoid large errors
in the residual. Other independent estimates of Kc from remote sensing are lacking in the literature, so
we cannot contrast to the results from other studies. In order to obtain realistic crop coefficient values
from remote sensing data, the model inputs and outputs must be correlated. It is critically important to
model and evaluate correlations of input variables. Ignoring the correlation between the input
parameters (NDVI) will add noise in the model output. Another reason for the errors in the Kc
estimation might be due to unknown irrigation schedules and high evaporation, because of
precipitation. As the prominent research already suggested, ±10% error prediction of the Kc values is
considered to represent levels of ET under pristine growing conditions. In the case above, the
inaccuracy of the measurement results is expressed as permissible error. The statistical data shows
some minor discrepancies between the observed and estimated Kc (Table 3). The coefficient of
determination and RMSE between and observed using BREBS were exceptional low. The
Willmott index of agreement is high, 0.93 and 0.84, respectively, for year 2006 and 2007. This is an
indication of the accuracy of the model, which suggests it can be used to estimate Kc with different
management practices. Biases in the model estimates were consistent over time and on the same order
for irrigated and rainfed crops as the NDVI- Kc uncertainty, so that it was concluded that the technique
is valuable for estimating the Kc from the remote sensing data.

4. Conclusion

This study developed a simple linear regression model ( 1.457 0.1725 ) to

establish a general relationship between a normalized difference vegetation index (NDVI) from
moderate resolution satellite data (MODIS) and the crop coefficient (Kc). Furthermore, because NDVI
is specific to the crop at each pixel, Kc is a direct representation of actual crop growth conditions in the
field. The crop coefficients were estimated spatially and temporally using the remote sensing model
applied to MODIS images taken during the year 2007. Results showed that variations in Kc are well
explained by variations in NDVI, especially for well-watered agricultural cropping systems. Kc had a
strong relation to NDVI during mid-season periods.
NDVI is an indicator of the density of vegetative cover and plant vigor; therefore, it is not surprising
that it captures most of the variation observed in Kc when there is no water stress condition. Validation
of the simple regression model yielded an r2 of.90 and an RMSE less than 0.17 for an irrigated maize
Remote Sens. 2013, 5 1600

field at SCAL in NE for two consecutive years (2006 and 2007). The systematic bias (±5%) may be
due to wet (2006) and dry (2007) effects on the crop evaporation, not represented in this Kc
formulation. The procedure for quantifying crop coefficients from NDVI data presented in this paper
should be useful in other regions of the globe to understand regional irrigation water consumption.
Biases and uncertainties in the model estimates were consistent over time and of the same order for
irrigated and rainfed crops as the NDVI-Kc model, so that it was concluded that the technique is
valuable for estimating the Kc from the remote sensing data. We have provided evidence that crop
coefficients quantified from remote sensing using this NDVI-Kc equation may be useful for irrigation
scheduling, evaluating irrigation/project performance, agricultural water budgets and estimating water
use efficiency. The NDVI-Kc model shows promise for use in water conservation in agriculture at the
local, regional and continental scales of measurement.

References

1. Allen, R.G.; Pereira, L.S.; Raes, D.; Smith, M. Crop Evapotranspiration—Guidelines for Computing
Crop Water Requirements—FAO Irrigation and Drainage Paper 56; FAO: Rome, Italy, 1998.
2. Allen, R.G.; Pereira, L.S.; Smith, M.; Raes, D.; Wright, J.L. The FAO-56 dual crop coefficient
method for predicting evaporation from soil and application extensions. J. Irrig. Drain. Eng.
2005, 131, 2–13.
3. Allen, R.G.; Walter, I.A.; Elliot, R.L.; Howell, T.A.; Itenfisu, D.; Jensen, M.E.; Snyder, R.L. The
ASCE Standardized Reference Evaportranspiration Equation; American Society of Civil
Engineers: Danvers, MA, USA, 2005; p. 59.
4. Justice, C.O.; Townshend, J.R.G. Special issue on the Moderate Resolution Imaging
Spectroradiometer (MODIS): A new generation of land surface monitoring. Remote Sens.
Environ. 2002, 83, 1–2.
5. Doorenbos, J.; Pruitt, W.O. Guidelines for Predicting Crop Water Requirements; Irrigation and
Drainge Paper 24; FAO: Rome, Italy, 1975.
6. Kamble, B.; Chemin, Y.H. GIPE in GRASS Raster Addons. Available online: https://svn.osgeo.org/
grass/grass-addons/grass6/imagery/gipe/i.vi.grid/description.html (accessed on 1 October 2012).
7. Kamble, B.; Irmak, A. Assimilating Remote Sensing-Based ET into SWAP Model for
Improved Estimation of Hydrological Predictions. In Proceeding of the 2008 IEEE International
Geoscience and Remote Sensing Symposium, Boston, MA, USA, 7–11 July 2008; Volume 3,
doi: 10.1109/IGARSS.2008.4779530.
8. Sellers, P.J. Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens. 1985, 6,
1335–1372.
9. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation.
Remote Sens. Environ. 1979, 8, 127–150.
10. Boegh, E.; Soegaard, H.; Hanan, N.; Kabat, O.; Lesch, L. A remote sensing study of the NDVI-Ts
relationship and the transpiration from sparse vegetation in the Sahel based on high resolution
data. Remote Sens. Environ. 1998, 69, 224–240.
11. Hunsaker, D.J.; Pinter, P.J.; Jr.; Kimball, B.A. Wheat basal crop coefficients determined by
normalized difference vegetation index. Irrig. Sci. 2005, 24, 1–14.
Remote Sens. 2013, 5 1601

12. Bastiaanssen, W.G.M. SEBAL-based sensible and latent heat fluxes in the irrigated Gediz Basin,
Turkey. J. Hydrol. 2000, 229, 87–100.
13. Irmak, S. Nebraska water and energy flux measurement, modeling, and research network
(NEBFLUX). Trans. ASABE 2010, 53, 1097–1115.
14. Jensen, M.E., Burman, R.D., Allen, R.G., Eds. Evapotranspiration and Irrigation Water
Requirements; ASCE: Reston, VA, USA, 1990.
15. Bausch, W.C.; Neale, C.M.U. Spectral inputs improve Maize crop coefficients and irrigation
scheduling. Trans. ASAE 1989, 32, 1901–1908.
16. Jayanthi, H.; Neale, C.M.U.; Wright, J.L. Seasonal Evapotranspiration Estimation Using Canopy
Reflectance: A Case Study Involving Pink Beans. In Proceedings of Remote Sensing and
Hydrology 2000, Santa Fe, NM, USA, 2–7 April 2000; pp. 302–305.
17. Irmak, A.; Ratcliffe, I.; Ranade, P.; Hubbard, K.; Singh, R.K.; Kamble, B.; Kjaersgaard, J.
Estimation of land surface evapotranspiration with a satellite remote sensing procedure. Great
Plains Res. 2011, 21, 73–88.
18. Benedetti, R.; Rossinni, P. On the use of NDVI profiles as a tool for agricultural statistics: The
case study of wheat yield estimate and forecast in Emilia Romagna. Remote Sens. Environ. 1993,
45, 311–326.
19. Choudhury, B.J.; Ahmed, N.U.; Idso, S.B.; Reginato, R.J.; Daughtry, C.S.T. Relations between
evaporation coefficients and vegetation indices studies by model simulations. Remote Sens.
Environ. 1994, 50, 1–17.
20. Irmak, A.; Kamble, B. Evapotranspiration data assimilation with genetic algorithms and SWAP
model for on-demand irrigation. Irrig. Sci. 2009, 28, 101–112.
21. Hubbard, K.G., Sivakumar, M.V.K., Eds. Automated Weather Stations for Applications in
Agriculture and Water Resources Management: Current Use and Future Perspectives. In
Proceedings of an International Workshop, Lincoln, NE, USA, 6–10 March 2000.
22. Allen, R.G.; Clemmens, A.J.; Burt, C.M.; Solomon, K.; O’Halloran, T. Prediction accuracy
for projectwide evapotranspiration using crop coefficients and reference evapotranspiration.
J. Irrig. Drain. Eng. 2005, 131, 24–36.
23. Gamon, J.A.; Field, C.B.; Goulden, M.; Griffn, K.; Hartley, A.; Joel, G.; Penuelas, J.; Valentini, R.
Relationships between NDVI, canopy structure and photosynthesis in three Californian vegetation
types. Ecol. Appl. 1995, 5, 28–41.
24. Tasumi, M.; Allen, R.; Trezza, R.; Wright, J. Satellite-Based energy balance to assess
within-population variance of crop coefficient curves. J. Irrig. Drain. Eng. 2005, 131, 94–109.
25. Hubbard, K.G. Climatic factors that limit daily evapotranspiration in sorghum. Clim. Res. 1992, 2,
73–80.
26. Baldocchi, D.; Falge, E.; Gu, L.H.; Olson, R.; Hollinger, D.; Running, S.; Anthoni, P.; Bernhofer,
C.; Davis, K.; Evans, R.; et al. FLUXNET: A new tool to study the temporal and spatial
variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. Bull. Am.
Meteorol. Soc. 2001, 82, 2415–2434
27. Verma, S.B.; Dobermann, A.; Cassman, K.G.; Walters, D.T.; Knops, J.M.H.; Arkebauer, T.J.;
Suyker, A.E.; Burba, G.G.; Amos, B.; Yang, H.S. Annual carbon dioxide exchange in irrigated
and rainfed maize based agro-ecosystems. Agric. For. Meteorol. 2005, 131, 77–96.
Remote Sens. 2013, 5 1602

28. Singh, R.; Irmak, A. Estimation of crop coefficients using satellite remote sensing. J. Irrig. Drain.
Eng. 2009, 135, 597–608.
29. Hargreaves, G.H.; Samani, Z.A. Reference crop evapotranspiration from temperature. Appl. Eng.
Agric. 1985, 1, 96–99.
30. Hargreaves, G.H.; Allen, R.G. History and evaluation of Hargreaves evapotranspiration equation.
J. Irrig. Drain. Eng. 2003, 129, 53–63.
31. Duffie, J.A.; Beckman, W.A. Solar Engineering of Thermal Processes; John Wiley & Sons:
New York, NY, USA, 1980.
32. Rouse, J.W.; Haas, R.H.; Schell, J.A.; Deering, D.W. Monitoring Vegetation Systems in the Great
Plains with ERTS. In Proceedings of Third ERTS Symposium, Washington, DC, USA, 10–14
December 1973; Volume 1, pp. 309–317.
33. Gitelson, A.; Vina, A.; Masek, J.; Verma, S.; Suyker, A. Synoptic monitoring of gross primary
productivity of maize using Landsat data. IEEE Geosci. Remote Sens. Lett. 2008, 5, 133–137.

© 2013 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution license
(http://creativecommons.org/licenses/by/3.0/).