Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Agricultural Water Management 252 (2021) 106880

Contents lists available at ScienceDirect

Agricultural Water Management


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

Parameterization of the AquaCrop model for cowpea and assessing the


impact of sowing dates normally used on yield
H.G.G.C. Nunes a, *, V.D.S. Farias a, D.P. Sousa a, D.L.P. Costa a, J.V.N. Pinto a, V.B. Moura a, E.
O. Teixeira a, M.J.A. Lima a, S. Ortega-Farias b, P.J.O.P. Souza a
a
ISPAAm-UFRA, Research Group Soil-Plant-Atmosphere Interaction in the Amazon, Federal Rural University of the Amazon, Perimetral Avenue, 2501, 66077-830
Belém, Brazil
b
CITRA, Research and Extension Center for Irrigation and Agroclimatology and Research Program on Adaptation of Agriculture to Climate Change (A2C2), Universidad
de Talca, Chile

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

Handling Editor - Dr Z Xiying The AquaCrop model was evaluated considering a data set from a cowpea (BR3-Tracuateua cultivar) cultivation
in Castanhal/Brazil under different irrigations during the reproductive phase, which is sensitive to water deficit.
Keywords: Data on leaf area index (LAI), soil water content (SWC), biomass, and final yield of two harvests were used for
Agricultural modeling model calibration (2013) and validation (2014). LAI data were used for parameterization of canopy cover (CC),
Vigna unguicula L. Walp
and some crop parameters obtained in the field. The model was also used to evaluate the impacts of sowing dates,
Drip irrigation
as well as the strategy of using an economical irrigation threshold (60% of Readily Available Water) on the water
Water-use efficiency
Crop yield use and crop yield. Regarding the results after parameterization of the canopy cover curve, the model efficiency
index (EF) presented a satisfactory performance (0.98), which is due to the high correlation coefficient (R2 =
0.99) between the simulated and observed values and the low values of normalized root mean square error
(NRMSE < 6.33%). For model performance in simulating SWC, there was high efficiency (EF > 0.70) for all
treatments, with low estimated errors and NRMSE representing 8.12% of mean SWC observations during vali­
dation, indicating high precision for simulation of biomass and cowpea yield. The model performance index
ranged from 0.92 to 0.94 for biomass and from 0.92 to 0.99 for final yield. These results showed that AquaCrop is
suitable for simulating biomass production and yield of cowpea. Additionally, sowing in early April is appro­
priate for better water use and productivity and the tested irrigation threshold can be indicated to significantly
improve final crop yields.

1. Introduction However, cowpea still has relatively low average yield in the region
(821 kg ha− 1) (Conab, 2018), generally due to crop mismanagement and
The cowpea (Vigna unguiculata L. Walp) is a fast-growing leguminous water deficit, especially during the reproductive phase. It is because
species that produces pod-type fruits. This plant plays an important seeds are formed at this stage, and when cowpea is under water deficit,
socioeconomic role as one of the crops most cultivated by family farmers there is a decrease in biomass and stomatal conductance that leads to a
in underdeveloped regions (Farias et al., 2017). In Brazil, it is cultivated relative reduction in grain production (Souza et al., 2017).
especially in the north and northeast regions, where it is a staple food for In order to supply the amount of water required by the cowpea crop
millions of families. In addition, it is a relevant source of employment in the region, irrigation has been applied especially over the reproduc­
and income (Lima Filho et al., 2013a, Lima et al., 2016). tive phase and in the least rainy season. In this way, water deficit im­
The Northeastern Pará state is the main production hub in the pacts can be avoided, and both the quality and quantity of the cowpea
Northern Brazil (Souza et al., 2017). In this region, planting is commonly yield can be positively influenced (Louzada and Jobim, 2011; Lima
performed at the end of the rainy season (April), aiming at reducing et al., 2016; Farias et al., 2017). Moreover, other sowing dates can be
precipitation-related risks to the harvests; therefore, production is proposed for cowpea cultivation in the region.
limited to only one harvest (Conab, 2018, Nunes et al., 2019). Tools capable of assisting the proper crop management and

* Corresponding author.
E-mail address: garibalde13@gmail.com (H.G.G.C. Nunes).

https://doi.org/10.1016/j.agwat.2021.106880
Received 16 April 2020; Received in revised form 9 March 2021; Accepted 20 March 2021
Available online 26 March 2021
0378-3774/© 2021 Elsevier B.V. All rights reserved.
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

simulating biomass and yields have been presented as important stra­ Table 1
tegies for planning and reducing risks in agriculture, as well as assisting Crop coefficient used in experiment.
in decision making. Agrometeorological or dynamic crop models, also Phenological stages Kc
known as agricultural forecasting models, are good examples of such
Initial 0.8
tools (Martins et al., 2014; Lima et al., 2016). These models are used in Growth 0.8 a 1.1
studies on climatic risks (Moradi et al., 2013). In addition, they are also Reproductive 1.1 a 1.4
used to obtain the best yield considering different factors, such as the Final 1.4 a 0.3
number of individuals at sowing and different fertilizations, irrigation
Initial and growth stages are considered the vegetative
systems and water depths (Barterng et al., 2010; Lima Filho et al.,
phase.
2013b).
Some models were applied in cowpea yield simulation, such as
(temperature, RH, U2, global radiation (GR, W m− 2), PRP and volu­
BEANGRO (Hoogenboom et al., 1992) and CROPGRO (Oliveira et al.,
metric soil water content in the 0.30 m depth (SWC, cm3 cm− 3). There
2012; Lima Filho et al., 2013a, 2013b). However, the FAO’s model for
were sensors attached to a CR10X datalogger and an AM16/32 multi­
agricultural growth and production, AquaCrop (Raes et al., 2012; Par­
plexer (Campbell Scientific, Inc.) that recorded and stored the averages
edes and Torres, 2016), currently used for several crops, has been hardly
of these variables every 10 min. The PRP was measured in the experi­
tested on cowpea (Nunes et al., 2019). Since the parameterization of
mental area to calculate the net water depth that should be applied
AquaCrop rarely considers field-based experimental data using irriga­
according to the daily crop evapotranspiration (ETc, mm d− 1) of the
tion under different water depths, this study evaluated the AquaCrop
previous day, as proposed by Farias et al. (2017). This value was sub­
model’s ability to simulate biomass and yield of cowpea in the north­
tracted from the full irrigation to be applied on a particular day of the
eastern Pará State.
experiment. For the calculation of ETc during the experiment, the single
The experiment was carried out in the municipality of Castanhal,
crop coefficient (Kc) suggested by Bastos et al. (2008) was used
assessing different water depths over the reproductive phase. The effects
(Table 1).
on yield and soil water dynamics were evaluated, and this information
The area was cultivated with the BR3-Tracuateua cultivar, which has
was compared to the data collected in the field. The objectives of this
prostrate and indeterminate growth, whose cycle lasts between 60 and
study were to use the field data of cowpea from two harvests to: (1)
70 days. Freire Filho et al. (2009) reported that this cultivar is the most
parameterize the AquaCrop model for simulation of canopy cover and
recommended for the region and the one most used by farmers in the
soil water; (2) evaluate the accuracy of the model for biomass and yield
state, because it is moderately tolerant to high temperatures and to the
predictions under different irrigations during the reproductive phase to
local climate (Souza et al., 2017). Sowing was performed mechanically
simulate the yields based on the sowing dates used by farmers; (3)
with a spacing of 0.50 m between the planting rows and 10 plants per
evaluate supplementary irrigation strategies for increasing crop pro­
linear meter (total of 200,000 plants per ha), on October 1, 2013 and
duction and water use efficiency.
September 9, 2014, during the dry season in the region. The harvest was
carried out on December 6 and November 11, in 2013 and 2014,
2. Material and methods
respectively.
The soil of the experimental site was classified as Dystrophic Yellow
2.1. Experimental area and crop management
Latosol (Lima et al., 2016) with sandy loam texture (Ramos et al., 2016).
The texture and other physical properties of the soil were determined in
The field experiment was carried out at an Experimental Farm
laboratories from the Brazilian Agricultural Research Corporation
belonging to the Federal Rural University of the Amazon (UFRA),
(EMBRAPA, Brazil). For this purpose, 15 undisturbed soil samples were
located in the municipality of Castanhal (1.32◦ S, 47.96◦ W, and 41 m
collected using the Richards’ pressure chamber to determine the per­
above sea level), Northeastern Pará State, during 2013 and 2014. Ac­
manent wilting point (θWP, cm3 cm− 3). A tension table was used to
cording to the Koppen climatic classification provided by Alvares et al.
measure saturation (θSAT, cm3 cm− 3) and field capacity (θFC, cm3 cm− 3)
(2013), the region climate is characterized as Am,1 with an average
(Table 2). For calculating θWP, a pressure value of 1500 kPa was applied.
annual air temperature of 26 ◦ C and maximum and minimum air tem­
Regarding θSAT and θFC, pressure values of 0 and 10 kPa were applied,
peratures corresponding to 28 and 22 ◦ C, respectively. The average
respectively. In these methods, soil samples lose water when the po­
annual precipitation in the study period was 1781.60 mm, with a rainy
tential is higher than the retention forces; therefore, the sample moisture
season from January to May (70% of total precipitation) and a dry
is obtained by gravimetry. Each potential (until equilibrium is reached)
season from August to December (16% of total precipitation) (Farias
applied to the samples is evaluated through successive weighings or
et al., 2017).
until there is no more flow (Teixeira and Behring, 2017).
Daily meteorological data (2003–2014) were collected from an
Saturated hydraulic conductivity (KSAT, cm d− 1) values were deter­
automatic meteorological station on a grassy surface, instrumented
mined by pedotransfer functions based on organic matter, bulk density,
approximately 3 km from the experimental field, belonging to the Na­
and texture data of soil layers (Saxton and Rawls, 2006; Ramos et al.,
tional Institute of Meteorology (INMET, Brazil). The model used data on
2014). Available soil water content (AWC, mm) was also calculated
average, maximum, and minimum temperatures (Tmed, Tmax and
using the soil physical and hydrological properties and effective root
Tmin, ◦ C), relative humidity (RH, %), wind speed at 2 m (U2, m s− 1),
zone (Zr, m), which was 0.30 m in the present study. Souza et al. (2017),
solar radiation (SR, W m− 2 or MJ m− 2 day− 1) and precipitation (PRP,
in a study in the same area, found cowpea effective root of 0.25 m in
mm). The daily reference evapotranspiration (ET0, mm day− 1) was
length, that is, 0.05 m less than the minimum value allowed by the
calculated using the methodology proposed by Allen et al. (1998).
model and the value used in this study. After parameterization, the
During the experiments, an automatic meteorological station was
model simulation of soil water content (SWC) was evaluated. Therefore,
installed to monitor the meteorological variables in the field
the simulated values were compared to those observed in the field.
Fertilization was carried out according to methodology proposed by
EMBRAPA for cultivation of cowpea, based on the information provided
1
Tropical Humid or Tropical Monsoon climate are regions characterized by by chemical analysis of the soil samples collected in the study area
annual precipitation greater than 1500 mm, with the driest month showing (Farias et al., 2017). It was divided and performed before and 30 days
average precipitation below 60 mm. They are also mega thermic regions, whose after sowing (DAS) using N-P-K chemical fertilizers with formulations of
average temperatures in the coldest month is always above 18 ◦ C (Alvares et al.,
0–60–45 in the first year and 0–40–45 in the second year (115 and 85 kg
2013).

2
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Table 2
Textural and hydraulic soil properties in the experimental field.
Layer (m) Sand (%) Silt (%) Clay (%) θSAT θFC θWP KSAT (cm d− 1)

Experimental farm 0.0–0.20 80 12 8 0.29 0.22 0.07 153


0.20–0.40 71 13 16 0.31 0.24 0.08 61
0.40–0.60 71 17 12 0.26 0.21 0.08 88
0.60–0.80 72 12 16 0.27 0.21 0.06 62
0.80–1.00 71 15 14 0.27 0.23 0.08 72

θSAT is volumetric water content at saturation; θFC is volumetric water content at field capacity; θWP is volumetric water content at wilting point.

Table 3
Irrigation (I) applied in the treatments and precipitation (PRP) in the two years of experiment, divided into cowpea phases.
2013 2014

I (mm) PRP (mm) I (mm) PRP (mm)

T1 T2 T3 T4 T1 T2 T3 T4

Phase

Veg 47.76 111.75 75.66 126.00


47.76 47.76 47.76 75.66 75.66 75.66
Rep 96.77 0 153.41 83.04 0 36.82
48.39 24.19 41.52 20.76
Total 144.53 47.76 265.16 158.70 75.66 162.82
96.15 71.95 117.18 96.42

Veg is the vegetative phase and Rep is the reproductive phase.

ha− 1, respectively). During the experiments, management treatments Therefore, irrigation was applied to each treatment again when the
were performed to keep the area free of weeds, diseases and pests. values lost by the crop (ETc) reached the time and quantity that allowed
another application, that is, the daily PRP in the area was always
2.2. Experimental design and treatments considered, subtracted from the full amount of irrigation to be applied.

The experimental design was randomized with six blocks and four 2.3. Irrigation system
treatments, in area of 100 m2. These treatments had different irrigations
applied during the cowpea reproductive phase, which is more sensitive The system used was drip irrigation, with a flow rate of 1.03 l h− 1,
to water deficit (Carvalho et al., 2014; Lima et al., 2016). On the other service pressure of 29.4 kPa, and emitters spaced 0.20 m apart. The drip
hand, over the vegetative phase, all treatments received the same water tapes were made of polyethylene with a diameter of 16 mm and built-in
depth (100% ETc replacement) in order to keep them at field capacity emitters. It is noteworthy that the water depths were controlled in all
(FC). Treatment 1 (T1) consisted in replacing 100% of the water lost by experiments as a function of time, aided by manometers, spigots, and
ETc; in treatment 2 (T2), 50% of the water lost by ETc was replaced; in timers, according to the flow rate of the emitters. Hydraulic evaluations
treatment 3 (T3), 25% of the water lost by ETc was replaced; and in were performed to determine the performance of the system according
treatment 4 (T4, control), there was no water replacement through to the methodology proposed by Bernardo et al. (2006).
irrigation, aiming at representing the scenario found by most farmers in This analysis was performed in all treatments and blocks by col­
the region. lecting containers (1000 ml) positioned at the beginning, middle and
The irrigation depths adopted in 2013 and 2014 were applied daily end of the irrigation lines, for 20 min and with two repetitions. These
and every two days, respectively, aiming to maintain the treatments in values were used to obtain: The Christiansen uniformity coefficient
function of the FC (0.22 cm3 cm− 3), considering water losses through (CUC, %) (96% and 94% in 2013 and 2014, respectively); the distri­
evapotranspiration, and to ensure that the analyzes occurred in function bution uniformity coefficient (DUC, %) (92% in 2013 and 89% in 2014);
of the decrease of this water replacement and not by the deficit arising and Ae (94% in 2013 and 92% in 2014). Despite the decrease in eval­
from several days. These treatments started at 35 DAS and extended to uation values between the years, which is normal in a system used over
58 and 56 DAS in 2013 and 2014, respectively. Irrigation was stopped several years, the system still had acceptable performance in both
after these days, since the beginning of the cowpea maturation stages experimental years (Merriam and Keller, 1978).
was reached.
Total required or full irrigation (CI, mm) was based on the net water 2.4. Crop data
depth to the application efficiency (Ae, %) of the irrigation system in the
different treatments. These values are shown in Table 3, together with Growth was monitored daily from emergence to the end of the
PRP in the years of the experiment and cowpea phases. Although the cowpea cycle. In order to achieve it, the Gepts and Fernández scale
experiment was carried out over the dry season in both years, the (Farias et al., 2017) was used, in which 10 plants were observed in a
occurrence of PRP was noticed during both phases. The vegetative phase linear 1 m line, used exclusively for monitoring purposes. The growth
of 2014 (126.00 mm) had more precipitation than in the previous year measures were collected between 15 and 9 DAS in 2013 and 2014,
(111.75 mm), which assisted to maintain the treatments at FC. However, respectively and weekly, by selecting two 20-meter lines in each treat­
during the reproductive phase, the PRP (153.41 and 36.82 mm in 2013 ment. Five plants per treatment were collected (0.50 m linear) following
and 2014, respectively) disrupted the maintenance of the target treat­ the experimental design adopted for each block (6 replications). The
ments. Overall, it added a greater amount of water in both years (30, 45 organs were separated into stem, petiole, leaf, discs, peduncle, flower,
and 60% more for T2, T3 and T4, respectively), especially in 2013, since pod and grain (when present).
these PRP events occurred after the irrigation in the treatments Yield was computed when 90% of the plants were at maturation
(Table 3). stages (66 and 63 DAS in 2013 and 2014, respectively). Three central

3
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

linear lines with 2 m were selected for yield analysis in each treatment. effects (CC*, %), by the maximum transpiration coefficient (KcTr,x,
Therefore, treatment edges were not used in any analysis to avoid border dimensionless) (in this study, the value of 1.14 was used, which was
effects. Pods and grains were collected and counted. All the samples adjusted by the value obtained by Farias et al. 2017) and ET0. The KcTr,x
were weighed with a precision scale (0.001 g) to determine fresh weight is adjusted by the model, taking into account the phenological stages
and 1000-grain weight (when there were grains). Samples were taken to and crop senescence. The actual or adjusted transpiration (TA, mm d− 1)
70 ◦ C aerated stoves for 72 h in order to determine constant dry weight, can be obtained by adjusting Tc to soil water stress conditions, using the
total dry matter (DMt) (above-ground biomass and yield, kg ha− 1), and water stress coefficient (KS, whose values range from 0 to 1), through TA
weighed again. = KSTC. KS describes the effects of soil water stress on the following crop
Leaf area index (LAI, cm2 cm2) was determined by the disc method growth processes (Raes et al., 2012): (1) reduction of the canopy
for all phenological stages, in which three leaf discs with a radius of 0.01 expansion rate; (2) acceleration of leaf senescence; (3) stomatal closure;
m were removed from each plant (ten in each treatment) using leaf dry and (4) by the penalties on HI after the beginning of the reproductive
matter samples (Farias et al., 2017). These values were used to calculate phase.
canopy cover (CC, %), which was used to compare the values simulated This value is defined in terms of a stress indicator (p, dimensionless)
by the AquaCrop model (Paredes et al., 2015; Pereira et al., 2015). The specifying an upper limit (0.40 was used in the present study) and a
CC values were derived using an exponential function of temporal lower limit (Raes et al., 2012; Vanuytrecht et al., 2014). This is different
degradation (Paredes and Torres, 2016): from the FAO56 approach, which considers a linear relationship be­
tween water stress and Tc. Therefore, the p values proposed by FAO56
CC = [1 − exp( − αLAI) ] (1)
were not used in the AquaCrop model, since the processes involved do
Where α (dimensionless) is the light extinction coefficient of the canopy. not have a linear relationship in most crops (Paredes and Torres, 2016).
This coefficient was initially discussed by Jeuffroy and Ney, who sug­ Moreover, the AquaCrop model performs a daily soil water balance
gested several ways for obtaining. It depends and varies for each crop in the root zone in terms of depletion (DR, mm). DR is expressed as a
and for the density of plants (Katerji et al., 2013; Pereira et al., 2015). fraction of total available water (TAW, mm) (Raes et al., 2012). DR
Several values of α are reported for cowpea, i.e., San José et al. (2004), ranges from zero (when soil water is at FC, i.e., 0% depletion, KS = 1) to
studying M-28–6–6 cultivar, obtained an average value of 0.66, and one (when soil water is 100% depleted KS = 0). Water stress begins to
Tesfaye et al. (2006), studying the Roba-1 cultivar in Ethiopia, found a affect plant growth when DR exceeds readily available water (RAW =
value of 0.86. Both cases were under ideal conditions (without water pTAW) (Vanuytrecht et al., 2014).
deficit). Teixeira et al. (2015), studying the Pérola and Radiante bean Thus, several irrigation management programs can be designed with
cultivars, found values of 0.79 and 0.74, respectively. the assistance of the model. When it is necessary to avoid water stress,
In this study, considering the LAI and the fraction of photosynthetic the model immediately applies irrigation before DR reaches RAW. On
radiation intercepted by the plant, the value of α was determined by the the other hand, when water stress needs to be allowed, the threshold for
coefficient of the exponential function over the cowpea cycle (Pengelly water application is higher than RAW. Each process affected by soil
et al., 1999). The α values were 0.83, 0.80, 0.77 and 0.74 for T1, T2, T3 water stress (senescence, stomatal opening, among others) has its own
and T4, respectively. C3 plants, such as cowpea, when compared to C4, upper and lower threshold levels (Raes et al., 2012). Es is also obtained
present different behaviors for light and water during the vegetative from CC* and ET0, through the equation:
development, which explains the high value for T1, since the higher the ES = KR (1 − CC∗ )Kex ET 0 (3)
α, the greater the interception of radiation by the canopy, reducing the
loss of water in the soil by evaporation (Casaroli et al., 2007), providing Where Kex (dimensionless) is the maximum soil evaporation coefficient
a higher LAI and consequently a higher CC, as found in the experiment. and KR (which ranges from 0 to 1) is the evaporation reduction coeffi­
Harvest index (HI, %) was determined by the ratio of grain yield cient; when KR < 1, there is water availability in the soil to respond to
(economically marketed part) on plants used to calculate the yield in the evaporative demand from the atmosphere.
each treatment to total above-ground biomass production, according to In the model, ETC directly depends on CC. Therefore, the model’s
the methodology proposed by Meireles (Souza et al., 2014). This index ability to produce good predictions of biomass and yield depends on the
evaluates the performance of the cultivar submitted to different condi­ proper parameterization of CC (Paredes et al., 2015; Pereira et al.,
tions, aiming at verifying the possibility of use in a given region. 2015).
The simulation of CC growth occurs through the combination of four
parts (sub-models) (Steduto et al., 2012; Raes et al., 2012): (1) soil water
2.5. AquaCrop model balance; (2) crop development, growth and yield; (3) an atmospheric
sub-model, the rain cycle, evaporative demand (ET0) and CO2 concen­
The AquaCrop model (Steduto et al., 2012; Raes et al., 2012) simu­ tration; (4) a management sub-model, which includes irrigation and
lates the growth of several crops (cotton, barley, potato, beet, sorghum, fertilization (Paredes et al., 2014). More details about the model are
wheat and etc.) as a function of canopy cover (CC). It was developed by available in the studies carried out by Steduto et al. (2012), Raes et al.
the FAO’s Land and Water Division in 2009 to assist agricultural man­ (2012), Paredes et al. (2014).
agement and production regarding the effect of water on yield. This The estimation of CC occurs for the whole cycle, considering three
index establishes a balance between precision, simplicity and robust­ growth phases (Raes et al., 2012; Vanuytrecht et al., 2014). The first
ness. It has already been widely tested for crops such as corn and tomato phase starts at emergence, characterized by initial canopy cover in the
(Katerji et al., 2013, Paredes et al., 2014), and through the parameter­ soil (CC0, %), which ends when half of the maximum CC (CCX, %) is
ization for new crops such as pea (Paredes and Torres, 2016), it is reached. The exponential function of time describing the rate of CC
possible to assist in planting planning and to analyze future scenarios. growth uses a canopy growth coefficient (CGC, % GDD− 1). The second
The model considers the daily time step to calculate ETc by treating phase lasts until the CCX is reached, at the level in which the model
crop transpiration (TC, mm d− 1) and soil evaporation (ES, mm d− 1) adopts an exponential function using the same CGC. The third phase
separately. However, this separation is different from the one used by refers to the decline in CC after the beginning of senescence, using a
FAO56 and was widely discussed by Pereira et al. (2015). Thus, Tc is coefficient of canopy decrease (CDC, % GDD-1).
given by a simple equation: Since cowpea is harvested a few days after leaf senescence begins, a
TC = (CC∗ × KcTr,x ) ET 0 (2) detailed determination of CDC was not relevant because of the short
periods. Wellens et al. (2013) followed the same procedure when
Tc is calculated by the canopy cover adjusted by the micro-advective

4
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Table 4
Values of leaf area index (LAI) observed in the experiment and used in the derivation of CC, divided by year and days after sowing (DAS).
LAI

2013 2014

DAS T1 T2 T3 T4 DAS T1 T2 T3 T4
15 0.35 0.34 0.33 0.36 9 0.16 0.15 0.16 0.15
19 0.53 0.56 0.54 0.59 12 0.30 0.27 0.29 0.27
25 0.95 1.04 0.86 0.96 16 0.73 0.59 0.66 0.56
29 1.54 1.52 1.36 1.56 19 0.98 0.87 0.92 0.83
32 2.27 2.33 2.12 1.99 22 1.68 1.59 1.50 1.45
36 2.71 2.55 2.37 2.37 25 2.24 2.20 2.13 2.09
39 2.86 2.78 2.64 2.47 30 2.64 2.59 2.56 2.49
43 3.25 3.04 2.81 2.84 32 2.84 2.79 2.70 2.64
46 3.20 3.03 2.81 2.70 36 3.03 2.94 2.87 2.85
50 3.26 3.07 2.93 2.66 39 3.23 3.12 3.07 2.99
54 3.16 2.83 2.71 2.43 43 3.49 3.25 3.14 3.01
57 3.07 2.68 2.53 2.25 46 3.52 3.28 3.16 3.03
64 2.71 2.52 2.33 2.07 50 3.30 3.05 2.87 2.74
53 3.14 2.88 2.69 2.55
60 2.85 2.49 2.27 2.15

The α values of 0.83, 0.80, 0.77 and 0.74 were used for T1, T2, T3 and T4, respectively.

parameterizing the AquaCrop for cabbage and Paredes and Torres 2.6. Parameterization, calibration and validation
(2016) for peas. In this study, the derivation of CC (Paredes and Torres,
2016) was based on the observed LAI values (Table 4). Parameterization was performed using the independent data sets
The model simulates dry matter above ground considering the ratio related to the experimental site during 2013 (calibration) and 2014
of biomass production to water transpired by the crop over the day. Crop (validation). Initially, it was performed for the parameters that influence
yield (Y, kg ha− 1) is simulated by the model from the biomass (B, kg CC and the T1 treatment, which did not suffer water restrictions and
ha− 1), through the following equation: then for the other parameters.
The initial values for the conservative parameters of cowpea were
Y= ƒHI HI 0 B (4)
selected from the values reported for peas, which is a relatively similar
Where HI0 (%) is the reference harvest index (Souza et al., 2014) that grain legume crop, and by parameters obtained experimentally ac­
indicates the harvestable proportion, and ƒHI (dimensionless) is an cording to the methodology adopted by Steduto et al. (2012). The same
adjustment factor calculated internally and depending on the treatment approach was adopted by Paredes and Torres (2016), who considered
used, water stress and other penalties. This adjustment may include the conservative soybean parameters as initial values for pea. These
impacts of water and temperature on final yield. Since this parameter is values were later refined through a trial-and-error procedure (until the
closely linked to the characteristics of each region, some authors suggest simulation reached 80% adjustment), in order to guarantee the best
that the index can be used only for the region and cultivar studied performance between simulated and observed variables.
(Fanceli, 2003). In the present study, HI0 (29%) was considered for T1, After completing this procedure, another trial-and-error procedure
because there was no water restriction in the stages of cowpea under this was applied to determine the best value of BWP*(Biomass water pro­
treatment. Therefore, the model input data in this study included: ductivity, g m− 2), through the execution of successive model simulations
(final biomass) used in comparison with the values observed. In sum­
(I) Climatic data from the INMET automatic station, containing in­ mary, the trial-and-error procedure began with the adjustment of the CC
formation from 2003 to 2014, consisting of daily and average parameters, followed by KcTr,x, CN and, finally, BWP*.
annual concentrations of CO2 in the atmosphere. In the present Model calibration and validation required at least two distinct series
study, the CO2 concentration of the model database and the of real data of the crop studied. In the first step (calibration), model
calculated ET0 values were used; parameters were adjusted until the responses approached the observed
(II) Soil data referring to textural and hydraulic properties of five values. The parameters were considered adjusted when they presented
layers (1 m); low values of errors and high values of efficiency in the simulations.
(III) Readily evaporable water (REW, mm) and curve number (CN, Regarding the validation of the results obtained for the two-time
dimensionless), calibrated to characterize soil evaporation and series, comparisons were made using graphical methods, i.e., hydro­
surface runoff, respectively. REW was estimated by using the grams and regression lines, or statistical analysis such as Student’s T-
texture and retention characteristics of the upper layer test, Wilmott’s concordance index (d) and the efficiency of the model
(0–0.20 m) (Allen et al., 1998); (EF) by Nash and Sutcliffe (Pereira et al., 2015; Lima et al., 2016).
(IV) Data on soil management practices that influence soil water
balance, i.e., irrigation practices, vegetation cover and runoff 2.7. Model performance evaluation
reduction;
(V) Irrigation data: (1) net water depths; (2) when there was no Regarding the model evaluation, and to assist the best adjustment of
irrigation; (3) a pre-determined irrigation program; and (4) the parameters used in the crop, several statistical indices were used
generation of an irrigation program, based on time criteria used according to the methodology adopted by Paredes and Torres (2016):
in the experiments (fixed range, permissible depletion expressed
in mm or RAW percentage) and depth criteria (refilling to FC or (I) The determination coefficient (R2) of the ordinary least squares’
using a fixed net water application expressed in mm or in RAW regression and the slope (b0) of the regression forced to the origin
percentage). Further information required in the model input is (FTO).
provided by Raes et al. (2012). (II) Indicators of residual estimation errors: normalized root mean
square error (NRMSE), mean percentage bias (PBias) and mean
absolute error (MAE), all expressed as a percentage. These

5
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Fig. 1. Simulation of canopy cover (black line) (CC, %) over the AquaCrop calibration period (2013) (T1 = a; T2 = b; T3 = c; T4 = d) compared to values observed
in the field (black balls). b0 e R2 are the regression and determination coefficients; NRMSE normalized root mean square error, MAE - mean absolute error, PBIAS -
error percentage, these values are expressed as percentage; EF - efficiency of the model.

indicators were calculated for each interaction, supporting the 1. Irrigated (I) up to field capacity aiming at minimizing economic and
parameterization by showing the smallest error obtained. production losses. It is noteworthy that an irrigation threshold was
(III) Modeling efficiency (EF), which indicates the quality of the used for acceptable depletion in cowpea, corresponding to 60% of
modeling and provides an indication of the relative magnitude of RAW; at this point, the automatic irrigation of the model was trig­
the mean square error compared to the data variance observed. gered to restore θFC;
All these values were calculated on a MS Excel spreadsheet. 2. Non-irrigated (NI) in order to evaluate the stress caused by lack of
water on yield, representing the way that cowpea is currently pro­
Since there are few observations, NRMSE could not be calculated for duced in the region.
the yield forecasts. Therefore, only the gross and the relative differences
were used, respectively, through the following equations: The Kolmogorov-Smirnov normality test was applied (Assistat soft­
ware, version 7.7) to verify if the dependent variable data (estimated
GD = (Pi − Oi) (5)
yield) followed a normal distribution at the significance level of 5%
(Silva and Azevedo, 2016). The estimated yields in all years did not
(Pi − Oi)
RD = (6) present a normal distribution. Thus, the non-parametric Kruskal-Wallis
Oi
test was applied at the p < 0.05 level to demonstrate that the supple­
Where GD (kg ha− 1) is the gross difference and RD (%) is the relative mentary irrigation strategy achieved high yields in the region.
difference. Oi (kg ha− 1) represents the observations and Pi (kg ha− 1) the All simulations were also evaluated in terms of relative yield
yield forecasts. decrease (RYD, %) (Paredes and Torres, 2016), calculated by:
( )
Ym − Ya
RYD = × 100 (7)
2.8. Evaluation of sowing dates and supplementary irrigation strategy Ym
adopted in the region
Where Ym (kg ha− 1) is the potential yield and Ya (kg ha− 1) is the actual,
Considering the ability of the AquaCrop model to forecast cowpea observed or estimated yield. The water productivity indicators (Pereira
yield and the simulation of soil water content under irrigation, the im­ et al., 2015; Paredes and Torres, 2016) were also used to analyze the
pacts on yield and use of water were evaluated for three sowing dates simulations:
(April 1st, 10th and 20th), which are used by producers in the region Irrigation water productivity (WPIrrig, kg m− 3).
(Conab, 2018). Both years of 2013 and 2014 and the series of 12 years
Ya
(2003–2014) were evaluated in the automatic meteorological station WPIrrig = (8)
IWU
from the municipality of Castanhal. Since farmers aim to achieve high
yields of cowpea and cost reduction, only two scenarios of automatic Where IWU (mm) is the water used in irrigation. It was based on the
programming by the model were evaluated for the three dates:

6
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Fig. 2. Simulation of canopy cover (black line) (CC, %) over the AquaCrop validation period (2014) (T1 = a; T2 = b; T3 = c; T4 = d) compared to values observed in
the field (black balls). b0 e R2 are the regression and determination coefficients; NRMSE - normalized root mean square error, MAE - mean absolute error, PBIAS -
error percentage, these values are expressed as percentage; EF - efficiency of the model.

ratio of the applied or estimated net water depth to the beneficial water (NIR, mm) for the three sowing dates were calculated considering the
use fraction (BWUF, dimensionless), which is the fraction of irrigation meteorological data series (12 years). These NIR values were obtained
applied to the soil water storage in the root zone. It was obtained as a from the simulations of the scenarios used in the study, at the level of
function of the seasonal distribution efficiency (0.9) of the system, probability of occurrence during the simulations. Therefore, NIR series
measured from the water captured along the crop root zone (Pereira were generated for all dates analyzed (Paredes and Torres, 2016).
et al., 2015).
Crop water productivity (WPET, kg m− 3). 3. Results
Ya
WPET = (9) 3.1. Model parameterization
ET Cact

Where ETC act (mm) is the actual crop evapotranspiration estimated by The parameters (such as CGC, CDG, α and etc.) that influence the
the model. simulation of CC by the model were evaluated by different indicators of
Farm water productivity (WP, kg m− 3). adjustment quality (threshold values > 80% adjustment), which are
shown in Figs. 1 and 2, for the comparison of observed CC values at the
WP =
Ya
=
Ya
(10) different treatments adopted and during the model calibration (2013)
TWU (Pe + CR + ∆SW + IWU) and validation (2014), respectively.
In the year of calibration (Fig. 1), the parameter values adopted for
Where TWU (mm) is the total water use, obtained by the sum of the
the simulation of the CC presented low estimative errors (NRMSE <
effective precipitation (Pe, mm), capillary rise (CR, mm), variation in
4.75%) and high EF > 0.97 in all treatments. These results were due to
soil moisture (ΔSW, mm) and IWU. Since the groundwater table is very
the high correlation (R2 > 0.98) between the observed and simulated
deep in the experimental site, CR was null. Pe was obtained by sub­
values. In addition, they showed small underestimation trends in the
tracting the water lost by surface runoff (RO, mm), which was estimated
simulation of CC (b0 < 0.93 and - 2.69% < PBIAS > 0.13%) in most
by AquaCrop through the PRP. ΔSW was calculated by subtracting the
treatments (T1, T2 and T4). It contributed to a low MAE (MAE < 5.25%)
final soil water (SWend) from the initial soil water (SWini) values, both
in all treatments.
computed in the field and compared with those values estimated by the
During the validation year (Fig. 2), there was a tendency to under­
model.
estimate the CC (b0 < 0.98% and 1.13% < PBIAS > 3.17%) in T2, T3 and
After analyzing the sowing dates adopted by producers in the region,
T4. However, the estimation errors were low and acceptable (NRMSE <
the water balance simulated by the model was demonstrated for the
6.33%) and EF was high (EF > 0.97), for all treatments (Figs. 1 and 2).
scenarios and years studied. In addition, the net irrigation requirements
These results indicate a good parameterization obtained for CC.

7
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Fig. 3. Simulation of soil water content (black line) (SWC, cm3 cm− 3) over the AquaCrop calibration period (2013) (T1 = a; T2 = b; T3 = c; T4 = d) compared to
values observed in the field (black balls). b0 e R2 are the regression and determination coefficients; NRMSE - normalized root mean square error, MAE - mean absolute
error, PBIAS - error percentage, these values are expressed as percentage; EF - efficiency of the model.

Fig. 4. Simulation of soil water content (black line) (SWC, cm3 cm− 3) during the AquaCrop model validation (2014) (T1 = a; T2 = b; T3 = c; T4 = d) compared to
values observed in the field (black balls). b0 and R2 - are the regression and determination coefficients; NRMSE - normalized root mean square error, MAE - mean
absolute error, PBIAS - error percentage, these values are expressed as percentage; EF - efficiency of the model.

The simulation of irrigation by the model can be assessed through the The adjustment quality indicators show a slight tendency to under­
dynamics of soil water content (SWC). Figs. 3 and 4 show this dynamics, estimate the SWC in the calibration year, with b0 > 0.73 and low values
simulated by the model, in relation to the dynamics observed in the of PBIAS (− 1.03% < PBIAS > 1.96%) for all treatments. Moreover,
field. estimation errors are low and acceptable (NRSME < 10.73% and MAE <

8
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Table 5 (2012).
Values of the AquaCrop parameters used for simulations. The best simulation by model in the initial cowpea phase was due to
Description Symbols and Units Value Determining the use of the same amount of water over this phase, which was also
method found by Pereira et al. (2015). In the reproductive phase, in which there
Canopy curve parameters was differentiation in the treatments, there were low underestimation
Canopy cover with 90% CC0, % 3.7 Adjusted values due to the treatments adopted, but they did not affect the model
emergency performance. It is due to the short last stage of cowpea, which facilitates
Maximum canopy cover CCX, % 93 Observed the parameterization (CC) and, consequently, the simulation of SWC, TA
(Experimental)
Canopy growth CGC, Fraction per degree 0.97 Adjusted
and ES, as found by Paredes and Torres (2016).
coefficient of daily growth The second phase lasts until CCX (in this study, the value of 93% was
Canopy reduction CDC, Fraction per degree 0.53 Adjusted used) is reached at the level in which the model adopts an exponential
coefficient of daily growth function using the same CGC; and finally, the third phase refers to the
Crop parameters
decline in CC after the beginning of senescence and uses a canopy
Basal coefficient of the KC TR X, Coefficient of 1.14 Observed
crop for transpiration crop transpiration, (Experimental) decrease coefficient (CDC, % GDD− 1). The CGC and CDC parameters
on CCX dimensionless (0.97% and 0.53% GDD− 1), respectively, were adjusted and used for
Adjusted biomass BWP*, g m− 2 16.5 Adjusted both years to simulate CC.
productivity (Water)
Maximum effective root Ze, m 0.3 Observed
zone (Experimental)
3.2. Biomass and yield simulations
Reference harvest index HI0, % 29 Observed
(Experimental) Biomass (B) and grain yield (Y) of cowpea, both simulated by the
Inhibition coefficient of HI increase by inhibition 10 adjusted model, were used to validate the predictions in comparison to the
leaf growth due to the of leaf growth at anthesis
observed values. Since CC was well simulated by the AquaCrop model
HI
Inhibition coefficient of HI increase by inhibition 8 Adjusted (Figs. 1 and 2), the simulation and accuracy for both B (Fig. 5) and Y
stomata due to the HI of leaf stomata at (Table 6) were also satisfactory.
anthesis In general, biomass simulations along the two harvests used for
Water depletion thresholds in the soil calibration (2013) and validation (2014) showed an underestimation
Water depletion Pexp upper, TAW Fraction 0.1 Adjusted
threshold in the soil to at which CGC becomes
trend. The regression coefficient (b0) ranged between 0.93 and 1.16,
leaf expansion null with negative biases (PBIAS) ranging from − 3.91 to − 19.13%. These
Water depletion Pexp lower, TAW fraction 0.35 Adjusted trends did not affect the model performance, since it had high efficiency
threshold in the soil to at which CGC starts (EF) in the simulation of B, for both crops, which resulted in values
leaf growth getting low
higher than 0.88. Additionally, estimation errors were acceptable
Coefficient curve form of Highly convex curve 1 Model default
leaf growth (15.81% < NRMSE > 21.82% and 18.20% < MAE > 23.71%) for these
Water depletion Psto, TAW fraction at 0.5 Adjusted predictions due to the high EF of the model in biomass simulation
threshold in the soil which stomata starts (Fig. 5).
for stomatal closure closing This underestimation was found over the early cowpea stages. It was
Coefficient curve form of Highly convex curve, 3 Model default
stomatal conductance dimensionless
particularly related to the underestimation of the CC (Figs. 1 and 2)
Water depletion Psen, TAW fraction at 0.38 Adjusted curve at these stages. A similar pattern was found in the study by Paredes
threshold in the soil which senescence starts and Torres (2016).
for senescence In general, the model also showed a satisfactory efficiency (EF) in the
Water depletion Ppol, TAW fraction at 0.6 Model default
simulation of cowpea yield, both in calibration and validation for all
threshold in the soil which there is a
for pollination pollination failure treatments adopted. During 2013, the efficiency value was 0.95, due to
Soil water balance parameters the trend of overestimating the yield for T4 (difference of 8.81% to the
Easily evaporated water REW, mm 9 Observed observed value). On the other hand, the other treatments showed a small
(Experimental) difference between the observed and simulated yields, ranging from
Curve number CN, dimensionless 70 Adjusted
− 2.68 to 1.13% (considering the observed mean values).
T1 and T3 showed a tendency of underestimation in calibration
8.91%), since they did not affect the EF of the model, which was higher (− 2.65% and − 2.48% of the observed mean values). However, during
than 0.77 for all treatments (Fig. 3). validation, the model was more efficient (EF = 0.99) in the simulation of
During the validation, the results were also satisfactory for the Y. It is due to the small tendency of overestimation of the simulated
simulation of SWC. The adjustment quality indicators showed a small values of Y, which ranged from 0.34% to 5.39% of the observed average
tendency of underestimation, with values of b0 close to 0.92 and values yield (Table 6).
of PBIAS ranging from 2.32% to 5.16% for all treatments. Overall, the
errors are acceptable (6.65% < NRMSE > 8.12% and 4.17% < MAE > 3.3. Evaluation of sowing dates adopted in the region and supplementary
7.66%) for these treatments. In summary, these error values did not irrigation strategy
influence the model efficiency in the simulation of SWC during the
validation step, since the EF values were higher than 0.70 (T1) (Fig. 4). Table 7 shows the supplementary irrigation strategy for 2013 and
These results demonstrate that the model can support irrigation 2014 and the meteorological series (2003–2014) included in the model,
strategies in the study area, since the observed SWC variation was well considering the three sowing dates (April 1st, 10th and 20th).
simulated by the model. Regarding the results, for scenario I (irrigated), the water values used
Satisfactory results were achieved during model calibration and in irrigation and estimated actual evapotranspiration (IWU and ETC act)
validation. Table 5 shows all the parameter values obtained and used for increased according to the delay in the sowing date, from the beginning
CC, crop water use and yield, fractional thresholds of soil water deple­ to the end of April, in the two simulated years (2013 and 2014) and for
tion, and soil water balance. The approaches used to obtain these the series analyzed (2003–2014). It is due to the decrease in precipita­
parameter values also include field observations. The parameterization tion at the end of April, which had increased the need for irrigation and
procedure and the default values were from the study by Raes et al. the amount of water lost by the crop. On the other hand, values of total
water use (TWU) had decreased in most simulated years (2014 and

9
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Fig. 5. Simulation of biomass (B, kg ha− 1) by the AquaCrop model over the calibration (2013) and validation (2014) stages, compared to the values observed in the
field. b0 e R2 - are the regression and determination coefficients; NRMSE - normalized root mean square error, MAE - mean absolute error, PBIAS - error percentage,
these values are expressed as percentage; EF - efficiency of the model.

Table 6
Differences between the observed and simulated yield for cowpea and the efficiency of the AquaCrop model for this prediction.
Year Treatments Observed Simulated Indicators

Yield Yield Difference


− 1 − 1
Y (kg ha ) Ya (kg ha ) (kg ha− 1) (%) EF

2013 (Calibration) T1 1319.97 1285.00 − 34.97 − 2.65 0.95


T2 1222.16 1236.00 13.84 1.13
T3 1188.49 1159.00 − 29.49 − 2.48
T4 817.94 890.00 72.06 8.81
2014 (Validation) T1 1569.18 1590.00 20.82 1.33 0.99
T2 1233.54 1300.00 66.46 5.39
T3 1002.26 1013.00 10.74 1.07
T4 792.34 795.00 2.66 0.34

EF refers to model efficiency.

2003–2014) with the delay of sowing, which may be due to the reduc­ ETc values with the delay in sowing. Consequently, there was a decrease
tion of precipitation in the region. in the water use efficiency by the crop (WPET) (Table 7).
Therefore, yields are higher on ending sowing dates for 2013 Considering the results of WP on both scenarios I and NI in 2013, the
(1700 kg ha− 1) and for the meteorological series (1598 kg ha− 1), with highest values were obtained for the sowing of April 1st (0.62 and
the exception of 2014 (higher yield on April 10th). RYD ranged from 0.63 kg m− 3, respectively). However, in 2014, the values referring to
5.6% to 7.8% in 2013 and from 3.9% to 9.9% in 2014, with average sowing on April 10th (0.64 and 1.21 kg m− 3) were the highest in the
(2003–2014) that increased as a function of the delay in sowing. Thus, study. For the meteorological data series (2003–2014), April 20th (0.43
when considering scenario I, sowing at the beginning of April has a and 0.46 kg m− 3) showed the highest values. All of them were due to
lower chance of RYD and a lower use of IWU. Both results were high yields in these days and lower TWU values. These results show that
confirmed by the WPIrrig values, which are higher with earlier sowing in a year without water deficit (2013 and 2003–2014), there is a lower
(April 1st). Therefore, under the first scenario, values between 1.41 and variation in water productivity. When the scenarios in use and years
4.69 kg m− 3 were reached (Table 7). with a lower precipitation distribution (2014) are considered, there is a
Regarding the NI (non-irrigated) scenario, which is predominant in greater variation in this efficiency (Table 7).
the region, RYD had a smaller variation between sowing dates. How­ The simulation of water balance by the model for these adopted
ever, they were quite high. During the experiment performed in 2013, scenarios (I and NI) shows that with the delay in sowing, a greater
they ranged from 11.5% to 18%. On the other hand, they ranged from amount of water is lost through infiltration (Infil), drain and runoff in
17.6% to 19.4% in 2014. Considering the series of meteorological data, both simulated scenarios and years analyzed. Sowing on April 1st also
they ranged from 9.1% to 16.5%. Therefore, there is a noticeable rela­ ensures a better use of water by the crop through transpiration (Tc) and
tion with the distribution of water in the soil. When sowing was delayed, fewer losses of water by evaporation (Es). This result shows that, in
under the NI scenario, there was a greater risk of higher RYD (RYD = addition to leading to a higher production gain (Table 7), this date de­
16.5%, for the series from 2003–2014), especially due to the lack of creases the losses of water by other balance components (Table 8).
water for the crop during stages with a greater water requirement Fig. 6 shows the probability curves of the net irrigation requirements
(reproductive phase). Likewise, in scenario I, NI showed an increase of (NIR) for the three sowing dates, resulting from model simulation for the

10
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Table 7
Indicators of yield and water use efficiency of the supplementary irrigation strategy and different sowing dates adopted in the region for 2013, 2014 and the series of
the INMET automatic station (2003–2014).
Year Sowing Scenario Ya (kg RYD Pe (mm) IWU ETC act TWU WPIrrig (kg WPET (kg WP (kg
ha− 1) (%) (mm) (mm) (mm) m− 3) m− 3) m− 3)

2013 April 1st* I 1585 6.8 219 42 196 256 3.77 0.81 0.62
NI 1400 17.6 219 – 173 221 – 0.81 0.63
April I 1585 7.8 264 44 201 327 3.60 0.79 0.48
10th* NI 1410 18 264 – 178 285 – 0.79 0.49
April I 1700 5.6 292 47 221 349 3.62 0.77 0.49
20th* NI 1593 11.5 292 – 207 315 – 0.77 0.51
2014 April 1st* I 1633 3.9 180 116 133 290 1.41 1.23 0.56
NI 1400 17.6 180 – 114 159 – 1.23 0.88
April I 1648 4.2 117 131 138 259 1.26 1.19 0.64
10th* NI 1400 18.6 117 – 118 116 – 1.19 1.21
April I 1622 9.9 137 136 152 284 1.19 1.07 0.57
20th* NI 1450 19.4 137 – 136 153 – 1.07 0.95
2003–2014 April 1st* I 1590 6.5 354 34 202 397 4.69 0.79 0.40
NI 1546 9.1 354 – 197 362 – 0.79 0.43
April I 1588 7.7 335 40 204 386 3.99 0.78 0.41
10th* NI 1533 10.9 335 – 197 335 – 0.78 0.46
April I 1598 11.2 319 48 210 372 3.31 0.76 0.43
20th* NI 1503 16.5 319 – 198 326 – 0.76 0.46

Ya – Estimated yield, RYD – Relative yield decreasing, Pe – Effective precipitation, IWU – Water used in irrigation, ETC act – Actual estimated evapotranspiration, TWU –
Total water use, WPIrrig – Irrigation water productivity, WPET – Crop water productivity, WP – Farm water productivity, I – Irrigated, NI – Non irrigated. *- It represents
that there was a difference between the averages (estimated yield) for the sowing dates and the scenario, using the Kruskal-Wallis test (p < 0.05%).

Table 8
Water balance of the supplementary irrigation strategy and different sowing dates adopted in the region for 2013, 2014 and the series of the INMET automatic station
(2003–2014).
Year Sowing Scenario PRP (mm) ETC act (mm) Irrig (mm) ES (mm) TC (mm) Drain (mm) Infil (mm) Runoff (mm)

2013 April 1st I 254 196 58 90 106 25 81 10


NI 254 173 – 112 61 20 46 15
April 10th I 310 201 53 94 107 27 116 19
NI 310 178 – 116 62 22 86 24
April 20th I 342 221 42 99 122 30 113 20
NI 342 207 – 101 106 25 85 25
2014 April 1st I 210 133 118 35 98 62 103 30
NI 210 114 – 62 52 4 66 26
April 10th I 128 138 122 47 91 8 101 3
NI 128 118 – 61 57 2 7 1
April 20th I 171 152 104 54 98 10 89 24
NI 171 136 – 88 48 14 1 20
2003–2014 April 1st I 440 202 33 92 110 50 185 36
NI 440 197 – 98 99 45 157 41
April 10th I 425 204 23 103 101 50 154 40
NI 425 197 – 105 92 60 138 30
April 20th I 391 210 38 97 113 40 147 32
NI 391 198 – 100 98 50 121 22

PRP – Precipitation, ETC act – Actual estimated evapotranspiration, Irrig – Irrigation, Es – Soil evaporation, Tc – Crop transpiration, Infil – Infiltration, I – Irrigated, NI –
Non-irrigated.

series included. They show that NIR varied highly from 5 to 104, 9–118 yield prediction (Pereira et al., 2015).
and 10–122 mm for April 1st, April 10th and the last date of sowing, In this study, suitable SWC simulation results were obtained for both
respectively. This variability results mainly from the high variation in years (calibration and validation). It shows that these values are
the effective precipitation between the years in the region, which ranged consistent with the values observed in the field, which also indicates that
from 180 to 513, 117–482 and 137–442 mm when sowing was per­ the parameterization of the model for CC and soil water content pa­
formed on April 1st, 10th and 20th, respectively (Fig. 6). rameters (KC TR X, coefficient of water stress in soil Ks, REW e CN) had
Therefore, farmers who sow at the end of April are more likely to satisfactory adjustments and are adequate.
apply irrigation, especially in severe summer years (122 mm). In gen­ When compared to models tested in the simulation of SWC, the
eral, although the first date experiences a greater amount of effective AquaCrop model demonstrated a better performance. Dallacort et al.
precipitation, farmers must apply a small water depth, which demon­ (2010), for example, using the CS-CROPGRO model for different con­
strates the need for irrigation. ditions and water depths in common bean cultivation, found a correla­
tion coefficient (R2) ranging from 0.70 to 0.74. In addition, these authors
4. Discussion found considerable differences between the observed and simulated
values, which indicates inefficiency in the simulation of water absorp­
Many authors state that a proper parameterization of the CC curve tion and redistribution into the soil. More recently, Santos et al. (2016),
and the water use parameters are necessary to obtain the soil water using the same model (CS-CROPGRO) for the IAC-Carioca cultivar in
balance provided by the model (Paredes et al. 2015; Pereira et al. 2015). Jaboticabal (SP), also reported low model efficiency in the simulation of
Thus, the AquaCrop model can assist in irrigation strategies and crop SWC, which was due to low correlation between the values (R2 = 0.55)

11
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Fig. 6. Probability curves for the net irrigation requirements (NIR) until the cowpea harvest, during sowing in April for the 2003–2014 series.

and high estimation errors (RMSE = 0.037 cm3 cm− 3) for the (2017).
approached years. Therefore, the AquaCrop model proved to be useful in Since the model was also applied for the series of climatic conditions
estimating SWC for local and experimental conditions. of the region (2003–2014), the duration of cowpea cycle for sowing until
During parameterization, the AquaCrop model does not follow the April 1st ranged from 67 to 69 days. On the other hand, for the sowing
approach suggested by FAO56 and uses different thresholds for the on April 10th, the cycle ranged from 65 to 66 days and had small du­
impacts of soil water stress on the various physiological processes. If the rations, from 64 to 65 days, which corresponded to sowing until April
upper limit value proposed by Allen et al. (1998) (pupper = 0.45) were 20th. Such differences are due to the fact that the crop cycle develops in
taken into account, the model would consider cowpea as extremely advance under high temperatures (Paredes and Torres, 2016). In the
tolerant to water stress regarding leaf expansion and moderately sensi­ study carried out by Farias et al. (2017), it was found that the cowpea
tive to senescence. Consequently, the model performance would not cycle is anticipated when it is sown in the months with warmer tem­
agree with results found by Fernandes et al. (2015) and Farias et al. peratures, and yield may decrease due to the lack of water during the
(2017). Moreover, for the pollination failure (pupper) in the AquaCrop, a reproductive phase of the crop. This response is similar to the results
minimum value of 0.75 is established. Thus, according to several found in the simulation performed by the model in the present study.
studies, the cowpea crop is considered as sensitive to water stress (Souza Late sowing in Castanhal can decrease not only the period available for
et al., 2017). In relation to this condition, in this study, values for these harvest but also final yield. This result may increase the demand for
parameters (Pupper and Ks curve forms) were adopted by conviction and supplementary irrigation (especially during the reproductive phase),
references (Table 5). which could increase final yield for farmers and lead to optimal crop
Therefore, it is noticeable that the parameterization adopted in this water use.
study is proper. Moreover, it emphasizes that the AquaCrop model The simulation of the water balance by the model showed that it can
demonstrates high efficiency for simulation of cowpea biomass pro­ provide further insights into the strategies adopted by farmers in the
duction and yield predictions for all adopted treatments. These results region. It is important to mention that these results are simulated and
are similar to other model applications for cowpea: Bastos et al. (2002) the users must be careful, since there was no validation with a field
found differences ranging from − 2.0 to − 9.8%using the CROPGRO experiment. Finally, the evaluated scenarios demonstrate that when
model for the BR 14 Mulato cowpea cultivar; Dallacort et al. (2011), farmers perform late sowing, they are more likely to use supplementary
using the CS-CROPGRO model, found a difference of 29%; Santos et al. irrigation, reaching up to 122 mm. It reinforces that the study of the
(2016), also using the CS-CROPGRO model in the simulation of common climatic variability of the region has a great importance when high
bean yield, found a difference ranging from − 16.7–51.6% in the treat­ yields and better water use are intended. Therefore, the AquaCrop
ments adopted. model can assist in this understanding and management.
More recently, in the study by Nunes et al. (2019), with the Aqua­
Crop model for assessing climate risk and the performance of the model 5. Conclusions
for the simulation of cowpea yield, the results showed that the model
can be used in the simulation of yields for the region, since it was sen­ This study presents an approach to the parameterization of the
sitive to the seasonality and proved to be consistent in these predictions, AquaCrop model for the cowpea crop focusing on water use, biomass
although it was not validated in the field. yield under different irrigations during the reproductive phase, and the
In turn, the yield results found in this study demonstrate that the use impact of sowing dates normally used. The ability of the model to
of irrigation during the dry season in the region can be an alternative not simulate canopy cover (CC), soil water content, biomass, and final grain
only for managing to produce a new crop but also for minimizing the yield was assessed using soil water content and crop observations during
effects of lack of water during the reproductive period of the season, as two years (calibration and validation). Despite the limited dataset used,
found by Farias et al. 2017, Nunes et al. 2019. parameterization was successfully tested, resulting in very low de­
The differences between the adopted scenarios are relevant viations in biomass and yield prediction.
(Table 7), as well as the sowing dates, which indicate that later sowing The satisfactory predictions for biomass and yield obtained in the
should be avoided (April 20th). Fortunately, sowing has been practiced present study refers to:
by most farmers (Conab, 2018). In addition, adopting supplementary
irrigation, especially over the reproductive period of cowpea, has been
suggested in other studies, such as Farias et al. (2017), Souza et al.

12
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

i. The data quality obtained in the field and the crop parameters. Dallacort, R., Freitas, P.S.L., Faria, R.T., Gonçalves, A.C.A., Rezende, R., Guimarães, R.M.
L., 2011. Simulation of bean crop growth, evapotranspiration and yield in Paraná
For studies on parameterization of crop models, such efficiency
state by the CROPGRO-Dry bean model. Acta Sci. Agron. 33 (3), 429–436.
can be achieved when using at least two-year series; Fanceli, A.L., 2003. Fisiologia, nutrição e adubação do milho para alto rendimento, first
ii. The suitable parameterization of the CC curve; and ed. ESALQ/USP, São Paulo.
iii. Optimal parameterization relative to soil water content Farias, V.D.S., Lima, M.J.A., Nunes, H.G.G.C., Sousa, D.P., Souza, P.J.O.P., 2017. Water
demand, crop coefficient and uncoupling factor of cowpea in the Eastern Amazon.
simulation. RC 30 (1), 190–200. https://doi.org/10.1590/1983-21252017v30n121rc.
Fernandes, F.B.P., Lacerda, C.F., Andrade, E.M., Neves, A.L.R., Sousa, C.H.C., 2015.
The results of this study provide a noteworthy basis for further Efeito de manejos do solo no déficit hídrico, trocas gasosas e rendimento do feijão-
de-corda no semiárido. Rev. Ciênc. Agron. 46 (3), 506–515.
research, including the need to refine the parameterization, considering Freire Filho, F.R., Cravo, M.S., Ribeiro, V.Q., Rocha, M.M., Castelo, E.O., Brandão, E.S.,
a broader range of environments and varieties of cowpea, as well as the Belmino, C.S., Melo, M.Í.S., 2009. BRS Milênio e BRS Urubuquara cultivares de
use of climatic series with more years of data, contributing with the feijão-caupi para a região Bragantina do Pará. Rev. Ceres 56 (6), 749–752.
Hoogenboom, G., Jones, J.W., Boote, K.J., 1992. Modeling growth, development, and
establishment of the most appropriate sowing period using the model. yield of grain legumes using soygro, pnutgro and beangro: a review. Trans. ASAE 35,
The parameterized model was used for evaluating supplementary 2043–2055.
irrigation for cowpea regarding yield and water productivity. The results Katerji, N., Campi, P., Mastrorilli, M., 2013. Productivity, evapotranspiration, and water
use efficiency of corn and tomato crops simulated by AquaCrop under contrasting
indicate that the use of a threshold that corresponded to 60% of readily water stress conditions in the Mediterranean region. Agric. Water Manag. 130,
available water (RAW), aiming at reducing irrigation costs and mini­ 14–26.
mizing yield losses, even in a rainy season, decreased the relative yield Lima, M.J.A., Farias, V.D.S., Costa, D.L.P., Sampaio, L.S., Souza, P.J.O.P., 2016. Efeito
combinado das variáveis meteorológicas sobre a condutância estomática do feijão-
loss (RYD) by up to 10%. Consequently, it resulted in increased cowpea
caupi. Hort. Bras. 34, 547–553. https://doi.org/10.1590/S0102-053620160414.
production (average increase of 65 kg ha− 1 in production). The best Lima Filho, A.F.L., Coelho Filho, M.A., Heinemann, A.B., 2013a. Calibração e avaliação
sowing date corresponds to early April, while late sowing should be do modelo CROPGRO para a cultura do feijão caupi no Recôncavo Baiano. Rev. Bras.
avoided. In addition, after assessing the sowing dates for the experiment Eng. Agríc. Ambient. 17 (12), 1286–1293.
Lima Filho, A.F.L., Coelho Filho, M.A., Heinemann, A.B., 2013b. Determinação de épocas
years, supplementary irrigation should be adopted to ensure high yields. de semeadura do feijão caupi no recôncavo baiano através do modelo CROPGRO.
However, the proper application requires support to farmers’ decisions Rev. Bras. Eng. Agríc. Ambient. 17 (12), 1294–1300.
on irrigation values. Therefore, the AquaCrop model can be a useful tool Louzada, J.A., Jobim, C.I.P., 2011. Demanda de irrigação suplementar e rendimento
relativo de grãos de feijão no Rio Grande do Sul. Rev. Bras. Agrometeorol. 16,
to help decision-making for cowpea irrigation management. 237–247.
Martins, M.A., Alvalá, R.C.S., Tomasella, J., 2014. Modelos de produtividade agrícola
aplicados à agricultura de sequeiro: limitações e vantagens para avaliação de colapso
Declaration of Competing Interest
de safras. ACSA 10 (4), 41–48.
Merriam, J.L., Keller, J., 1978. Farm Irrigation System Evaluation: A Guide for
The authors declare that they have no known competing financial Management, first ed. Department of Agricultural Engineering, California.
interests or personal relationships that could have appeared to influence Moradi, R., Koocheki, A., Mahallati, M.N., Mansoori, H., 2013. Adaptation strategies for
maize cultivation under climate change in Iran: irrigation and planting date
the work reported in this paper. management. Mitig. Adapt. Strateg. Glob. Change 18, 265–284.
Nunes, H.G.G.C., de Pinho Sousa, D., Moura, V.B., Ferreira, D.P., de Nóvoa Pinto, J.V., de
Acknowledgements Oliveira Vieira, I.C., da Silva Farias, V.D., de Oliveira, E.C., de Oliveira Ponte de
Souza, P.J., 2019. Perfomance of the AquaCrop model in the climate risk analysis
and yield prediction of cowpea (Vigna Unguiculatta L. Walp). Aust. J. Crop Sci. 13
We thank the Federal Rural University of the Amazon, the Program (07), 1105–1112. https://doi.org/10.21475/ajcs.19.13.07.p1590.
of Post Graduate in Agronomy and the National Counsel of Technolog­ Oliveira, E.C., Costa, J.M.N., Paula Jr., T.J., Ferreira, W.P.M., Justino, F.B., Neves, L.O.,
2012. The performance of the CROPGRO model for bean (Phaseolus vulgaris L.)
ical and Scientific Development (CNPq, Brazil) for the financial support yield simulation. Acta Sci. Agron. 34 (3), 239–246.
to this research through the Universal Project (process No. 483402/ Paredes, P., Torres, M.O., 2016. Parameterization of AquaCrop model for vining pea
2012-5). In addition, the author HGGCN thanks to PCI-MCTIC/MPEG biomass and yield predictions and assessing impacts of irrigation strategies
considering various sowing dates (Online). Irrig. Sci. 35 (1), 27–41. https://doi.org/
and to the CNPq for the scholarship (process No. 302200/2020-7). 10.1007/s00271-016-0520-x.
Paredes, P., Melo-Abreu, J.P., Alves, I., Pereira, L.S., 2014. Assessing the performance of
References the FAO AquaCrop model to estimate maize yields and water use under full and
deficit irrigation with focus on model parameterization. Agric. Water Manag. 144,
81–97.
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration. Guidelines
Paredes, P., Wei, Z., Liu, Y., Chi, W.W., Pereira, L.S., 2015. Performance assessment of
for computing crop water requirements. Irrigation and Drainage Paper No. 56. FAO,
the FAO AquaCrop model for soil water, soil evaporation, biomass and yield of
Rome.
soybeans in North China plain. Agric. Water Manag. 152, 57–71.
Alvares, C.A., Stape, J.L., Sentelhas, P.C., Gonçalves, J.L.M., Sparovek, G., 2013.
Pengelly, B.C., Blamey, F.P.C., Muchow, R.C., 1999. Radiation interception and the
Köppen’s climate classification map for Brazil. Meteorol. Z. 22, 711–728. https://
accumulation of biomass and nitrogen by soybean and three tropical annual forage
doi.org/10.1127/0941-2948/2013/0507.
legumes. Field Crop. Res. 63, 99–112.
Barterng, P., Hoogenboom, G., Patanothai, A., Singh, P., Wani, S.P., Pathak, P.,
Pereira, L.S., Paredes, P., Rodrigues, G.C., Neves, M., 2015. Modeling malt barley water
Tongpoonpol, S., Atichart, S., Srihaban, P., Buranaviriykul, S., Jintrawet, A.,
use and evapotranspiration partitioning in two contrasting rainfall years. Assessing
Nguyen, T.C., 2010. Application of the cropping system model CSM-CROPGRO-
AquaCrop and SIMDualKc models. Agric. Water Manag. 159, 239–254. https://doi.
Soybean for determining optimum management strategies for soybean in tropical
org/10.1016/j.agwat.2015.08.017.
environments. J. Agron. Crop Sci. 196, 231–242.
Raes, D., Steduto, P., Hsiao, T.C., Fereres, E., 2012. AquaCrop Model. Crop Water
Bastos, E.A., Folegatti, M.V., Faria, R.T., Júnior, A.S.A., Cardoso, M.J., 2002. Simulation
productivity. Calculation Procedures and Calibration Guidance. Reference Manual.
of growth and development of irrigated cowpea in Piauí state by CROPGRO model.
Version 4.0. FAO, Rome.
Pesq. Agropec. Bras. 37 (10), 1381–1387.
Ramos, T.B., Gonçalves, M.C., Martins, J.C., Pereira, L.S., 2014. Comparação de
Bastos, E.A., Ferreira, V.M., Silva, C.R., Andrade Jr., A.S., 2008. Evapotranspiração e
diferentes funções de pedotransferência para estimar as propriedades hidráulicas em
coeficiente de cultivo do feijão-caupi no vale do Gurguéia. Irriga 13 (2), 182–190.
Portugal. In: Gonçalves, M.C., Ramos, T.B., Martins, J.C. (Eds.), Solo. Produção
Bernardo, S., Soares, A.A., Mantovani, E.C., 2006. Manual de Irrigação, eighth ed. UFV,
Agrária e Sustentabilidade dos Ecossistemas. INIA-V, Oeiras, pp. 29–34.
Viçosa.
Ramos, T.B., Gonçalves, M.C., Pereira, L.S., 2016. Características de retenção de água no
Carvalho, J.J., Saad, J.C.C., Bastos, A.V.S., Naves, S.S., Soares, F.A.L., Vidal, V.M., 2014.
solo para utilização na rega das culturas. INIA-V, Oeiras. 〈http://www.researchgate.
Teor e acúmulo de nutrientes em grãos de feijão comum em semeadura direta sob
net/publication/309735078〉 (accessed 20 March 2017).
déficit hídrico. Irriga 1, 104–117.
San José, J.J., Montes, R., Nikonova, N., Valladares, N., Buendia, C., Malave, V.,
Casaroli, D.D., Fagan, E.B., Simon, J., Medeiros, S.P., Manfron, P.A., Neto, D., Van
Bracho, R., 2004. Dry-matter partitioning and radiation-use efficiency in cowpea
Lier, Q.J., Müller, L., Martin, T.N., 2007. Radiação solar e aspectos fisiológicos na
cultivars (Vigna unguiculata (L.) Walp. cvs TC-9-6 and M-28-6-6) during consecutive
cultura de soja - uma revisão. Rev. FZVA 14 (2), 102–120.
seasonal courses in the Orinoco llanos. J. Agric. Sci. 142, 163–175.
Conab – Companhia Nacional de Abastecimento, 2018. Monitoramento Agrícola. Safra
Santos, M.G., Faria, R.T., Palaretti, L.F., Dantas, G.F., Dalri, A.B., Lopes, A.S., 2016.
2017/18. Acompanhamento da safra brasileira (grãos). Conab, Brasília.
Calibration and testing of CS-CROPGRO model for common beans. Eng. Agríc. 36
Dallacort, R., Freitas, P.S.L., Faria, R.T., Gonçalves, A.C.A., Jácome, G.A., Rezende, R.,
(6), 1239–1249. https://doi.org/10.1590/1809-4430-Eng.Agric.v36n6p1239-1249/
2010. Soil water balance simulated by CROPGRO-Dry bean model for
2016.
edaphoclimatic conditions in Maringá. Rev. Bras. Eng. Agríc. Ambient. 14 (4),
351–357.

13
H.G.G.C. Nunes et al. Agricultural Water Management 252 (2021) 106880

Saxton, K.E., Rawls, W.J., 2006. Soil water characteristic estimates by texture and Teixeira, W.G., Behring, S.B., 2017. Retenção de água no solo pelos métodos da mesa de
organic matter for hydrologic solutions. Soil Sci. Soc. Am. J. 70, 1569–1578. tensão e da câmara de Richards, in: Teixeira, P.C. (Ed.), Manual de métodos de
Silva, F.A.S., Azevedo, C.A.V., 2016. The assistat software Version 7.7 and its use in the análise de solo. Embrapa, Brasília, pp. 34–46.
analysis of experimental data. Afr. J. Agric. Res. 11, 3733–3740. Tesfaye, K., Walker, S., Tsubo, M., 2006. Radiation interception and radiation use
Souza, L.H.B., Peixoto, C.P., Silveira, P.S., Ledo, C.A.S., Lima, V.P., Santos, A.P.S.G., efficiency of three grain legumes under water deficit conditions in a semi-arid
2014. Características agronômicas e rendimento de girassol em diferentes épocas de environment. Eur. J. Agron. 25 (1), 60–70.
semeadura e populações de plantas no recôncavo da Bahia. Biosci. J. 30, 90-10. Vanuytrecht, E., Raes, D., Steduto, P., Hsiao, T.C., Fereres, E., Heng, L.K., Garcia-Vila, M.,
Souza, P.J.O.P., Farias, V.D.S., Lima, M.J.A., Ramos, T.F., Sousa, A.M.L., 2017. Cowpea Mejias Moreno, P., 2014. AquaCrop: FAO’S crop water productivity and yield
leaf area, biomass production and productivity under different water regimes in response model. Environ. Model. Softw. 62, 351–360.
Castanhal, Pará, Brazil. RC 30 (3), 748–759. Wellens, J., Raes, D., Traore, F., Denis, A., Djaby, B., Tychon, B., 2013. Perfomance
Steduto, P., Hsiao, T.C., Fereres, E., Raes, D., 2012. Crop Yield Response to Water. assessment of the FAO AquaCrop model for irrigated cabbage on farmer plots in a
Irrigation and Drainage Paper No. 66. FAO, Rome. semi-arid environment. Agric. Water Manag. 127, 40–47.
Teixeira, G.C.S., Stone, L.F., Heinemann, A.B., 2015. Eficiência do uso da radiação solar e
índices morfofisiológicos em cultivares de feijoeiro. Pesq. Agropec. Trop. 45 (1),
9–17.

14

You might also like