Journal of Arid Environments 188 (2021) 104461

Contents lists available at ScienceDirect

Journal of Arid Environments

journal homepage: www.elsevier.com/locate/jaridenv

Assessment of the agricultural water budget in southern Iran using

Sentinel-2 to Landsat-8 datasets
Arnaud Caiserman a, *, Farshad Amiraslani b, c, Dominique Dumas a
a
Geography and Planning Research Center, Faculty of Geography, Jean Moulin Lyon 3 University, 7th Chevreul Street, Lyon, 69007, France
b
Department of Remote Sensing and Geographic Information, Faculty of Geography, University of Tehran, Tehran, 1417853933, Iran
c
Ulster University, School of Geography and Environmental Sciences, Coleraine, UK

A R T I C L E I N F O A B S T R A C T

Keywords: This paper is a first attempt to compute the total water needs of an agricultural plain with remote sensing and
PYSEBAL ground data in Iran. The cropping areas were mapped with Sentinels-2 images, based on NDVI profiles classi­
Crop mapping fication. This model was validated and 85% of the areas were correctly classified. Second, the crop water needs
Iran
were computed using PYSEBAL and Landsat-8 images. Crop evapotranspiration (ETseason) and Irrigation Re­
Irrigation
Sentinel-2
quirements (IRseason) were calculated for each crop and then validated by comparing IR collected in the field from
Landast-8 farmers with computed IRPYSEBAL on 5 plots. IRPYSEBAL underestimated the reality with an average of 10% while
Remote sensing the overestimation average was 17%. The second validation was the comparison of Daily ET from FAO-56
method and Daily ET PYSEBAL showed a RMSE of 0.67 mm/day and MAE of 0.52 mm/day, which assesses
the accuracy of PYSEBAL. ETseason varies according to weather parameters in the plain and IRseason, according to
different irrigation practices. The most water demanding crops were identified: rice (IR: 1427 mm) and corn
(669). The total water balance of Marvdasht was negative in 2018 with 0.2859 km3 of extracted groundwater for
irrigation for only 0.098 km3 of available water for aquifers recharge.

1. Introduction using remote sensing. In addition, such an analysis of crop water needs
enabled us to assess the total use of groundwater during that year. The
Countries are not equally affected by water shortage issues and they Marvdasht Plain is fully in the prism of climate change as it records a
adopt different approaches for agricultural adaptation to droughts and decrease in rainfall of 1.1 mm/decades over the period 1988–2015
climate change. Thereby, this study puts forward the case of Iran in the (Roshan and Negahban, 2015) as well as an increase in temperature of
West of Asia as one of the countries severely affected by water issues 0.05–0.99C◦ /decades since 1975 (Soltani et al., 2016). Droughts have
(Faramarzi, 2010; Karimi et al., 2018; Keshavarz et al., 2014; Madani, also been frequent over the last forty years, particularly in 1981, 1982,
2014; Madani et al., 2016; Zehtabian et al., 2010). The country is facing 1983, 1985, 1987, 2003, 2004 2008 and 2011, during which drought
a double challenge: (1) an increasing water needs due to growing pop­ severity strongly affected agricultural production (Ahani et al., 2012;
ulation (Motamed, 2017; Neuve-Eglise, 2007; Saatsaz, 2019) and (2) an Keshavarz et al., 2014; Keshavarz and Karami, 2013). These past and
increase of drought frequency over the last two decades and future future climate changes make it essential to estimate water use by agri­
climate change scenarios (Amiraslani and Caiserman, 2018; Golian culture, as support to political decision-making for the decades to come.
et al., 2015; Keshavarz and Karami, 2013; Tabari et al., 2012). However, Monitoring water consumption of crops appears as a key issue to high­
Iran has attempted to increase the agricultural productivity with greater light the crops which might exacerbate water shortage, in the name of
access to water for irrigation to support food security. food security. Moreover, the assessment of water balance and IR per
The objective of this paper is to identify sensitive crops which require crop type is the first attempt in this region according to our literature
significant amounts of irrigation in one of the most important agricul­ review in English. Remote sensing is a useful tool and has already proven
tural zones of Iran (Hassanshahi et al., 2015; Moameni, 1999): the plain its relevancy to monitor agriculture and water issues, especially under
of Marvdasht in the Fars Province. For this purpose crop Evapotranspi­ arid and semi-arid conditions.
ration (ETseason) and Irrigation Requirements (IRseason) were computed The application of remote sensing in agriculture are subdivided as

* Corresponding author.
E-mail address: [email protected] (A. Caiserman).

https://doi.org/10.1016/j.jaridenv.2021.104461
Received 7 April 2020; Received in revised form 6 October 2020; Accepted 2 February 2021
Available online 18 February 2021
0140-1963/© 2021 Elsevier Ltd. All rights reserved.
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

follows (Asgarian et al., 2016): (1) agricultural dynamics and the evo­ been assessed in many countries between 85% and 95% on experimental
lution of crop areas with low resolution images such as MODIS (250 m), fields (Liou and Kar, 2014). The validation was conducted with lysim­
(2) precision agriculture with high resolutions images such as Quickbird eter measurements in several countries with Root Mean Square Error
(0.65), Pleiade (0.7 m) or RapidEye images (5 m) for yields estimations, (RMSE) of 0.7 in Spain, 0,03 in China, 0.14 in Nigeria or 0,6 in Italy
soil humidity assessment or weed prevention and (3) crop type classi­ (Water Watch, 2019).
fication with medium resolution images such as Landsat-8 (30 m) or The second section of the paper introduces the chosen region in Iran
Sentinel-2 (10 m). The present paper is considering the third approach of and the requisite data from to ground to the satelite images to compute
agriculture through crop mapping and crop water needs estimation. the agricultural water budget. In the third section, the results of crop
Numerous studies have already developed methodology to map crop mapping and PYSEBAL will be explained and interpreted. Eventually,
areas with satelite images (Belgiu and Csillik, 2018; Hao et al., 2018; the fourth section consists in the discussion of the paper, namely the
Heupel et al., 2018; Kenduiywo et al., 2018; Lamb and Brown, 2001; validation of crop mapping and PYSEBAL through field works and global
Panigrahy and Sharma, 1997; Song et al., 2017; Waldhoff et al., 2017; literature review, and the perspectives of these models will be shown in
Xie et al., 2007; Zhong, 2012). This paper used a new process, recently the same part.
developed for another case of study in Lebanon using Sentinel-2 images
for its good resolution (10 m) (Caiserman et al., 2019). This method was 2. Methods and materials
divided in three steps: (a) a new way to extract fields boundaries by
stacking monthly high NDVI pixels to highlight the cultivated areas, (b) 2.1. Study area: Marvdasht plain
the retrieval of crop calendars and (c) the classification of pixels. The
novelty of this methodology was its simplicity and reproducibility. In The study area is located in the Fars Province, southern Iran
addition, this crop mapping process was based on field works, increasing (29◦ 52′ 34N - 52◦ 48′ 22E, elevation: 1600 m) and covers 95000 ha
the reliability of the outputs. Remote sensing has another role for water (Fig. 1). The current climate is the Mediterranean characterised by two
needs estimations. Numerous algorithms have already been developed contrasted seasons between wet winters and dry and hot summers. Ac­
such as SEBI (Menenti and Choudhury, 1993), SEBS (Su, 2002), S-SEBI cording to local climate stations, Marvdasht annually receives 440 mm
(Roerink et al., 2000), METRIC (Allen et al., 2007), TSM (Norman and in the northern part and 275 mm in the more arid area in the southern
Becker, 1995), SAMIR (Simonneaux et al., 2009) and PYSEBAL using part and 73% of the precipitation occurs in winter (based on annual
Landsat-8 images (Bastiaanssen et al., 1998a, 1998b; Hessel, 2019; average on 1990–2017 period). Thereby, the irrigation is necessary from
Hessel et al., 2017). This paper selected the latest version of PYSEBAL May until October and the annual average of potential evapotranspira­
since it does not require a significant amount of data and its accuracy has tion reaches 1680 mm (Attarod et al., 2016). Nevertheless,

Fig. 1. Study area: the Marvdahst Plain, Fars Province and its annual average precipitations from 1990 to 2017 (Sources (Islamic Republic of Iran Meteorological
Organisation, 2018):

2
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

precipitations are highly variable and numerous droughts occurred in conduct this comparison, the Euclidean Distance (ED) was computed
the recent decades (Ahani et al., 2012; Keshavarz et al., 2014; Keshavarz (Equation (1)).
and Karami, 2013; Khosravi et al., 2017). √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑n
ED = (a − b)2 (1)
2.2. Sentinel-2 and Landsat-8 imagery i=1

This study has required a double dataset of images: monthly Where a is the NDVI average of sampled fields and b, the unknown NDVI
Sentinel-2 images for crop mapping in spring (images from January to profiles to classify. n, is the number of month from the beginning until
June) and summer (July to December) 2018 and monthly Landsat-8 the end of the season, depending on crop.
imagery over the same year for crop water needs according to PYSE­ types. When the ED is close to 0, the unknown NDVI profiles is
BAL inputs requisites (Fig. 2). First, 11 Sentinels-2 images were down­ considered as the crop type from the NDVI average of sampled fields.
loaded (images in February were too cloudy, therefore the average was According to the results, a value of 0.4 was used as a threshold to define
computed between January and March NDVI) to assess the evolution of how close to 0 the ED was. If the ED classification could not find any crop
pixel’s greenness throughout the season. It appeared that one image per type, our research classified the pixel as “spring vegetables” since there
month was sufficient to depict and differentiate the NDVI profiles of were numerous fields of garlic, carrots, onions or other vegetables ac­
each crop type (Caiserman et al., 2019). In addition, winter images were cording to fields observations. Contaminated pixels on the limits of the
most of the time too cloudy (>80%) to enable the use of several images fields were removed by splitting fields with roads map, digitalized from
per month throughout the whole season. The images were downloaded GoogleEarth images, with a broad buffer which separates every field
using the USGS Data Explorer (USGS, 2019). This evolution enabled to (roads and ditches).
distinguish them according to crop types collected on the ground. A field The last step was the validation of the classification. This was
survey was conducted across the Marvdasht plain during the agricultural computed by using the sampled and known fields. They were classed
season (January–August) in 2018. The aim was to record GPS-based into two equal groups: known fields and false unknown fields. For
locations of each crop type in spring and summer (Table 1). The crop instance for wheat, the 244 known fields were divided into two groups of
calendars were extracted from interviews with 60 farmers and also from 122. The first 122 were used to compute the NDVI profile of wheat (crop
NDVI temporal profiles of sampled crop types (based on Table 1). calendar) and the other 122 were tested for classification according to
On each ‘Day Of Year’ timeframe, hourly ground data from an ED to the first group. The outputs of this validation are presented in the
indicative station – Sad Doroudzan (X: 30,17, Y: 52,78, Z:1600) (Fig. 3) – results.
were acquired to compute an instantaneous ET (ETinst), during the time
of overpass (11:00 GMT): solar radiation (solar panel), wind speed 2.4. PYSEBAL ETseason and IRseason
(anemometer), temperature (thermometer) and relative humidity
(shelter with humidity sensor, Fig. 3). Sad Doroudzan station in Marv­ PYSEBAL estimates the transfer of energy from the solar radiation to
dasht plain was chosen as the reference station due to its proximity to the water transfer to the atmosphere. This transfer can be quantified
fields and daily data recordings. PYSEBAL computes ET with a Stan­ with the estimation of crop evapotranspiration (ET) and PYSEBAL re­
dardized Penman-Monteith Equation (Waters et al., 2002). quires the calculation of ET reference (ET0) from ground datasets. In
Marvdasht, previous studies on rice evapotranspiration suggested using
2.3. Crop mapping the Equation of Hargreaves-Samani (Fooladmand et al., 2008; Fool­
admand and Haghighat, 2007) for its accuracy in semi-arid climates and
Before any classification, the cultivated areas needed to be retrieved in Iran in general (Valipour and Eslamian, 2014). The Equation of ET0 is
from satelite images without manual digitizing. Normalized Difference written as follows (Equation (2)):
Vegetation Index was used on Sentinel-2 images (NDVI) to delineate
fields boundaries. The cultivated areas had a high NDVI (over 0.3) from ET0 = 0.00256∗(Tmean + 17.8)∗(Tmax − Tmin )0.5 Ra (2)
March to November. The selection of these areas with the Raster Where Tmean is the daily average temperature, Tmax and Tmin are the
Calculator in Arcmap (version 10.5.1) on each image, the polygonisation maximum and minimum temperature during the day. Ra is the daily
of these green areas and the stack of the monthly boundaries gave a final extraterrestrial radiation (MJ/m2/day) and assesses the income radia­
map of field boundaries. tion to the atmosphere. Ra is obtained from the original FAO Equation
From the sampled fields and crop types, the evolution of NDVI (FAO, 1986):
temporal profiles was extracted to differentiate crops calendar. The
average of NDVI within the sampled fields were used to construct crops 24*60
Ra = Gsc dr [ωs sin(ϕ)sin(δ) + cos(ϕ) cos(δ) sin ​ (ωs )] (3)
NDVI profiles (Fig. 4). π
Then, the crop type of unknown pixels needed to be classified. For Where Gsc is the solar constant 0.0820 MJ/m2/min, dr the inverse
this purpose, the NDVI profiles of all the previously delimited fields were relative distance Earth-Sun, ωs the solar angle and ϕ the radian latitude.
computed and compared to the NDVI profiles from sampled fields. To Once the data have been compiled (ET0, Landsat-8 images, Solar

Fig. 2. Timeframe of Landsat-8 and Sentinel-2 images for assessing crop water needs and crop mapping in Marvdasht (2018).

3
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Table 1
Sampled crop types and land uses in Marvdasht plain during the field work in 2018 (January–August).
Spring Alfalfa Canola Orchard Wheat Bare soil Urban

(534) 20 24 134 244 26 86

Summer Alfalfa Orchard Corn Rice Sugar beet Tomato Fallow Bare soil Urban
(719) 20 134 60 174 100 79 40 26 86

Fig. 3. Daily ground data from Sad Doroudzan station for ETinst estimation in PYSEBAL model in 2018.

radiation, Wind speed, Temperature and Relative humidity), one can Comparably, RL↑ Equation uses Ts the surface temperature instead of
run the new version of PYSEBAL: PYSEBAL 3.4.0.0 (Hessel et al., 2017). Ta (K). Waters et al. (2002) suggested to compute surface temperature
ETinst is first processed from the Latent Heat Flux (λET), based on the using the first band thermal (band 10) and the K1 (774.8853) and K2
subtraction of Net Radiation (Rn), Soil Heat Flux (G) and Sensible Heat (1321.0789) constants of each image. The corrected thermal radiance
Flux (H): λET will be later converted into an amount of evaporated from the surface is estimated with the Equation of Wukelikc (Wukelic
water, calibrated with the ET0 from the station. et al., 1989) using the spectral radiance of band 6 to convert Digital
Numbers (DN) to Radiance.
λET = ​ Rn ​ – ​ G ​ – ​ H ​ (4)
Once Rn has been computed, one can proceed with the Soil Heat Flux
The overall steps of PYSEBAL are summed up in a flowchart (Fig. 5). (G) in W/m2, defined as a ratio with Rn:
Rn (W/m2) is the result of the subtraction of all of the outcoming / / ( )( )
G Rn = ​ Ts α 0.0038α + 0.0074α2 1 − 0.98NDVI4 (7)
radiation (through reflection) to all of the incoming radiations (long and
shortwaves): The multiplication of this ratio with Rn gives G. Second, the Sensible
Rn = ​ RS↓ − α ​ RS↓ + ​ RL↓ − ​ RL↑ − (1 − εo )RL↓ (5) Heat Flux (H) in W/m2 is assessed by Equation (17). This step is the most
complex since it uses some ground data which must be correctly
Where RS↓ matches to the shortwave incoming radiation in W/m2 recorded. In addition, one needs to select the hot and cold pixels to
estimated from the solar constant (the theoretical amount of solar en­ compute H (selection according to criteria Table 2), which determines
ergy on 1 m2), the relative distance Earth-Sun and the atmospheric the quality of the outputs.
transmissivity which describes the transparency of the atmosphere, α is ( )/
the surface albedo (the reflected fraction of sunlight from the Earth H ​ = ρ × cp × ​ dT rah (8)
surface, written by Waters (Waters et al., 2002) and εo, thermal emis­ Where ρ is the air density (1.225 kg/m3), cp the amount of heat
sivity in W/m. required to change the temperature of 1C◦ (1004 J/kg/K), dT the tem­
Second, RL↓ (the incoming longwave radiation) is estimated from Ta perature difference between two heights (z1: 0.1 m and z2: 2 m above
the near-surface air temperature at the station (in Kelvin, K), εa the at­ surface) and rah the aerodynamic resistance to heat transport in s/m (the
mospheric emissivity derived from the atmospheric transmissivity with heat and vapour transfer from the surface to the edge of the canopy)
a coefficient of 0.85 (Hessel et al., 2017; Waters et al., 2002), and σ, the under the influence of wind speed (from Sad Doroudzan) and surface
content of Stefan-Boltzmann to describe reflected energy from a black roughness using NDVI and albedo in PYSEBAL. The definition of tem­
body on the ground (5.67 × 10− 8 W/m2/K): perature at 2 m of height must be computed with indicative pixels: cold
RL↓ = εa × σ × Ta 4 (6) and hot.

4
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 4. Sentinel-2 NDVI profiles of crop types and land uses from sampled fields in Marvdasht plain in 2018.

Once Rn, G and H were computed, then λET (Equation (4)) could be from the Sad Doroudzan (when Landsat-8 overpasses) is expressed as
calculated. The latent heat flux is converted in an amount of ET at the follows:
time of Landsat-8 overpass:
ETinst
ETr F = (10)
λET ET0
ETinst = 3600 (9)
λ Where ETrF is named ET fraction. ETrF values are close to crop co­
The conversion from second to the hour is necessary to compute the efficients and enable to calibrate ET to different crop in the next Equa­
hourly ET when the satellite overpasses the plain. λ is the necessary tion (Waters et al., 2002). This ratio of ETrF is then used to calibrate the
latent heat to change a kilo of water from liquid to gaseous state daily ET0 of Sad Doroudzan over the whole seasons (the length depends
(326.508 J/kg). The ratio between ETinst from PYSEBAL and the ET0 on the crop calendars of each crop). Since ET0 to the station is available

5
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 5. Flowchart of PYSEBAL, modified after Waters et al. (2002).

Table 2
Criteria which were used to select hot and cold pixels in PYSEBAL.
Criteria Values Purpose Remarks

Cold LAI 4–6 Delineate a threshold of ET on the most Cold pixels excluded water bodies where there was no transpiration. Thereby, pixels
pixels NDVI NDVImax- well-watered pixels, where ET is the covered by well-irrigated vegetation were selected.
0.1xNDVIstd highest.
Ts 284–295 K
albedo 0.22–0.24
Elevation 1400–1800 Removing the mountainous areas
range

Criteria Values Purpose Remarks

hot LAI 0–0.4 Delineate a threshold of ET on the less well- Dry fields with a weak vegetation coverage were favoured. Transpiration and
pixels NDVI 0.03–0.20 watered pixels, where ET is the lowest evaporation would be low, but not null. Hot pixels did no cover deserts since the
Ts 302–310 K surface temperature was too high
albedo 0.13–0.15
Elevation 1400–1800 Removing the mountainous areas
range

every hour of the season, the ration between ETinst and ETr can be version 2.18.3.
calculated on each image and converted into a seasonal ET of several Afterwards, the monthly and seasonal Irrigation Requirements (mm
months: or m3) can be estimated on each farm from the Equation of the Food and
Agriculture Organisation by the subtraction of ET and net rainfalls. Net
∑
n
ETseason = ETr Fseason ETr24 (11) rainfalls were computed with the CROPWAT software version 8.0 using
1 USDA Soil Conservation Service methodology which requires pre­
Where ETrFseason is ETrF within the season of one crop or another and cipitations as an input (Ewaid et al., 2019):
n
∑
ETr24 , the sum of daily ETr over the season. In order to delineate the Pnet =
(P ∗ (125 − 0, 2∗3∗P))
for P < =
250
(12)
1 125 3
beginning and the end of the season, PYSEBAL recommends to retain the
first day of the first month of the season of wheat for instance and the 125 250
Pnet = + 0, 1*P for P ​ > (13)
last day of the last month. Ground data of Sad Doroudzan on the day pass 3 3
of Landsat-8 are necessary to interpolate ETinst to the all plain. The sum Were Pnet is the effective rainfalls and P, the seasonal precipitations
of ET24 gives the total evapotranspiration of crops (ETseason). For each recorded in Doroudzan station in 2018 (in mm). And IR Equation:
crop areas, the average, the minimum and the maximum of ET are
computed inside the fields polygons using Zonal Statistics from Qgis IRseason = ETseason − Pnet season (14)

6
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

The calculation of IRseason for rice is somewhat different. Indeed, in calculating Root Mean Square Error (RMSE) and Mean Average Error
the FAO method (Brouwer et al., 2001), rice is an exception to the (MEA) between Daily ET FAO-56 and the Daily PYSEBAL average of
ET-Pnet Equation. Irrigation inputs for rice are calculated with the plots larger than 4 ha.
following Equation (15). The root zone of the rice is saturated during the
sowing period with an initial supply of 200 mm (SAT). Then farmers 3. Results
apply a 100 mm water layer to keep the seedlings in water (WL). Finally,
to compensate for the water lost through percolation in the soil, a daily 3.1. Crop classification accuracy
supply of several mm of water is needed to ensure soil water saturation
(PERC). This percolation has already been calculated and recorded with Regarding the farm boundaries, with R: 0.95 and R2: 0.91 (Fig. 6),
lysimeter data in the Marvdasht plain (Pirmoradian et al., 2002) and is one can consider the model as accurate enough and these automated
measured at 3.4 mm/day for 4 months due to the fine-textured alluvial fields limits can be further used for crop areas classification. Also, the
soils in the area. We, therefore, used this study as a reference to calculate crop classification accuracy was estimated based on data precision,
the water requirement of rice and this value to compute PERC on our recall and overall accuracy (Table 3).
same study area. From spring to summer, the minimal overall accuracy is 73.56%
(rice), the minimal recall 0.57 (alfalfa) and the minimal precision 0.56
2.5. Crop classification validation (corn). In addition, rice and corn could be mixed up for their very similar
crop calendars. Nonetheless, higher NDVI values of rice at mid-summer
It was necessary to assess the accuracy of the farm boundaries (Fig. 4) enabled the distinction of these two crop types. Otherwise, the
delineation through the comparison of areas from 60 automatically classification of crop areas appeared as accurate enough to be used to
extracted and manually digitized fields with GoogleEarth images. R and compute the crop and eventually the total water budget in Marvdasht
R2 were calculated to assess the reliability of the fields boundaries plain.
extraction. Once the farm boundaries are validated, the classification Crop mapping can be improved, as showed in the previous case of
itself should be assessed with the computation of precision, recall and study that used this methodology with Sentinel-2 images (Caiserman
overall accuracy. et al., 2019). In this study, it was assumed that a greater number of
GPS-based points per crop types (for crop calendars extraction) would
2.6. Crop water needs validation enhance the accuracy of maps. Thereby, this paper showed that crops
with similar agricultural calendars remain difficult to be distinguished,
The key point of this research was the validation of PYSEBAL outputs but the precision, recall and overall accuracy of the crop maps in
through the comparison of IRPYSEBAL and IR from the field surveys with Marvdasht plain were still satisfying and make these maps convenient
farmers. The surveys targeted information about the yields and when­ for crop water needs estimations.
ever possible, the irrigation calendar. More importantly, the surveys
aimed to target the amount of irrigated water on crops. Whenever 3.2. PYSEBAL’s results accuracy
farmers knew the amount of irrigation they applied on their farms, the
GPS coordinates were collected and the Irrigation Requirements (IR) In most of the cases, PYSEBAL underestimates the reality with an
from the surveys on these fields were compared to the IR of PYSEBAL on average of 10% (Table 4). The best estimation is the plot n◦ 4 (wheat)
the same fields. These information collected from farmers might be where PYSEBAL only underestimated the reality of 1.96%. The state­
incorrect or incomplete. Consequently, not all of the 60 surveys were ments of the farmers and PYSEBAL outputs were highly correlated. On
retained as accurate and relaible enough to validate the IR. However, a the other hand, the worst example was another wheat field (over­
group of 5 farmers wrote down and precisely knew these information estimation of 17%) where IRPYSEBAL was 459 mm/ha/season and IRfields,
(times and volume) that were used to validate IRPYSEBAL: three fields of 384 mm. This might be due to errors from the farmers who probably
wheat, one of alfalfa and one of corn. On the sampled plots (where underestimated the amount of irrigated water. The pixels of the outputs
farmers knew exactly the amount of irrigation water they brought), this could be also overlapped with other fields and the estimation not ac­
study located and estimated the cultivated surface with GoogleEarth curate. Nonetheless, the overall estimation is satisfying and PYSEBAL is
images, then multiplied the frequency of irrigation (number of irrigation therefore considered as reliable enough to compute crop water needs.
session) with the amount of irrigation water per session (m3) and the Moreover, the comparison between Daily ET FAO-56 and PYSEBAL
output obtained was the IR of these fields. Then, IRfield with IRPYSEBAL confirms the underestimation of PYSEBAL in most cases (Table 5, Fig. 7).
were compared on the same plot. PYSEBAL outputs match to the ET and Indeed, all Daily ET PYSEBAL values are lower than those of Daily ET
then to the IR/ha/season (after net rainfall subtraction) per pixel. FAO-56 except for orchards due to a wider range of ET PYSEBAL average
Therefore, this study compared IRfields and IRPYSEBAL per ha and per
season. Consequently, despite the lack of lysimeter data, the accuracy of
ETPYSEBAL could be assessed in this study through the comparison of
IRfield and IRPYSEBAL.
This lack of lysimetric data led us to consider a second process to
validate ETseason of PYSEBAL. Indeed, this difficulty of the lack of data
collected on the ground has already been encountered in other studies
that recommend comparing Daily ET of PYSEBAL with Daily ET of the
FAO-56 method (Stancalie et al., 2010) using Allen’s method (Allen
et al., 1998). This method consists in calculating Daily ET by calibrating
ET0 of the climatic station representative of the plain – here the Dor­
oudzan station – with the crop coefficients (Kc) taken from the litera­
ture. Crop coefficients are indeed not the same from one crop to another
and vary over time according to the crop phenological stage. It is
important to compare the results of PYSEBAL in the Marvdasht Plain
only from plots large enough to match the spatial resolution of Landsat-8
images. Thus, selected plots in the Marvdasht Plain should be larger than Fig. 6. Comparison of automated and digitized farm boundaries in 60 random
4 ha (Tasumi, 2019). The accuracy of PYSEBAL was assessed by plots in Marvdasht plain.

7
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Table 3
Validation of crop areas classification in Marvdahst plain, 2018
(spring) Precision Recall Overall accuracy (summer) Precision Recall Overall accuracy

Alfalfa 66.67 0.57 80 Corn 0.56 0.63 80

Canola 69.23 0.6 75 Rice 0.73 0.98 73.56
Orchard 93.75 0.94 89.55 Sugar beet 0.84 0.79 84
Wheat 98.1 0.94 84.43 Tomato 0.86 0.97 82.05
Urban 100 1 100 Fallow 0.9 1 90
Bare soil 81.25 0.81 100

Table 4
Comparison of IRPYSEBAL and IRfields from the agricultural season in Marvdasht (2018).
Plot n◦ Crop type X Y Area Frequency Amount m3 IRfields mm/ha IRPYSEBAL mm/ha Over/under-estimation of IR (%)

1 Alfalfa 52.84 29.84 2.443 15 95 1425 1343 − 5.92

2 Corn 52.80 29.93 2.294 6 86.4 518 473 − 9.08
3 Wheat 52.79 29.93 1.418 4 86.4 346 268 − 25.4
4 Wheat 52.84 29.84 1.395 4 90 360 353 − 1.96
5 Wheat 52.84 29.84 3.513 4 96 384 459 17.70

in Table 5. However, despite the lack of expensive and scarce lysimetric

Table 5
data in the field, the relatively low values show the relative accuracy of
Crop coefficients retrieved from the literature for FAO-56 method and Daily ET
PYSEBAL in its estimate of Daily ET in 2018 in the Marvdasht Plain.
from FAO-56 and PYSEBAL in Marvdasht plain in 2018.
Crops Kcin Kcmid Kcend Length Plots FAO- PYSEBAL
(days) over 4 56 Daily ET 3.3. Water balance of Marvdasht plain
ha Daily
ET
The results of the crop classification provide an agricultural census of
Alfalfa 0.4 0.95 0.9 60 35 1.72 1.33 Marvdasht plain in 2018. Table 6 shows the areas per crop type and
Canola 0.35 1.15 0.35 175 12 3.02 2.99 Figs. 8 and 9 locate each plot per crop type. In spring, over 32250 ha was
Orchard 0.4 1.1 0.45 150 57 3.43 3.61
cultivated, mostly wheat (17811 ha, 50.5% of the plain, Fig. 8), as one of
Wheat 0.3 1.15 0.32 240 917 1.96 1.78
Corn 0.3 1.2 0.75 150 267 5.49 4.17 the key crops for food security and self-sufficiency in Iran. Rice is also a
Rice 1.05 1.2 0.75 150 60 7.13 6.17 key-crop for food security and is intensively cultivated in Marvdasht. It
Sugar 0.35 1.2 0.7 160 122 6.10 5.74 is a traditional crop, especially in the northern part of the plain, but the
beet
construction of a dam in Doroudzan with a maximum capacity of one
Tomato 0.6 1.15 0.8 140 167 4.73 3.94
billion m3 (Figs. 8 and 9) in the 1970s drastically increased the area of
rice, as another key crop of food security in Iran (Moameni, 1999). Rice
cultivation had become almost industrial and remains as one of the most
profitable crops in this region. Summer vegetables are exclusively
composed of tomatoes and sugar beets on mid-areas, 12.8 and 8.6%,
respectively. Overall, the crop choices in Marvdasht are not too diverse
and follow clear trends of food production within a legitimate food se­
curity perspective.
The map (Fig. 10) shows the seasonal spatial distribution of the
PYSEBAL ETseason and Table 7 shows the water balance information for
the Marvdasht Plain. Firstly, it appears that the plain is more intensively
cultivated in spring than in summer due to respective rainfall of 181 mm
and 119 mm. The results of PYSEBAL thus make it possible to calculate
the total irrigation needs of the plain in 2018 (Table 7).
The last column of Table 7 allows to prioritize the crops according to
their pressures on the groundwater resource by dividing IRseason of each
crop with the volume precipitated and available for aquifer recharge on
each surface. It is clear that rice exerts the greatest pressure because of

Fig. 7. Comparaison of Daily ET FAO-56 and Daily ET PYSEBAL in the Marv­ Table 6
dasht plain in 2018. Spring and summer crop areas in Marvdasht plain based on crop classification
with Sentinel-2 images in 2018.
on account on the variety of fruit trees species in that class. The number Crops (spring) Area Area Crops Area Area
of plots compared by crop type varied according to the importance of the (ha) (%) (summer) (ha) (%)
plants. For example, only 12 plots of canola larger than 4 ha were
Wheat 17811 50.5 Corn 7184 22.2
compared as canola is only marginally grown in the plain, compared to Spring 14014 39.8 Rice 5433 16.8
917 plots of wheat, a major crop in the plain. In total, the RMSE between vegetable
ET Daily FAO-56 and PYSEBAL was 0.67 mm/day and the MAE 0.52 Orchard 1818 5.2 Tomato 4140 12.8
Alfalfa 1548 4.4 Sugar beet 2768 8.6
mm/day. One must take into account the bias of the FAO-56 method
Canola 59 0.2 Orchard 1818 5.6
(Allen et al., 1998), which only considers well-irrigated plots with no Total 35250 100 Alfalfa 1548 4.8
water deficit, which is not necessarily the case for all the plots compared Total 32307 100

8
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 8. Crop map in spring 2018.

its total irrigation needs. Indeed, the volume needed to irrigate all rice necessarily leads to a decrease in the piezometric level of the Marvdasht
plots in 2018 was 11.92 times the volume of water available for Plain. For this reason, 60 farmers were asked in the surveys the current
groundwater recharge. Rice IRseason was between 770 and 907 mm depth of their wells as well as the depth estimated some 30 years ago.
depending on the plots with IRseason ranging from 1359 to 1496 mm According to these surveys, the average drawdown of the water tables
according to Brouwer’s Equation (Brouwer et al., 2001). Rice is thus a would have been 125 m over the last thirty years due to the intense use
crop that consumed too much water compared to the renewable water of groundwater. This trend can only be confirmed by the negative water
resource and therefore does not seem to be adapted to the water resource balance of PYSEBAL in 2018 and the intensive cultivation of these plants
of this semi-arid context. On the other hand, all the plants that appear in in the Marvdasht plain every year for the last 50 years or so, as already
red in Table 7 are in this same case of over-consumption of water to shown by Momeni’s report in that study area twenty years ago (Moa­
different degrees, from corn (pressure 5.61 times higher) to tomatoes meni, 1999).
(pressure 1.44 higher).
In addition, these IRseason values obtained with rice are lower than 4. Discussion
those obtained by lysimetric measurements in the Pirmoradian study
(Pirmoradian et al., 2002): 1983 and 2361 mm/season, which confirms 4.1. PYSEBAL ETseason variability
the underestimates of PYSEBAL. These plants, because of their surface
areas and their ETseason and IRseason are too large and the results of Table 7 reports some variability in ETseason and IRseason proportion­
PYSEBAL allow us to identify the crops on which action should be taken ally, based on the net precipitation for each plant. The Hargreaves-
either by reducing the cultivated areas or by improving irrigation Samani Equation at Sad Doroudzan station was used as a reference to
techniques by adopting, for example, the drip or sprinkler systems which calibrate ETinst from PYSEBAL at 11:00 GMT each day took into account
are very little present in the plain according to surveys with farmers. On temperature and solar radiation while the Standardized Penman-
the other hand, only the orchards have a water consumption adapted to Monteith Equation used temperature, wind speed, radiation and rela­
the water available for recharging (pressure less than 1 in green in tive humidity. This study therefore compared Daily ET from Hargreaves-
Table 7). Orchards benefit indeed from a more modest ETseason ranging Samani with ETr24 PYSEBAL on a grassy (assumed to be well watered)
from 443 to 869 but especially from a long rainy season (332 mm) as it is plot of 11 ha next to Sad Doroudzan station (Fig. 11). This comparison
an annual crop. In theory, all the plants cultivated in the plain should first explains the heterogeneity of PYSEBAL’s results over the whole
have a IRseason that is lower than the volume available for recharge. This plain as PYSEBAL takes into account the calibration between two
negative water balance leads to a total groundwater consumption of different Equations in the calculation of ETseason: the reference Equation
0.2859 km3 for a precipitation volume available for recharge of only at the station and the reference Equation calculated by PYSEBAL.
0.098 km3. If such water use is repeated every year, this water balance Indeed, the comparison of daily ET results shows some differences at

9
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 9. Crop map in summer 2018.

the scale of the hydrological year. The RMSE is 0.63 mm/day and the paper compared these parameters between the recorded data (regarding
MAE is 0.46 mm/day. The difference between the two ET is thus slight temperature, Air Temperature of the station at 2 m as compared to Land
and ETr24 estimated by PYSEBAL remains very close to the ETHargeaves surface Temprature of PYSEBAL) at Sad Doroudzan with a well-irrigated
observed on the day of the Landsat-8 overpass. Only days 42 (March), alfalfa plot close to the station (over a large area of 11 ha) with an alfalfa
106 (May), 138 (June), 138 (July) and 170 (August) showed an over­ plot south of 8 ha and an alfalfa plot north of 5 ha (Fig. 12).
estimation of ETr24 compared to the ETHargreaves of the station. However, Location near the station (LNS) is 4 km to Sad Doroudzan, location
this slight difference may have consequences on the ETseason of the south 55 km and location north 19 km. Logically, the daily temperatures
Marvasht Plain. Moreover, ETseason of PYSEBAL remains very dependent at LNS are the closest to those recorded at Sad Doroudzan station. On the
on the climatic data recorded at the Sad Doroudzan station. The corre­ other hand, the further away from the station, the more different the
lation was calculated between ETr24 of the nine largest and closest plots temperatures are at location north (RMSE 7.2 K/day compared to
to the Sad Doroudzan station (one sample plot per crop type) and the Temperature at LNS) and south (RMSE 7.7 K/day compared to Tem­
daily climatic parameters used by PYSEBAL at the time of the Landsat-8 perature location near the station), which in this case are higher than the
run: Relative Humidity, Temperature, Wind Speed and Radiation temperatures at LNS. The warmest surface temperatures are also located
(Fig. 12). There is a strong correlation between ETr24 and RH with a at the south location 55 km to the south, so a stronger ETseason can be
minimum R2 of 0.75, and Temperature (R22: 0.87) and Radiation (R2: expected (ETseason LNS 1066 mm, south location: 1518 mm and north
0.86). Only the relationship between Wind Speed and ETr24 seems location: 1315 mm). Not all Surface Temperatures are the same every­
weaker because Wind Speed can be very variable from one area to where, including in the PYSEBAL results, and ETseason will depend on
another and it seems that PYSEBAL minimizes the weight of WS in its this spatial distribution. On the other hand, the Radiation layer of
estimate of ETr24. Recall here that these correlations are such that they PYSEBAL seems to be more homogeneous than the Surface Temperature
are for the plots closest to the reference station used in our case study. layer (Fig. 14). Although ETr24 was correlated with Radiation (Fig. 13),
The variability of the ETseason by crop type on the plots closest to the it would appear that the Radiation does not vary much in the plain.
station because the climatic data are homogeneous up to a certain However, the PYSEBAL Radiation of near, south and north locations
radius. Indeed, the climatic conditions are not exactly the same between records a high RMSE of 54–57 MJ/m2/day which may limit the reli­
the north, the centre and the south of the plain. Logically, Relative ability of this layer at the scale of a whole plain, which is essential for the
Humidity is lower in the drier areas where Temperatures and Radiation ETseason estimate.
are higher. The four climatic parameters were not available at other In the PYSEBAL results, Daily ETr24 from the three alfalfa plots
stations in the Marvdasht Plain. Nevertheless, the PYSEBAL outputs on correlated well with both Surface Temperatures with a minimum R2 of
Daily Radiation from each satellite image as well as Daily Surface 0.57 and solar radiation with a minimum R2 of 0.68 (Fig. 15). This
Temperature were compared with three different locations. Indeed, this confirms the importance of climatic parameters in the variability of

10
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 10. Maps of annual, spring and summer ETseason based on PYSEBAL in Marvdasht plain in 2018.

ETseason but also the fact that the variation comes more from the spatial 4.2. IRseason variability: groundwater economy
distribution of Land Surface Temperature than from Radiation with
which ETr24. Indeed, if the whole plain was characterised by the same The Marvdasht Plain is also characterised by IRseason variabilities.
Surface Temperature on each Landsat-8 image for each day of passage, Although the soils in the region are predominantly fine-textured alluvial
ETseason would probably record much smaller variabilities. This is a soils, the soil characteristics are important to consider. For example, the
strength of the PYSEBAL model, whose spatial variations reflect plots at the foot of the mountains are on deeper, sandier soils on which
different evapotranspiration realities over an entire plain. farmers will prefer to grow orchards rather than rice. Indeed, a rice plot
The predominance of climatic parameters is to be questioned in this on sandy soil would lead to higher percolation values than those ob­
variability but Table 7 on the water balance of the Marvdasht plain also tained by lysimeters in the centre of the plain in previous studies (Pir­
showed the variability of the water used by the farmers, i.e. IRseason. moradian et al., 2002). These values could thus reach 8 mm/day
Although IRseason is proportional to the ETseason in terms of available net according to Brouwer (Brouwer et al., 2001), which would considerably
precipitation, it would appear that there are other factors responsible for increase the total percolation of the rice plot. However, beyond the soil
this variability. conditions of the study area, farmers practices should be questioned,
especially irrigation schedules and irrigation systems (Calera et al.,
2017; Costa et al., 2019; Hess et al., 2016; Zwart and Bastiaanssen,
2007). For this purpose, the irrigation systems of 60 farmers were

11
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Table 7
Water balance from ETseason and IRseason of the cultivated crops in the Marvdasht plain 2018.

*Net rainfall is the amount of net precipitation that occurred during the season of the matching crop.
Total IR is the product of multiplying the average IRseason with the cultivated area.
Total available precipitations is the volume of precipitations that occurred during the season and available for aquifer recharge.
Pressure on groundwater resource is the number of times the IRseason exceeded the matching Netrainfall. This highlights the sensitive crops regarding
water in this semi-arid plain.

positive but this would imply drastic reductions in cultivated areas in

addition to the adoption of lower IRseason. However, this option does not
seem to be economically feasible for farmers. However, even in this
scenario, for all crops except orchards, the road to water savings is still
long as IRseason remains higher than the volume available for aquifer
recharge. This theoretical scenario is a true illustration of the usefulness
of PYSEBAL in a water-saving objective, without reducing the cultivated
areas. These results make it possible to set the objectives necessary to
reduce water use in the Marvdasht plain.

5. Conclusion

This study showed the contributions of remote sensing to the esti­

mation of the water balance of an important agricultural plain such as
Marvdasht in southern Iran. This free technology requires some field­
work, particularly for mapping crops from one season to the next. The
fieldwork and NDVI classification mapping of each plot are largely
Fig. 11. Comparison of Daily ETHargreaves at Sad Doroudzan station and dependent on agricultural calendars, but at the end of the agricultural
ETr24 PYSEBAL. year this paper showed that it was possible to map the plots quickly and
with high accuracy. The second step, ETseason estimation, can also only
collected in the surveys. Of the 60 farmers surveyed, 52 use the furrow be done at the end of the crop year. PYSEBAL proved to be a relevant tool
system which is more sensitive to evaporation. The variability of IRseason in the estimation of ETseason and IRseason. The validation of PYSEBAL
can therefore be explained by this practice. Indeed, it would seem that with the field (comparison of IRPYSEBAL and IRFIELD) and with Daily ET
the distribution of net rainfall varies very little between the north and FAO-56 was the biggest challenge of this study. PYSEBAL seems to un­
the south of the plain. For example, the precipitation for the Januar­ derestimates the reality. Indeed, the lack of lysimetric data can limit the
y–June period is 287 mm in the north of the plain and 272 mm in the reliability of such models, but PYSEBAL seems robust in the case of
south. Thus, wheat plots are found with low IRseason of 227 mm/season Marvdasht as the differences between IRPYSEBAL and IRfield, and between
in both the north and the south and IRseason of 466 mm/season in both Daily ET PYSEBAL et Daily ET FAO-56 remains low, and this study
the north and the south. This means that some farmers irrigate more allowed to understand that the variability of ETseason came primarily
than their neighbours. This finding should put us on the track of the from the variability of the Land Surface Temperature layer generated by
groundwater economy. Indeed, it would seem that farmers who irrigate PYSEBAL than from the Radiation Layer. Indeed, climatic conditions are
more on their plots could apply smaller amounts of water, closer to those not exactly homogeneous in all parts of an agricultural plain, which
of their neighbours in the lower IRseason. This shift from high IRseason to explains the heterogeneity of ETseason results. As such, the Radiation
low IRseason can only be achieved by improving irrigation, starting with Layer, due to its homogeneity, does not seem to reflect this climatic
the amounts applied. Thus, this study has estimated the groundwater diversity as much as the Land Surface Temperature layer.
savings that could be achieved in the plain if all farmers irrigated, Thanks to this methodological combination, it was possible to
theoretically, with the lowest IRseason found in the plain (Fig. 16). characterize crops according to their water requirements. Crops such as
In this way, the pressure on water resources would be more modest rice or maize have too high consumption in relation to the volume
than in the current context of the year 2018. Such a change would result available for groundwater recharge. In this case, most of the crops had a
in a total groundwater use of 0.2208 km3 instead of the current 0.2859 negative water balance, which in 2018 led to the overutilisation of
km3. In absolute terms, the water balance would have to become groundwater in the plain. Such practices can only lead to groundwater

12
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 12. Strong positive correlation between T (◦ C), RH (%), R (MJ/m2) and ETr24 PYSEBAL.

Fig. 13. Comparison of 4 locations temperatures between observed air tem­

perature data at Sad Doroudzan station and land surface temperature of 3 Fig. 14. Comparison of 4 locations temperatures between observed daily ra­
different locations in the plain. diation data at Sad Doroudzan station and land surface temperature of 3
different locations in the plain.

13
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

Fig. 15. Strong positive correlation between ETr24 and PYSEBAL Land Surface Temperature and Radiation.

Fig. 16. Possible groundwater economy in the Marvdasht plain in 2018.

drawdown, but the advantage of this study is that it targeted the sen­ Declaration of competing interest
sitive crops least adapted to the semi-arid context. In this respect, the
results of this study are intended to be used by political decision-makers The authors declare that they have no known competing financial
in the field of agriculture and water and are intended to be reproducible interests or personal relationships that could have appeared to influence
every year on any plain in the world. The variability of the IRseason the work reported in this paper.
showed us that some farmers were able to irrigate less than their
neighbours despite similar climatic conditions. It would seem that it is References
above all on practices that agricultural policies should act, in order to
minimize the use of groundwater by tending towards these lower vol­ Ahani, H., Kherad, M., Kousari, M.R., Rezaeian-Zadeh, M., Karampour, M.A., Ejraee, F.,
umes already applied by some farmers. Kamali, S., 2012. An investigation of trends in precipitation volume for the last three
decades in different regions of Fars province, Iran. Theor. Appl. Climatol. 109,
361–382.
CRediT authorship contribution statement Allen, G.R., Pereira, L.S., Raes, D., Smith, M., 1998. Crop Evapotranspiration –
Guidelines for Computing Crop Water Requirements. FAO, Rome, Italy.
Allen, R., Tasumi, M., Trezza, R., 2007. Satellite-based energy balance for mapping
Arnaud Caiserman: Conceptualization, Methodology, Software, evapotranspiration with internalized calibration (METRIC)—Model. J. Irrigat. Drain.
Validation, Formal analysis, Investigation, Resources, Data curation, Eng. 133, 380–394. https://doi.org/10.1061/(ASCE)0733-9437 (2007)133:4(380).
Writing - original draft, Writing - review & editing. Farshad Amir­ Amiraslani, F., Caiserman, A., 2018. Multi-stakeholder and multi-level interventions to
tackle climate change and land degradation: the case of Iran. Sustainability 10.
aslani: Writing - review & editing, Visualization. Dominique Dumas: https://doi.org/10.3390/su10062000, 2000.
Supervision, Project administration, Funding acquisition. Asgarian, A., Soffianian, A., Pourmanafi, S., 2016. Crop type mapping in a highly
fragmented and heterogeneous agricultural landscape: a case of central Iran using

14
A. Caiserman et al. Journal of Arid Environments 188 (2021) 104461

multi-temporal Landsat 8 imagery. Comput. Electron. Agric. 127, 531–540. https:// Liou, Y.-A., Kar, S.K., 2014. Evapotranspiration estimation with remote sensing and
doi.org/10.1016/j.compag.2016.07.019. various surface energy balance algorithms—a review. Energies 7, 2821–2849.
Attarod, P., Rostami, F., Dolatshahi, A., Sadeghi, S.M.M., Amiri, G.Z., Bayramzadeh, V., https://doi.org/10.3390/en7052821.
2016. Do changes in meteorological parameters and evapotranspiration affect Madani, K., 2014. Water management in Iran: what is causing the looming crisis?
declining oak forests of Iran? J. For. Sci. 62, 553–561. https://doi.org/10.17221/83/ J. Environ. Soc. Sci. 4, 315–328. https://doi.org/10.1007/s13412-014-0182-z.
2016-JFS, 2016. Madani, K., AghaKouchak, A., Mirchi, A., 2016. Iran’s socio-economic drought:
Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A., Holtslag, A.A.M., 1998a. A remote challenges of a water-bankrupt nation. Iran. Stud. 49, 997–1016. https://doi.org/
sensing surface energy balance algorithm for land (SEBAL). 1. Formulation. 10.1080/00210862.2016.1259286.
J. Hydrol. 212 (213), 198–212. https://doi.org/10.1016/S0022-1694(98)00253-4. Menenti, M., Choudhury, B.J., 1993. Parameterization of land surface evaporation by
Bastiaanssen, W.G.M., Pelgrum, H., Wang, J., Ma, Y., Moreno, J.F., Roerink, G.J., van der means of location dependent potential evaporation and surface temperature range.
Wal, T., 1998b. A remote sensing surface energy balance algorithm for land Dep. Environ. Food Rural Aff. (Defra) 212, 561–568.
(SEBAL).: Part 2: Validation. J. Hydrol. 212 (213), 213–229. https://doi.org/ Moameni, A., 1999. Soil Quality Changes under Long - Term Wheat Cultivation in the
10.1016/S0022-1694(98)00254-6. Marvdasht Plain. Ghent University, South - Central Iran (Ghent, Belgium).
Belgiu, M., Csillik, O., 2018. Sentinel-2 cropland mapping using pixel-based and object- Motamed, M., 2017. Developments in Iran’s Agriculture Sector and Prospects for U.S.
based time-weighted dynamic time warping analysis. Remote Sens. Environ. 204, Trade. United States Depar Tment of Agriculture. Washington D.C.
509–523. https://doi.org/10.1016/j.rse.2017.10.005. Neuve-Eglise, A., 2007. L’agriculture Iranienne : Une Modernisation Inachevée. La Revue
Brouwer, C., Prins, K., Kay, M., Heibloem, M., 2001. Irrigation Water Management: de Téhéran.
Irrigation Methods. FAO, Wageningen, Netherlands. Norman, J.M., Becker, F., 1995. Terminology in thermal infrared remote sensing of
Caiserman, A., Dumas, D., Bennafla, K., Faour, G., Amiraslani, F., 2019. Application of natural surfaces. Rem. Sens. Rev. 12, 159–173. https://doi.org/10.1080/
remotely sensed imagery and socioeconomic surveys to map crop choices in the 02757259509532284.
bekaa valley (Lebanon). Agriculture 9, 57. https://doi.org/10.3390/ Panigrahy, S., Sharma, S.A., 1997. Mapping of crop rotation using multidate Indian
agriculture9030057. Remote Sensing Satellite digital data. ISPRS J. Photogrammetry Remote Sens. 52,
Calera, A., Campos, I., Osann, A., D’Urso, G., Menenti, M., 2017. Remote sensing for crop 85–91. https://doi.org/10.1016/S0924-2716(97)83003-1.
water management: from ET modelling to services for the end users. Sensors (Basel) Pirmoradian, N., Kamgar-Haghighi, A.A., Sepaskhah, A.R., 2002. Crop coefficient and
17. https://doi.org/10.3390/s17051104. water requirement of rice in kooshkak area, Fars province. J. Sci. Technol. Agric.
Costa, J. de O., Coelho, R.D., Wolff, W., José, J.V., Folegatti, M.V., Ferraz, S.F. de B., Nat. Resourc. 1, 10–22.
Costa, J. de O., Coelho, R.D., Wolff, W., José, J.V., Folegatti, M.V., Ferraz, S.F. de B., Roerink, G.J., Su, Z., Menenti, M., 2000. S-SEBI: a simple remote sensing algorithm to
2019. Spatial variability of coffee plant water consumption based on the SEBAL estimate the surface energy balance. Phys. Chem. Earth - Part B Hydrol., Oceans
algorithm. Sci. Agric. 76, 93–101. https://doi.org/10.1590/1678-992x-2017-0158. Atmos. 25, 147–157. https://doi.org/10.1016/S1464-1909(99)00128-8.
Ewaid, S.H., Abed, S.A., Al-Ansari, N., 2019. Crop water requirements and irrigation Roshan, G., Negahban, S., 2015. Modeling of the effects of climate change on rainy and
schedules for some major crops in southern Iraq. Water 11, 756. https://doi.org/ gully erosion potential of Kor-chamriz watershed in Fars province. Model. Earth Syst.
10.3390/w11040756. Environ. 1, 26. https://doi.org/10.1007/s40808-015-0031-4.
FAO, 1986. Chapter 4: irrigation water needs [WWW Document]. URL. http://www.fao. Saatsaz, M., 2019. A historical investigation on water resources management in Iran.
org/docrep/S2022E/s2022e08.htm. Environ. Dev. Sustain. https://doi.org/10.1007/s10668-018-00307-y.
Faramarzi, M., 2010. Assessment of Regional Water Endowments, Crop Water Simonneaux, V., Lepage, M., Helson, D., Métral, J., Thomas, S., Duchemin, B.,
Productivity, and Implications for Intra-country Virtual Water Trade in Iran. Cherkaoui, M., Kharrou, H., Berjami, B., Chebhouni, A., 2009. Estimation spatialisée
University of Zurich, Zurich, Switzerland. de l’évapotranspiration des cultures irriguées par télédétection : application à la
Fooladmand, H.R., Haghighat, M., 2007. Spatial and temporal calibration of Hargreaves gestion de l’irrigation dans la plaine du Haouz (Marrakech, Maroc). Secheresse 20,
equation for calculating monthly ETo based on Penman-Monteith method. Irrigat. 123–130. https://doi.org/10.1684/sec.2009.0177.
Drain. 56, 439–449. https://doi.org/10.1002/ird.305. Soltani, M., Laux, P., Kunstmann, H., Stan, K., Sohrabi, M.M., Molanejad, M.,
Fooladmand, H.R., Zandilak, H., Ravanan, M.H., 2008. Comparison of different types of Sabziparvar, A.A., Ranjbar SaadatAbadi, A., Ranjbar, F., Rousta, I., Zawar-Reza, P.,
Hargreaves equation for estimating monthly evapotranspiration in the south of Iran. Khoshakhlagh, F., Soltanzadeh, I., Babu, C.A., Azizi, G.H., Martin, M.V., 2016.
Arch. Agron Soil Sci. 54, 321–330. https://doi.org/10.1080/03650340701793603. Assessment of climate variations in temperature and precipitation extreme events
Golian, S., Mazdiyasni, O., AghaKouchak, A., 2015. Trends in meteorological and over Iran. Theor. Appl. Climatol. 126, 775–795. https://doi.org/10.1007/s00704-
agricultural droughts in Iran. Theor. Appl. Climatol. 119, 679–688. https://doi.org/ 015-1609-5.
10.1007/s00704-014-1139-6. Song, Q., Hu, Q., Zhou, Q., Hovis, C., Xiang, M., Tang, H., Wu, W., 2017. In-season crop
Hao, P., Tang, H., Chen, Z., Liu, Z., 2018. Early-season crop mapping using improved mapping with GF-1/WFV data by combining object-based image analysis and
artificial immune network (IAIN) and Sentinel data. PeerJ 6. https://doi.org/ random forest. Rem. Sens. 9, 1184. https://doi.org/10.3390/rs9111184.
10.7717/peerj.5431. Stancalie, G., Marica, A., Toulios, L., 2010. Using earth observation data and CROPWAT
Hassanshahi, H., Iravani, H., Daneshvar Ameri, Z., Kalantari, K., 2015. Measure and model to estimate the actual crop evapotranspiration. Phys. Chem. Earth, Parts A/B/
comparison of economic, social and ecological sustainability of farming systems in C, Bio-, Agro, Urban Climatol. 35, 25–30. https://doi.org/10.1016/j.
the Marvdasht plain. Desert 20, 231–239. https://doi.org/10.22059/ pce.2010.03.013.
jdesert.2015.56485. Su, Z., 2002. The Surface Energy Balance System (SEBS) for estimation of turbulent heat
Hess, T., Daccache, A., Daneshkhah, A., Knox, J., 2016. Scale impacts on spatial fluxes. Hydrol. Earth Syst. Sci. 6, 85–100. https://doi.org/10.5194/hess-6-85-2002.
variability in reference evapotranspiration. Hydrol. Sci. J. 61, 601–609. https://doi. Tabari, H., Abghari, H., Talaee, P.H., 2012. Temporal trends and spatial characteristics of
org/10.1080/02626667.2015.1083105. drought and rainfall in arid and semiarid regions of Iran. Hydrol. Process. 26,
Hessel, T., 2019. Add SEBAL. Contribute to Wateraccounting/SEBAL Development by 3351–3361. https://doi.org/10.1002/hyp.8460.
Creating an Account on GitHub. Water Accounting. Tasumi, M., 2019. Estimating evapotranspiration using METRIC model and Landsat data
Hessel, T., van Opstal, J., Trambauer, P., Mohamed, Y., Bastiaanssen, W., 2017. User for better understandings of regional hydrology in the western Urmia Lake Basin.
Guide: pySEBAL for Landsat Imagery (Beta Version). UNESCO - IHE, Delft, the Agric. Water Manag. 226 https://doi.org/10.1016/j.agwat.2019.105805, 105805.
Netherlands. Valipour, M., Eslamian, S., 2014. Analysis of potential evapotranspiration using 11
Heupel, K., Spengler, D., Itzerott, S., 2018. A progressive crop-type classification using modified temperature-based models. Int. J. Hortic. Sci. Technol. 4, 192–207. https://
multitemporal remote sensing data and phenological information. Photogramm. doi.org/10.1504/IJHST.2014.067733.
Fernerkund. GeoInf. 86, 53–69. https://doi.org/10.1007/s41064-018-0050-7. Waldhoff, G., Lussem, U., Bareth, G., 2017. Multi-Data Approach for remote sensing-
Islamic Republic of Iran Meteorological Organisation, 2018. Climate Data from Sad based regional crop rotation mapping: a case study for the Rur catchment, Germany.
Doroudzan Station. Tehran, Iran. Int. J. Appl. Earth Obs. Geoinf. 61, 55–69. https://doi.org/10.1016/j.
Karimi, V., Karami, E., Keshavarz, M., 2018. Climate change and agriculture: impacts and jag.2017.04.009.
adaptive responses in Iran. J. Integr. Agric. 17, 1–15. https://doi.org/10.1016/ Water Watch, 2019. WaterWatch remote sensing services. Validation [WWW Document].
S2095-3119(17)61794-5. URL. http://www.waterwatch.nl/tools0/sebal/validation.html.
Kenduiywo, B.K., Bargiel, D., Soergel, U., 2018. Crop-type mapping from a sequence of Waters, R., Allen, R., Tasumi, M., Trezza, R., Bastiaanssen, W., 2002. SEBAL: Surface
Sentinel 1 images. Int. J. Rem. Sens. 39, 6383–6404. https://doi.org/10.1080/ Energy Balance Algorithms for Land Idaho Implementation Advanced Training and
01431161.2018.1460503. Users Manual. NASA and University of Idaho, Spokane, United States.
Keshavarz, M., Karami, E., 2013. Institutional adaptation to drought: the case of Fars Wukelic, G.E., Gibbons, D.E., Martucci, L.M., Foote, H.P., 1989. Radiometric calibration
agricultural organization. J. Environ. Manag. 127, 61–68. https://doi.org/10.1016/ of Landsat thematic mapper thermal band. Remote Sens. Environ. 28, 339–347.
j.jenvman.2013.04.032. https://doi.org/10.1016/0034-4257(89)90125-9.
Keshavarz, M., Karami, E., Zibaei, M., 2014. Adaptation of Iranian farmers to climate Xie, H., Tian, Y.Q., Granillo, J.A., Keller, G.R., 2007. Suitable remote sensing method and
variability and change. Reg. Environ. Change 14, 1163–1174. https://doi.org/ data for mapping and measuring active crop fields. Int. J. Rem. Sens. 28, 395–411.
10.1007/s10113-013-0558-8. https://doi.org/10.1080/01431160600702673.
Khosravi, H., Sajedi Hosseini, F., Nasrollahi, M., Gharechaee, H.R., 2017. Trend analysis Zehtabian, G., Khosravi, H., Ghodsi, M., 2010. High Demand in a Land of Water Scarcity:
and detection of precipitation fluctuations in arid and semi-arid regions. Desert 22, Iran, in: Water and Sustainability in Arid Regions. Springer, Dordrecht, pp. 75–86.
77–84. https://doi.org/10.22059/jdesert.2017.62173. https://doi.org/10.1007/978-90-481-2776-4_5.
Lamb, D.W., Brown, R.B., 2001. PA—precision agriculture: remote-sensing and mapping Zhong, L., 2012. Efficient Crop Type Mapping Based on Remote Sensing in the Central
of weeds in crops. J. Agric. Eng. Res. 78, 117–125. https://doi.org/10.1006/ Valley. California, UC Berkeley.
jaer.2000.0630. Zwart, S.J., Bastiaanssen, W.G.M., 2007. SEBAL for detecting spatial variation of water
productivity and scope for improvement in eight irrigated wheat systems. Agric.
Water Manag. 89, 287–296. https://doi.org/10.1016/j.agwat.2007.02.002.

15