Water Resources Management (2021) 35:2231–2247

https://doi.org/10.1007/s11269-021-02838-y

Estimation of Crop Water Requirement Based

on Planting Structure Extraction from Multi-Temporal
MODIS EVI

Changchun Xu 1 & Xicheng Zhang 1 & Jinxia Zhang 1 & Yapeng Chen 2 &
Teshome L. Yami 3 & Yang Hong 3

Received: 8 July 2020 / Accepted: 12 April 2021 / Published online: 6 May 2021
# The Author(s), under exclusive licence to Springer Nature B.V. 2021

Abstract
Estimation of crop water requirement (CWR) is key to the rational water use and
agricultural water conservation in arid regions. Using remote sensing data to obtain
long-term CWR over large areas helps water resources management in water-scarce
areas. This study, taking the Kaidu-Kongqi River basin in arid northwest China as the
study area, investigated the feasibility of synergistically using phenological characteris-
tics, Savitzky-Golay filter, harmonic analysis and decision tree to extract crop planting
structures (CPS) from MODIS EVI, and meanwhile analyzed the spatiotemporal variation
in the estimated CWR. The results show that the integrated method for CPS identification
and extraction is feasible and reliable with the classification accuracy over 80%. The mid-
season stage requires the most water and cash crops need more water than cereal crops.
Summer accounts for 69% the total growing season water use. The significant increase in
the area of high water demand crops such as cotton raised the total CWR of the basin
surging from 14.91× 108m3 in 2000 to 34.92×108m3 in 2017. The spatial distribution of
CWR was more related to crop types and area than to climatic conditions. Controlling the
expansion of arable land and optimizing the agricultural planting structure remain
important tasks for the sustainable management of water resources in the basin.

Keywords Crop water requirement . Penman-Monteith . Remote sensing . Phenological

characteristics . Savitzky-Golay filter . HANTS

1 Introduction

Water scarcity, a prominent environmental problem in arid and semi-arid regions in the world,
is seriously restricting the rapid development of regional socio-economy and eco-environment.

* Changchun Xu
[email protected]

Extended author information available on the last page of the article

2232 C. Xu et al.

It has and will be aggravated by the accelerated water cycle due to global warming and the
expansion of arable land (Schewe et al. 2014; Hagemann et al. 2013; Ruan et al. 2020). The
agricultural irrigation water accounts for more than 90% of total water consumption (Shen
et al. 2013) in these areas, which not only exacerbates the contradictions between supply and
demand and uneven distribution of water resources, but leads to a series of environmental
problems, such as the ecological degradation in the lower reaches of Tarim River from the
mid-1970s to the early 2000s. How to use the limited water wisely is an urgent issue for
people. Many water managers and scientists have been working on it and reached a consensus
that developing water-saving agriculture to improve water use efficiency and reduce the
proportion of agricultural water use is the most efficient solution, while this requires us to
know exactly how much water crops need.
Crop water requirement, equal to ETc in value, refers to the evapotranspiration from
disease-free, well-fertilized crops, grown in large fields, under optimum soil water conditions
and achieving full production under the given climatic conditions (Allen et al. 1998). The
measurements of ET mainly include hydrological method, micro-meteorological method, plant
physiology method and remote sensing method (such as surface heat balance and Penman-
Monteith model). Compared with others, the remote sensing method provides reliable infor-
mation for the quantification of evapotranspiration over large areas (Dalezios et al. 2019;
Supriyasilp et al. 2020; Mcnally et al. 2015; Gontia and Tiwari 2010; Pakhale et al. 2010). The
remote sensing-based evapotranspiration models, such as SEBAL (Bansouleh et al. 2015;
Bastiaanssen et al. 1998) and SEBS (Su 2002; Mahour et al. 2015) are considered as high
resolution models for ET estimation and applied to determine water demand (Kamali and
Nazari 2018; Samuel et al. 2018). However, the spatializations of a large number of parameters
often increase the complexity of estimation, reduce the accuracy of results and limit the
widespread use of models. The Penman–Monteith (P-M) method, a simple but physically
based approach for modelling the reference crop evapotranspiration (ET0), is rated better than
Priestley-Taylor and Hargreaves models (Ngongondo et al. 2013; Maina et al. 2012) and has
been widely used for ET estimation (Durdiev et al. 2021; Gong et al. 2020; Wang et al. 2018;
Sharifi and Dinpashoh 2014). In this paper, the P-M method was adopted to estimate ET0.
Then ETc was estimated by using the crop efficient approach.
The accurate extraction of crop planting structure is critical to the remote sensing estimation
of ETc. To achieve the purpose, most studies used higher spatial resolution images, such as
Landsat (Samuel et al. 2018), worldview-2 (Dalezios et al. 2019) and Sentinel-2 (Son et al.
2020; Liu et al. 2020a). Landsat images have a resolution of about 30 m and are available for
public use. However, good-quality Landsat data are often less available because of the
influence of heavy cloud cover, which results in the inability to obtain the optimal phenolog-
ical data from the few images. Worldview-2 images have a very high resolution of 0.5 m, but
they are commercial products which is costly and not suitable for large area remote sensing
estimation. Sentinel-2 has a high spatial resolution of 10 m, a high revisit time of 2–3 days and
is available for public use, but the time series is short (since 2015), which is not conducive to
understanding a long-term change in CPS and CWR in an area at present. Long series of
remote sensing data with high temporal resolution are still very important for practical
applications. The MODIS satellite has a very high temporal resolution, although the spatial
resolution is coarse (250m for NDVI/EVI). Most previous studies show that the MODIS data
could distinguish crops efficiently based on the phenological characteristics (Massey et al.
2017; Xu et al. 2014; Muhammad et al. 2015). It makes sense to employ some techniques to
improve the crop classification accuracy of the low-cost MODIS data for the ETc estimation
Estimation of Crop Water Requirement Based on Planting Structure... 2233

over large areas in data-sparse area. Many studies have been carried out and ways to improve
are various, such as learning algorithms (Feyisa et al. 2020), random forest (Hao et al. 2015),
classification and regression trees (Conrad et al. 2011), and support vector machine (Zhang
et al. 2014). Phenology-based crop classification could identify crop types by analyzing the
characteristics of key phenological periods in time series, which avoid the phenomenon of
misclassification caused by "the same spectrum of foreign objects" through using single
temporal image data. The high temporal resolution MODIS data are particularly suitable for
phenology-based crop classification. However, many studies based on phenological informa-
tion tend to use high spatial resolution data (Bargiel 2017) or fusion data (Li et al. 2021; Liu
et al. 2020b), fewer studies have focused on the joint use of phenological characteristics with
other methods to improve the accuracy of crop classification with low-cost MODIS data (Cao
et al. 2018). In this paper, we tested the reliability of synthetically using the phenological
characteristics, the Savitzky-Golay filter, the harmonic analysis of time-series and the decision
tree to enhance the identification accuracy of crops with MODIS EVI. Then, the classification
result was applied to estimate the CWR in the Kaidu-Kongqi River basin in Northwest China.
Finally, the spatiotemporal variation in CWR were analyzed. We expect the study to contribute
to the methodological improvement of crop classification, and to the integration of crop
classification, CWR estimation and agricultural water saving to improve water use efficiency
in water-scarce areas.

2 Data and Methods

2.1 Study Area

The Kaidu-Kongqi River, one of the tributaries of the Tarim River, locates on the southern
slope of Tianshan Mountains, Xinjiang, China (Fig. 1). The basin has an area of 7.73×104 km2
with longitude 82º56′ − 88º12′E and latitude 40º48′ − 43º20′N. The study area is the important
agricultural production base of Xinjiang, owing to the unique climate of high temperature, less
precipitation and abundant sunshine. The average annual precipitation is merely 155.6 mm,
but the heterogeneity is strong with 75% precipitation in summer, over 700mm in northwest
and 50mm in southeast. Rivers are fed by precipitation (45.4%) and snow and ice melt
(15.2%). River discharge is unevenly distributed throughout the year with 22.4% in spring,
44.9% in summer, 21.3% in autumn and 11.4% in winter. Agricultural irrigation accounts for
90% of the total water use.

2.2 Data and Materials

2.2.1 Remote Sensing Data

The remote sensing data includes MOD13Q1 (vegetation indices) and MCD12Q1 (land
cover) in 2000, 2005, 2010 and 2017 (https://modis.gsfc.nasa.gov). The MOD13Q1 data
at 250 m spatial resolution were used to extract crop types. There are a total of 356 images.
The orbit number involves h24v04, h24v05, h25v04 and h25v05. The MCD12Q1 land
cover data were used to help determine the distribution of crops. The MODIS Reprojection
Tool and the ArcGIS software were used to preprocess the images by splicing, projection
conversion and clipping.
2234 C. Xu et al.

Fig. 1 Map of the study area and the field survey sample sites

2.2.2 Meteorological Data

The daily meteorological data from 66 stations in Xinjiang were collected from China
Meteorological Data Sharing Network (http://data.cma.cn), including the minimum and
maximum temperatures, sunshine hours, relative humidity and average wind speed. The data
were used for ET0 calculation.

2.2.3 Field Survey Data

A field survey was carried out from July 1 to 15, 2017 in the study rea. A total of 275 sample
sites (Fig. 1) were selected for the training and validating of crop classification after eliminat-
ing the invalid data. There are seven major crops in the study area, wheat, corn, cotton, beet,
tomato, chilli and pear.

2.2.4 Statistical Data

The crop area data were collected from the Statistical Yearbook of Bayingolin Mongol
Autonomous Prefecture (2001, 2006, 2011 and 2018). The statistical data was used for the
accuracy verification of crop classification and extraction.

2.3 Methods

2.3.1 Savitzky-Golay Filter

The MODIS EVI is generated by the maximum value synthesis method, which can effectively
eliminate the interference of part cloud, aerosol and ground object bidirectional reflection, but
the effect of cloud cover remains (Wu et al. 2009), which will cause the jagged irregular
fluctuation in EVI curve and affect crop identification. The Savitzky-Golay (S-G) filter can be
used to smooth the EVI time-series and help improve the crop classification (Cao et al. 2018).
The details for the S-G filter process can be found in literature by Chen et al (2004).
Estimation of Crop Water Requirement Based on Planting Structure... 2235

To avoid the edge effect caused by the S-G filter, the EVI of the last three phases in
previous year and that of the first three phases in next year were added to the current
EVI sequence to form new EVI time-series before performing the S-G filter (Wang
et al. 2017).

2.3.2 Phenological Characteristics

The EVI time-series curve smoothed by the S-G filter can accurately reveal the characteristics
of annual dynamic change in crop growth (Li et al. 2011). The phenological parameters of
crops can be extracted from EVI curves by using the dynamic threshold method. In this paper,
four basic phenological parameters (start of growing season, SOS; end of growing season,
EOS; length of growing season, LOS; peak time, PT) were extracted by the TIMESAT
software.

2.3.3 Harmonic Analysis

Harmonic ANalysis of Time-Series (HANTS) is a time-series reconstruction method

based on the improved fast Fourier transform (Jakubauskas et al. 2001; Roerink et al.
2000). It discomposes the time series into an additive term (i.e., zero harmonic term,
without phase) and many sine or cosine curves with different frequencies. Each curve
(function) is defined by a unique amplitude and phase. HANTS can not only remove
cloud contamination points, but also require less stringent time-series images than the
fast Fourier transform (FFT). It has been widely used in satellite observation time-
series, such as NDVI, LAI, FPAR and LST (Wang et al. 2017; Zhou et al. 2015).
The amplitude and phase of harmonic components are quantitative measures of
vegetation phenology (Menenti et al. 1993). In this study, the basic parameters were
set as follows: the number of frequencies (NOF) is 4 (Jakubauskas et al. 2001), the fit
error tolerance (FET) is 0.05, and the degree of over determinedness (DOD) is 8.

2.3.4 Decision Tree Classification

Based on the field survey sampling data, according to the reconstructed EVI curves by the S-G
filter, the EVI values of key phases, the phenological parameters and the harmonic parameters,
the planting structure extraction models of seven major crops were constructed separately for
crop identification and extraction by using the decision tree classification method.

2.3.5 Estimation of Crop Water Requirement

The crop coefficient approach (Doorenbos and Pruitt 1977) was used to produce an estimate of
ETc which is expressed as follows,

ET c ¼ K c  ET 0 ð1Þ

Where, ETc is crop evapotranspiration; Kc is crop coefficient which is divided into four growth
stages: initial, crop development, mid-season and late season. Kc changes with natural climatic
conditions. ET0 is reference crop evapotranspiration which is computed by the Penman-
Monteith equation recommended by FAO-56 (Allen et al. 1998).
2236 C. Xu et al.

0:408ðRn  GÞ þ  Tþ273
900
U 2 ðes  ea Þ
ET 0 ¼ ð2Þ
þð1 þ 0:34U 2 Þ

Where, ET0, reference crop evapotranspiration (mm/day); Rn, net radiation at the crop surface
(MJ/(m2d)); G, soil heat flux density (MJ/(m2d)); T, mean daily air temperature at 2 m height
( ); U2, wind speed at 2 m height (m/s); es, saturation vapor pressure at Tc (kPa); ea, actual
vapor pressure (kPa); es-ea, saturation vapor pressure deficit (kPa); Δ, slope of saturation vapor
pressure and temperature relationship curve (kPa/ ); γ, psychrometric constant (kPa/ ) .

3 Results and Analysis

3.1 Phenological Characteristics

The phenological characteristics of the seven crops were extracted from the S-G filtered
EVI curves (Table 1). All the curves were found showing one peak throughout the year,
indicating that the seven crops are all annual plant (Fig. 2). In the entire growing season,
the EVI peak of wheat and pear was far lower than the other crops. Between wheat and
pear, PT of wheat was earlier, SOS of pear was earlier, and LOS of pear was longer. Thus,
wheat and pear were relatively easy to identify. In the remaining crops, the SOS and EOS
of cotton both were the latest, the EOS and LOS of corn was earlier and the shortest
respectively, so these two crops were also easy to distinguish. For the remaining three
crops (chilli, tomato and beet), the phenological differences between them were too small
to be identify. Therefore, not all crops can be identified with just the phenological
information and the S-G filter.

3.2 Harmonic Analysis of EVI Curves

The reconstructed EVI curves by the S-G filter were decomposed into several harmonic
components with HANTS (Table 2). The amplitude of the additive term reflects the mean of
EVI curve. Obviously, the amplitude of wheat was the smallest and beet was the largest, which
was consistent with the results shown in Fig. 2. The amplitude of the first term is the largest,
explaining the most information of the original curve. The amplitude of chilli, tomato and beet
was quite different from each other: beet was the largest, chilli was the second, and tomato was
the smallest, so they can be distinguished by amplitude. For the crops with similar EVI mean,
such as pear/tomato or corn/cotton/beet, the amplitude of the first term can also be applied to
distinguish them.

Table 1 Phenological parameters of the seven main crops in the Kaidu-Kongqi River basin

Crop SOS EOS LOS PT

Wheat 120 257 137 158

Corn 144 256 111 197
Cotton 150 287 136 206
Beet 131 273 142 206
Pear 112 288 176 197
Chilli 137 269 132 202
Tomato 128 265 137 192
Estimation of Crop Water Requirement Based on Planting Structure... 2237

Fig. 2 The S-G filtered and reconstructed EVI curves for the seven major crops

In terms of phase, values of the first term are the largest, corresponding to the peak time of
EVI. Cotton had the largest phase and wheat had the least, indicating that the EVI peak for
cotton appeared the latest, that for wheat occurred the earliest. Therefore, both amplitude and
phase of harmonic can provide additional useful information for crop identification. They can
not only reveal the phenological characteristics precisely, but also magnify the differences
among them to help enhance the accuracy of crop classification.

3.3 Accuracy Verification of Crop Classification

The accuracy of crop classification was verified by both the confusion matrix and the statistical data.
As to the confusion matrix, the user’s and producer’s accuracies were all more than 80%, the overall
accuracy was 87.86% and the Kappa coefficient was 0.86 (Table 3), showing that the crop
classification was good.
In terms of statistical verification, 19 out of 25 of the relative accuracy (RA) were more than
90%, 7 out of 19 RA were more than 98%, and the average RA was 91.2% (Table 4), further
indicating that the combination of the S-G filter, phenological characteristics, harmonic
analysis and decision tree was feasible and reliable.

Table 2 Harmonic parameters of the seven main crops in the Kaidu-Kongqi River basin

Crop Amplitude Phase

0 1 2 3 1 2 3

Wheat 2448 1933 711 152 169 60 157

Corn 3371 2571 396 189 201 197 124
Cotton 3482 2965 424 166 209 192 201
Beet 3503 3325 520 200 198 173 172
Pear 2927 1883 232 29 192 85 172
Chilli 3274 3008 474 182 198 231 211
Tomato 2940 2534 423 138 190 161 172

Notes: number 0–3 represents harmonic term.

2238 C. Xu et al.

Table 3 Confusion matrix verification of crop classification

Validation data User’s accuracy Producer’s accuracy

Cotton Pear Chilli Tomato Beet Wheat Corn Total

Cotton 18 0 0 0 2 0 0 20 90.0 90.0

Pear 0 18 0 0 1 0 0 19 94.7 90.0
Chilli 0 0 17 3 0 0 0 20 85.0 85.0
Tomato 0 0 2 17 0 2 0 21 81.0 85.0
Beet 0 0 1 0 16 0 1 18 88.9 80.0
Wheat 0 2 0 0 0 18 0 20 90.0 90.0
Corn 2 0 0 0 1 0 19 22 86.4 95.0
Total 20 20 20 20 20 20 20 140

3.4 Spatialization of ET0 and Kc

The meteorological data from 66 observation stations in the whole Xinjiang were used to
calculate ET0 by the Penman-Monteith formula. Then the calculated ET0 were interpolated by
the ANUSPLIN interpolation method (Hutchinson and Xu 2013) and the data within the study
area were extracted. The Kc used in this study were mainly referred to other related studies
(Duan et al. 2004) and the standard crop coefficients recommended by FAO (Table 5). To
obtain the spatial distribution of ET0 in the whole basin, Kc was spatially assigned to the seven
crops at different growth stage based on the obtained crop type distribution map (Fig. 3). Kc in
2017 was shown in Fig. 4 as an example.

3.5 Estimation and Analysis of CWR

3.5.1 CWR at Growth Stages for Different Crops

The CWR per unit area (CWRu, mm) of the seven crops in growing season (Fig. 5) showed
that the CWRu ranged between 20-60mm in the initial stage, 60-320mm in the crop develop-
ment stage, 170-470mm in the mid-season stage, and 50-300mm in the late season stage.
Overall, the mid-season stage required the most water, and the initial stage demanded the least.
There were large differences in CWRu among stages for all the crops except tomato. Through-
out the whole growing season, CWRu was greater than 600mm for all seven crops. Among

Table 4 Statistical verification of crop classification

Crops 2000 2005 2010 2017

SA EA RA SA EA RA SA EA RA SA EA RA

Wheat 326.7 355.9 91.1 393.7 245.7 62.4 308.5 203.8 66.0 361.9 312.4 86.3
Corn 124.6 133.3 93.0 162.4 161.8 99.6 216.4 204.6 94.6 323.3 319.7 98.9
Cotton 821.5 792.9 96.5 1262.4 1146.6 90.8 1775.3 1640.0 92.4 2790.7 2283.5 81.8
Beet 102.1 115.1 87.3 98.0 94.8 96.7 89.1 92.0 96.8 82.5 81.8 99.1
Pear / 378.0 289.5 76.9 460.2 436.8 94.9 480.8 481.1 99.9
Chilli / 80.9 81.5 99.3 215.8 211.9 98.2 449.4 434.4 96.7
Tomato / 121.5 113.3 93.3 200.9 220.8 90.1 165.4 168.0 98.4

Notes: SA, statistical area from yearbook (km2 ); EA, extracted area from images (km2 ); RA, relative accuracy
(%); /, no statistical data
Estimation of Crop Water Requirement Based on Planting Structure... 2239

Table 5 Crop coefficient at different growth stages for the seven major crops in the Kaidu-Kongqi River basin

Growth stage LOS Growing season

A B C D

Wheat 0.17 0.17–1.16 1.16 1.16–0.40 130 03/04–07/11

Corn 0.21 0.21–1.20 1.20 1.20–0.35 150 04/26−09/23
Cotton 0.26 0.26–1.20 1.20 1.20–0.70 180 04/20−10/16
Beet 0.34 0.34–1.21 1.21 1.21–0.70 190 03/20−09/25
Pear 0.45 0.45–0.95 0.95 0.95−0.70 205 04/07–10/27
Chilli 0.60 0.60–1.05 1.05 1.05–0.90 160 04/01–09/07
Tomato 0.60 0.60–1.15 1.15 1.15–0.80 140 04/01–08/28

Notes: A, initial stage; B, crop development stage; C, mid-season stage; D, late season stage

them, sugar beet needed the most water (910.73mm) and wheat needed the least (603.09mm).
In comparison, economic crops (such as cotton, beet and pear) required more water than cereal
crops (such as wheat and corn).

3.5.2 Annual Variation in CWR for Different Crops

The annual variation in CWRu of the seven crops changed synchronously during 2000–2017 with
less difference between years (Fig. 6a), indicating that they responded to the external climatic
environment consistently. However, the total CWR (CWRt) for each crop varied quite differently
(Fig. 6b). During 2000–2017, the CWRt of cash crops such as cotton, pear and chilli increased
significantly. The crop planting structure in the study area had changed greatly over the last 20 years.
Cotton is the mainstay of agriculture in Northwest China. In our study area, cotton area
continued to grow explosively, increasing from 133.31 km2 in 2000 to 161.75 km2 in 2005,
204.63 km2 in 2010 and 319.69 km2 in 2017, with a growth rate of 10.96 km2 per year.
Correspondingly, water demand for cotton climbed sharply from 6.29×108m3 in 2000 to
13.47×108m3 in 2010 and 20.01×108m3 in 2017. The growth rate was 0.81×108m3 per year.
Similarly, water demand for pear and chilli increased sharply too. They were the second and
third largest water consumer after cotton in the study area. Water demand for tomato had
dropped since 2010 after a quick increase from 2000 to 2010. These reflect changes in crop
cultivation structure in the study area in response to market changes.

3.5.3 Temporal Variation in CWR for All Crops

Except for the seven main crops, there were some other crops scattered in the study area. They
were too small in area to be identified, but the total area was large enough not to be ignored.
For this part, the CWRu of them was substituted with the mean of the seven crops. Then the
total water requirement of all crops was the sum of the two parts.
Consistent with the crop growth rhythm, the CWRt for all crops showed a single-peak
distribution characteristic, with 15.04×108m3 in summer, 3.53×108m3 in spring and 3.36×108m3
in fall, accounting for 69%, 16% and 15% water required in the whole growing season,
respectively (Fig. 7a). The CWRt for all crops soared from 14.91×108m3 in 2000 to
34.92×108m3 in 2017, approaching the average annual surface runoff over the last 30 years
(37.7×108m3). If irrigation water efficiency is considered, the total water demand of the crop alone
would exceed the water supply, not to mention the water demand of the natural vegetation. The
2240 C. Xu et al.

Fig. 3 Crop types and spatial distribution extracted from MODIS EVI
Estimation of Crop Water Requirement Based on Planting Structure... 2241

Fig. 3 (continued)
2242 C. Xu et al.

Fig. 4 Spatial distribution of crop coefficient (Kc) at the four growth stages in 2017

opposing trends in CWRu (Fig. 6a) and CWRt (Fig. 7b) from 2005–2010 feedback the under-
standing that the decrease in CWR because of changing climate conditions cannot offset the
increase in CWR due to increased human activities. Therefore, controlling the expansion of arable
land, improving the efficiency of water use and optimizing the agricultural cultivation structure
are still important tasks for the sustainable water resources management in the basin.

3.5.4 Spatial Variation in CWRt for All Crops

The annual CWRt for each pixel in the growing season was obtained by spatial calculation,
ranging between 34,000–66,000m3 (Fig. 8). For comparison, it was divided into four classes,

Fig. 5 Crop water requirement per unit area (CWRu) at growth stages (bar, left y-axis) and that in growing
season (line, right y-axis) for the seven main crops
Estimation of Crop Water Requirement Based on Planting Structure... 2243

Fig. 6 Interannual variations in the CWRu (a) and CWRt (b) for the seven main crops

each class spanning 8000 m3. The entire basin was mainly covered by the second class
(42,000 ~ 50,000 m3) and the third class (50,000 ~ 58,000 m3).
Large differences were found in the spatial distribution and variation in CWRt of the
two sub-basins. In the Kongqi River sub-basin, the CWRt was relatively stable with less
change. Since 2005, the basin was dominantly covered by the expanding third class. The
variation can be attributed to the simple and homogeneous crop types (Fig. 3) and the
similar water requirements per unit area of the two major crops (cotton and pear, Fig. 6a)
in the Kongqi sub-basin. In the Kaidu River sub-basin, however, the CWRt varied
drastically in space. There was a tendency for the third and fourth class to expand,
indicating water demand was increasing dramatically. The increase in water requirement
in the study area was more related to complex crop types and increased area than to
changes in climatic conditions.

4 Conclusions and Discussion

Crop water requirement is important basis for rational and effective allocation of water
resources in arid areas. It is necessary to use remote sensing to achieve large-scale crop water
requirement estimation. This study used a combination of several techniques to improve the
identification accuracy of low-cost MODIS data, which is important for water resources
management in water-scarce areas. Results showed that the combination use of S-G filter,
phenological characteristics, HANTS and decision tree was feasible and reliable. It can help
improve the crop classification and extraction. The mid-season stage and the initial stage
required the most and least water, respectively. Water requirement per unit area (CWRu)
throughout the growing season were all more than 600mm for the seven crops. Overall, cash
crops demand more water than cereal crops. The annual change in CWRu was less, but the

Fig. 7 Variations in monthly CWRt (a) and CWRt in growing season (b) for all crops
2244 C. Xu et al.

Fig. 8 Spatial distribution of CWRt in growing season for all crops in 2000, 2005, 2010 and 2017

total water requirement (CWRt) varied greatly due to the significant increase in the area of
some high water-demanding crops, such as cotton, pear and chilli, which caused the CWRt of
the whole basin surging from 14.91 × 108m3 in 2000 to 34.92 × 108m3 in 2017. Summer
demanded the most water, accounting for 69% of water required in the whole growing season.
The spatiotemporal distribution and variation in CWRt during 2000–2017 was more related to
the crop types and area than to the climatic conditions.
The integrated method for crop classification in the study showed good effects in
MODIS data, which is significantly important for getting crop types, area, water demand
and other indicators over large areas at low cost, particularly suitable for the regions with
large-scale fields and simple crop types. However, the problem of mixed pixels cannot be
completely overcome due to the spatial resolution of MODIS data itself. When the data
accuracy is demanding, the scale of farmland is small, or the homogeneity of underlying
surface is poor, the MODIS data could not meet the requirements well. To obtain better
results, employing the multi-source fusion data with higher temporal, spatial and spectral
resolution will be a good choice. With the development of remote sensing technology, the
extraction of a large number of parameters required for evapotranspiration models has
become possible. Thus, a combination use of evapotranspiration models with high-quality
fusion data will be the future direction of remote sensing hydrology or agricultural remote
sensing. In the process, downscaling or upscaling of data is still an issue that needs to be
addressed.

Acknowledgements This work was jointly supported by the National Natural Science Foundation of China
(Nos. 41561023, 42067062); the China Scholarship Council Program (No. 201808655036). Thanks to China
Meteorological Administration and NASA for providing free data. Thanks to the editors and anonymous
reviewers for their detailed and constructive comments.
Estimation of Crop Water Requirement Based on Planting Structure... 2245

Author Contributions All authors contributed to the study conception and design. Data collection and
processing was performed by Xicheng Zhang and Jinxia Zhang. Artworks were modified by Yapeng Chen.
The language was polished by Teshome L. Yami and Yang Hong. The draft of the manuscript was written by
Changchun Xu. All authors commented on previous versions of the manuscript and approved the final
manuscript.

Declarations

Conflict of Interest The authors declare that they have no conflict of interest.

References

Allen RG, Pereira LS, Raes D, Smith M (1998) Crop evapotranspiration: guidelines for computing crop water
requirements. FAO Irrigation and drainage paper 56. FAO, Rome
Bansouleh BF, Karimi AR, Hesadi H (2015) Evaluation of SEBAL and SEBS algorithms in the estimation of
maize evapotranspiration. Int J Plant Soil Sci 6:350–358. https://doi.org/10.9734/IJPSS/2015/15711
Bargiel D (2017) A new method for crop classification combining time series of radar images and crop
phenology information. Remote Sens Environ 198:369–383. https://doi.org/10.1016/j.rse.2017.06.022
Bastiaanssen WGM, Menenti M, Feddes RA, Holtslag AAM (1998) A remote sensing surface energy balance
algorithm for land (SEBAL): 1. Formulation. J Hydrol 212–213:198–212a. https://doi.org/10.1016/S0022-
1694(98)00253-4
Cao R, Chen Y, Shen M, Chen J, Zhou J, Wang C, Yang W (2018) A simple method to improve the quality of
NDVI time - series data by integrating spatiotemporal information with the Savitzky - Golay filter. Remote
Sens Environ 217:244–257. https://doi.org/10.1016/j.rse.2018.08.022
Chen J, Jönsson P, Tamura M, Gu Z, Matsushita B, Eklundh L (2004) A simple method for reconstructing a
high-quality NDVI time-series data set based on the Savitzky–Golay filter. Remote Sens Environ 91:332–
344. https://doi.org/10.1016/j.rse.2004.03.014
Conrad C, Colditz RR, Dech S, Klein D, Vlek PLG (2011) Temporal segmentation of MODIS time series for
improving crop classification in Central Asian irrigation systems. Int J Remote Sens 32:8763–8778. https://
doi.org/10.1080/01431161.2010.550647
Dalezios NR, Dercas N, Spyropoulos NV, Psomiadis E (2019) Remotely sensed methodologies for crop water
availability and requirements in precision farming of vulnerable agriculture. Water Resour Manag 33:1499–
1519. https://doi.org/10.1007/s11269-018-2161-8
Doorenbos J, Pruitt WO (1977) Guidelines for predicting crop water requirements. Irrigation and Drainage Paper
24. FAO, Rome
Duan A, Sun J, Liu Y, Xiao J, Liu Q, Qi X (2004) Irrigation quota of major crops for Northern China. China
Agricultural Science and Technology Press, Beijing
Durdiev K, Chen X, Huang Y, Ilkhom M, Liu T, Friday O, Abdullaev F, Gafforov K, Omurakunova G (2021)
Investigation of crop evapotranspiration and irrigation water requirement in the lower Amu Darya River
Basin, Central Asia. J Arid Land 13:23–39. https://doi.org/10.1007/s40333-021-0054-9
Feyisa GL, Palao LK, Nelson A, Gumma MK, Paliwal A, Win KT, Nge KH, Johnson DE (2020) Characterizing
and mapping cropping patterns in a complex agro-ecosystem: an interactive participatory mapping procedure
using machine learning algorithms and MODIS vegetation indices. Comput Electron Agric 175:105595.
https://doi.org/10.1016/j.compag.2020.105595
Gong X, Wang S, Xu C, Zhang H, Ge J (2020) Evaluation of several reference evapotranspiration models and
determination of crop water requirement for tomato in a solar greenhouse. Am Soc Horticult Sci 55(2):244–
250. https://doi.org/10.21273/HORTSCI14514-19
Gontia NK, Tiwari KN (2010) Estimation of crop coefficient and evapotranspiration of wheat (Triticum
aestivum) in an irrigation command using remote sensing and GIS. Water Resour Manag 24:1399–1414.
https://doi.org/10.1007/s11269-009-9505-3
Hagemann S, Chen C, Clark DB, Folwell S, Gosling SN, Haddeland I, Hanasaki N, Heinke J, Ludwig F, Voss F,
Wiltshire AJ (2013) Climate change impact on available water resources obtained using multiple global
climate and hydrology models. Earth Syst Dynam 4:129–144. https://doi.org/10.5194/esd-4-129-2013
Hao P, Zhan Y, Wang L, Niu Z, Shakir M (2015) Feature selection of time series MODIS data for early crop
classification using random forest: a case study in Kansas, USA. Remote Sens 7(5):5347–5369. https://doi.
org/10.3390/rs70505347
2246 C. Xu et al.

Hutchinson MF, Xu T (2013) ANUSPLIN version 4.4 user guide. The Australia National University, Fenner
Schoolof Environment and Society, Canberra
Jakubauskas ME, Legates DR, Kastens JH (2001) Harmonic analysis of time-series AVHRR NDVI data.
Photogramm Eng Remote Sens. 67:461–470
Kamali MI, Nazari R (2018) Determination of maize water requirement using remote sensing data and SEBAL
algorithm. Agric Water Manag 209:197–205. https://doi.org/10.1016/j.agwat.2018.07.035
Li Z, Tang H, Yang P, Zhou Q, Wu W, Zou J, Zhang L, Zhang X (2011) Responses of cropland phenophases to
agricultural thermal resources change in Northeast China. Acta Geogr Sin 66(7):928–939
Li R, Xu M, Chen Z, Gao B, Cai J, Shen F, He X, Zhang Y, Chen D (2021) Phenology-based classification of
crop species and rotation types using fused MODIS and Landsat data: the comparison of a random-forest-
based model and a decision-rule-based mole. Soil Tillage Res 206:104838. https://doi.org/10.1016/j.still.
2020.104838
Liu L, Xiao X, Qin Y, Wang J, Xu X, Hu Y, Qiao Z (2020a) Mapping cropping intensity in China using time
series Landsat and Sentinel-2 images and Google Earth Engine. Remote Sens Environ 239:111624. https://
doi.org/10.1016/j.rse.2019.111624
Liu X, Zhai H, Shen Y, Lou B, Jiang C, Li T, Hussain SB, Shen G (2020b) Large-scale crop mapping from
multisource remote sensing images in google earth engine. IEEE J Sel Top Appl Earth Obs Remote Sens 13:
414–427
Mahour M, Stein A, Sharifi A, Tolpekin V (2015) Integrating super resolution mapping and SEBS modeling for
evapotranspiration mapping at the field scale. Precision Agric 16:571–586. https://doi.org/10.1007/s11119-
015-9395-8
Maina M, Amin MSM, Wayayok A, Asha TS (2012) Evaluation of different ET0 calculation methods: a case
study in Kano state, Nigeria. Philipp Agric Scientist 95(4):378–382
Massey R, Sankey TT, Congalton RG, Yadav K, Thenkabail PS, Ozdogan M, Meador AJS (2017) MODIS
phenology-derived, multi-year distribution of conterminous U.S. crop types. Remote Sens Environ 198:490–
503. https://doi.org/10.1016/j.rse.2017.06.033
Mcnally A, Hustak GJ, Brown M et al (2015) Calculating crop water requirement satisfaction in the West Africa
Sahel with remotely sensed soil moisture. J Hydrometeorol 16:295–305. https://doi.org/10.1175/JHM-D-14-
0049.1
Menenti M, Azzali S, Verhoef W, Swol RV (1993) Mapping agroecological zones and time lag in vegetation
growth by means of Fourier analysis of time series of NDVI images. Adv Space Res 5:233–237. https://doi.
org/10.1016/0273-1177(93)90550-U
Muhammad S, Niu Z, Wang L, Aablikim A, Hao P, Wang C (2015) Crop classification based on time series
MODIS EVI and ground observation for three adjoining years in Xinjiang. Spectrosc Spectr Anal 35(5):
1345–1350
Ngongondo C, Xu C, Tallaksen L, Alemaw BF (2013) Evaluation of the FAO Penman-Monteith, Priestley-
Taylor and Hargreaves models for estimating reference evapotranspiration in southern Malawi. Hydrol Res
4:706–722. https://doi.org/10.2166/nh.2012.224
Pakhale G, Gupta P, Nale J (2010) Crop and irrigation water requirement estimation by remote sensing and GIS:
a case study of Karnal District, Haryana, India. Int J Eng Technol 2(4):207–211
Roerink GJ, Menenti M, Verhoef W (2000) Reconstructing cloud free NDVI composites using Fourier analysis
of time series. Int J Remote Sens 21:1911–1917. https://doi.org/10.1080/014311600209814
Ruan H, Yu J, Wang P, Wang T (2020) Increased crop water requirement have exacerbated water stress in the
arid transboundary rivers of Central Asia. Sci Total Environ 713:136585. https://doi.org/10.1016/j.scitotenv.
2020.136585
Samuel A, Girma A, Zenebe A, Ghebreyohannes T (2018) Spatio-temporal variability of evapotranspiration and
crop water requirement from space. J Hydrol 567:732–742. https://doi.org/10.1016/j.jhydrol.2018.01.058
Schewe J, Heinke J, Gerten D et al (2014) Multimodel assessment of water scarcity under climate change. PNAS
9:3245–3250. https://doi.org/10.1073/pnas.1222460110
Sharifi A, Dinpashoh Y (2014) Sensitivity analysis of the Penman-Monteith reference crop evapotranspiration to
climatic variables in Iran. Water Resour Manag 28:5465–5476. https://doi.org/10.1007/s11269-014-0813-x
Shen Y, Li S, Chen Y, Qi Y, Zhang S (2013) Estimation of regional irrigation water requirement and water
supply risk in the arid region of Northwestern China 1989–2010. Agric Water Manage 128:55–64. https://
doi.org/10.1016/j.agwat.2013.06.014
Son NT, Chen CF, Chen CR, Guo HY (2020) Classification of multitemporal Sentinel-2 data for field-level
monitoring of rice cropping practices in Taiwan. Adv Space Res 65:1910–1921. https://doi.org/10.1016/j.
asr. 2020.01.028
Su Z (2002) The surface energy balance system (SEBS) for estimation of turbulent heat fluxes. Hydrol Earth Syst
Sci 6:85–100. https://doi.org/10.5194/hess-6-85-2002
Estimation of Crop Water Requirement Based on Planting Structure... 2247

Supriyasilp T, Pongput K, Boonyanupong S, Suwanlertcharoen T (2020) Enhance water management for Muang
Fai irrigation system through remote sensing and SWOT analysis. Water Resour Manag. https://doi.org/10.
1007/s11269-020-02724-z
Wang P, Xun L, Li L, Xie Y, Wang L (2017) Extraction of planting areas of main crops based on Fourier
transformed characteristics of time series leaf area index products. Transactions of the Chinese Society of
Agricultural Engineering 21:207–215
Wang J, Liu X, Cheng K, Zhang X, Li L, Pan G (2018) Winter wheat water requirement and utilization
efficiency under simulated climate change conditions: a Penman-Monteith model evaluation. Agric Water
Manage 197:100–109. https://doi.org/10.1016/j.agwat.2017.11.015
Wu W, Yang P, Tang H (2009) Comparison of two fitting methods of NDVI time series datasets. Trans Chin Soc
Agric Eng 11:183–188
Xu Q, Yang G, Long H, Wang C, Li X, Huang D (2014) Crop information identification based on MODIS
NDVI time-series data. Trans Chin Soc Agric Eng 11:134–144
Zhang J, Feng L, Yao F (2014) Improved maize cultivated area estimation over a large scale combining MODIS-
EVI time series data and crop phenological information. ISRS J Photogramm Remote Sens 94:102–113
Zhou J, Li J, Massimo M (2015) Reconstruction of global MODIS NDVI time series: Performance of Harmonic
ANalysis of Time Series (HANTS). Remote Sens Environ 15:217–228. https://doi.org/10.1016/j.rse.2015.
03.018

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Affiliations

Changchun Xu 1 & Xicheng Zhang 1 & Jinxia Zhang 1 & Yapeng Chen 2 & Teshome L. Yami 3 &
Yang Hong 3
1
Key Laboratory of Oasis Ecology, College of Resource and Environmental Sciences, Xinjiang University,
Urumqi, Xinjiang, China
2
State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese
Academy of Sciences, Urumqi, Xinjiang, China
3
Hydrometeorology and Remote Sensing Laboratory, School of Civil Engineering and Environmental
Sciences, The University of Oklahoma, Norman, OK, USA