Heriot-Watt University Research Gateway

Prediction of Consumptive Use Under Different Soil MoistureContent and Soil Salinity Conditions Using Artificial NeuralNetwork Models

Citation for published version:Qi, Y, Huo, Z, Feng, S, Adeloye, AJ & Dai, X 2018, 'Prediction of Consumptive Use Under Different SoilMoisture Content and Soil Salinity Conditions Using Artificial Neural Network Models', Irrigation andDrainage, vol. 67, no. 4, pp. 615-624. https://doi.org/10.1002/ird.2270

Digital Object Identifier (DOI):10.1002/ird.2270

Link:Link to publication record in Heriot-Watt Research Portal

Document Version:Peer reviewed version

Published In:Irrigation and Drainage

Publisher Rights Statement:This is the peer reviewed version of the following article: Qi, Y., Huo, Z., Feng, S., Adeloye, A. J., and Dai, X.(2018) Prediction of Consumptive Use Under Different Soil Moisture Content and Soil Salinity Conditions UsingArtificial Neural Network Models. Irrig. and Drain., 67: 615–624, which has been published in final form athttps://doi.org/10.1002/ird.2270. This article may be used for non-commercial purposes in accordance with WileyTerms and Conditions for Use of Self-Archived Versions.

General rightsCopyright for the publications made accessible via Heriot-Watt Research Portal is retained by the author(s) and /or other copyright owners and it is a condition of accessing these publications that users recognise and abide bythe legal requirements associated with these rights.

Take down policyHeriot-Watt University has made every reasonable effort to ensure that the content in Heriot-Watt ResearchPortal complies with UK legislation. If you believe that the public display of this file breaches copyright pleasecontact open.access@hw.ac.uk providing details, and we will remove access to the work immediately andinvestigate your claim.

Download date: 01. Mar. 2021

1

PREDICTION OF CONSUMPTIVE USE UNDER DIFFERENT SOIL

MOISTURE CONTENT AND SOIL SALINITY CONDITIONS USING

ARTIFICIAL NEURAL NETWORK MODELS†

YANBING QI1, 2, ZAILIN HUO1, SHAOYUAN FENG3 , ADEBAYO J. ADELOYE4, AND

XIAOQIN DAI5

1Centre for Agricultural Water Research in China, China Agricultural University, Beijing, PR China

2Beijing Water Science &Technology Institute, Beijing, PR China

3School of Hydraulic, Energy and Power Engineering, Yangzhou University, Yangzhou, Jiangsu, PR

China

4Institute for Infrastructure and Environment, Heriot-Watt University, Edinburgh, United Kingdom

5Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Science, Beijing,

PR China

ABSTRACT

Response of water use of crop to soil moisture and salinity is complex to quantify using traditional

field experiments. Based on field experimental data for two years, artificial neural networks

models with five inputs including soil moisture content, total salt content, plant height, leaf area

index and crop reference evapotranspiration (ET0) were developed to estimate daily actual

† Prédiction de l’utilisation à des fins de consommation dans différentes conditions de teneur en humidité

du sol et de salinité du sol à l’aide de modèles de réseaux neuronaux artificiels

Correspondence to: Ass. Prof. Zailin Huo. Centre for Agricultural Water Research in China, China

Agricultural University. No.17 Qinghua East Road, Haidian, Beijing, 100083, PR China. Tel.: +86 10

62736762, Fax: +86 10 62736762. E-mail: huozl@cau.edu.cn.

2

evapotranspiration (ET). The models were later used to simulate the response of crop water

consumption to soil moisture and salinity stresses at different growth stages. The results showed

that the ANN model has a high precision with root mean squared error of 0.41 and 0.52 mm day-

1, relative error of 19.6 and 25.6%, and coefficient of determination of 0.87 and 0.79 for training

and testing samples, respectively. Furthermore, the simulation results showed that the seed corn

ET is sensitive to soil salt stress at all the growth stages, although the salinity threshold at which

the impact becomes felt and the extent of the impact vary for the different growth stages, with the

booting and tasseling stages being the most robust. The study offers a more direct approach of

evaluating actual crop evapotranspiration considering explicitly water and salinity stresses.

KEY WORDS: crop water consumption; soil moisture; salinity; artificial neural network.

RÉSUMÉ

La réponse de l’utilisation de l’eau de la culture à l’humidité et à la salinité du sol est complexe à

quantifier en utilisant des expériences de terrain traditionnelles. Basé sur des données

expérimentales de terrain depuis deux ans, des modèles de réseaux neuronaux artificiels avec cinq

entrées incluant l’humidité du sol, la teneur totale en sel, la hauteur des plantes, l’indice foliaire

et l’évapotranspiration de référence (ET0) ont été développés pour estimer l’évapotranspiration

(ET) réelle quotidienne. Les modèles ont ensuite été utilisés pour simuler la réponse de la

consommation d’eau des cultures aux contraintes d’humidité et de salinité du sol à différents

stades de croissance. Les résultats ont montré que le modèle ANN a une grande précision avec

une erreur quadratique moyenne de 0,41 et 0,52 mm jour-1, une erreur relative de 19,6 et 25,6%

et un coefficient de détermination de 0,87 et 0,79 respectivement pour les échantillons

d’entraînement et d’essai. En outre, les résultats de la simulation ont montré que le maïs de

semence ET est sensible au stress salin du sol à tous les stades de croissance, bien que le seuil de

salinité auquel l’impact se fait sentir et l’ampleur de l’impact varient pour les différents stades de

3

croissance, les stades de démarrage et floraison étant les plus robustes. L’étude offre une approche

plus directe de l’évaluation de l’évapotranspiration des cultures en tenant compte explicitement

des contraintes d’eau et de salinité.

MOTS CLÉS: consommation d’eau des cultures; l’humidité du sol; salinité; réseau neuronal

artificiel.

INTRODUCTION

Crop evapotranspiration (ET) is an important process of farmland hydrologic cycle and water

balance (Kit et al., 2000; Allen et al., 2011) and accurate estimation of the ET is the basis of

water-saving practices during irrigation. In the past several decades, many measurement and

calculation methods have been developed to estimate ET. Direct measurement methods of ET

include Lysimeter, eddy covariance and the Bowen ratio approach; the calculation methods

include water balance, aerodynamic, energy budget, combination of aerodynamic and energy

budget, remote sensing, and a large number of some empirical formula methods. The ET is

controlled by various factors, in which soil environment including soil water and salt is important;

consequently, any effort to understand response of crop ET to these soil environment factors is

useful for crop production.

Nowadays, saline water has become an important irrigation water resources as the shortage

of fresh water resources for sustainable agricultural production intensifies (Ai et al., 2007; Geerts

et al., 2009; Ould et al., 2010; Shahbaz et al., 2011). Note: I have put the references in

chronological order to better reflect the development of the science Water and salt stresses that

are inevitable in deficit irrigation employing saline water have an impact on ET and crop yields

(Jiang et al., 2012). The response of ET to soil moisture and salt is complex and nonlinear (FAO-

56). Furthermore, the sensitivity of crop to water and salt stresses depends on the stage of growth

of crops. The FAO methodology for estimating crop actual ET (Allen et al., 1998) recommends

4

the use of coefficients to account for water and salinity stresses when estimating ET from the

reference crop evapotranspiration (ET0). However, determining these coefficients for different

stages of crop growth is difficult and complicated thus limiting the usefulness of the approach.

Consequently, a procedure that can determine the ET directly from locally observed weather and

other data without the need for the endless calibration required for the crop coefficient approach

will be useful.

Artificial neural network (ANN) methodology is a modelling and simulation tool, which is

specially designed for dynamic nonlinear systems. One of the most important traits of ANN

models is their ability to adapt to recurrent changes and detect patterns in complex natural systems.

The neural networks approach has been applied to diverse fields and many studies have approved

the successful application of ANN in different hydrological and water resources investigations

such as rainfall runoff modelling (Hsu et al., 1995; Tokar and Johnson 1999), stream flow

prediction (Imrie et al., 2000), reservoir and water resources planning (Adeloye and De Munari,

2006; Adeloye, 2009), reservoir inflow forecasting (Jain et al., 1999; El-Shafie et al., 2007, 2009;

Sulaiman et al., 2011) and prediction of water quality parameters (Maier and Dandy, 1999;

Rustum and Adeloye, 2007). In addition, Artificial Neural networks have been widely applied to

predict crop yield. Kaul et al. (2005), Alvarez (2009) and Dai et al. (2011) predicted crop yield

using ANN and obtained good results. Recently, artificial neural networks have also been

successfully used in modelling the ET0 (Kumar et al., 2002; Trajkovic et al., 2003; Sudheer et al.,

2003; Zanetti et al., 2007; Jain et al., 2008; Dai et al., 2009; Liu et al., 2009; Adeloye et al., 2012;

Huo et al., 2012). However, the ANN approach rarely has been used to model ET responses to

soil environments, especially in relation to water and salt stresses.

The objectives of this study are: i) to develop ANN models to calculate daily actual ET

under water and salt stress with data from saline water and deficit irrigation field experiments; ii)

to use the developed ANN models to simulate the response of daily ET to different soil moisture

and saline environments.

5

MATERIALS AND METHODS

Field experiment

The field irrigation experiment with saline water was conducted at the Experimental Station

for Water-Saving in Agriculture and Ecology of the China Agricultural University (102°52′ E,

37°52′ N, and elevation 1557.5 m), located in Gansu Province, Northwest China. The study area

lies within the middle reaches of the Shiyang River Basin and is characterized by typical arid zone

climate with deep groundwater table. Annual sunshine duration is over 3000 hours, average

annual precipitation is around 160 mm, and open water evaporation is around 2000 mm. These

cause significant soil moisture deficits without irrigation.

Field experiments on seed corn were set up from 2012 to 2013. In total, there were 26

elementary micro-plots, each with area of 6.66 m2 (3.33 m × 2 m) and depth of 3 m. Basic physical

and chemical properties of tested soil are listed in Table I. Based on estimated crop water

requirement (ETc), deficit irrigation set at 1, 2/3 and 1/2 ETc (w1, w2 and w3) in combination

with salt concentration of 0.71, 3, 6 and 9 g l−1 (s1, s2, s3 and s4) were applied in the experiment.

Four salinity treatments (s1 – s4) for the full irrigation scenario (w1) and only three salinity levels

(s1 - s3) for each of the two deficit irrigations scenarios (w2 and w3) were conducted in the field

experiment. In total, there were ten irrigation treatments in the experiments. Irrigation was applied

five times according to local farming customs. Fresh water with salt concentration of 0.71 g l-1

was obtained from the local well, saline water with salt concentration of 3, 6 and 9 g l−1 were

obtained by adding NaCl, MgSO4, and CaSO4 (a mass ratio of 2:2:1) into fresh water. The detailed

irrigation treatments are listed in Table II.

Seed corn was planted on April 24, 2012 and harvested on September 23, 2012. In 2013,

planting was done on April 20, 2013 and harvested on September 13, 2013. Cultural practices,

such as fertilization, pest control, harrowing, etc. were implemented as necessary in accordance

with local practice.

Soil samples were obtained from six depths (0-10, 10-20, 20-40, 40-60, 60-80, and 80-100

cm). Each soil sample was divided into two parts, with one part being used for soil moisture

6

determination and the other part for measuring soil salinity. The soil sample for soil salinity

determination weighed 10 g after air-dried following which the electrical conductivity of 1:5 soil–

water extract (EC1:5) was determined. The electrical conductivity was converted to total salt

content using (Jiang et al., 2010).

𝑆 = 0.0275𝐸𝐶1:5 + 0.1366 (1)

Where S is the salinity g kg-1; EC1:5 is the electrical conductivity (μs cm-1).

The crop height and size of leaf area were measured once with rule every 10-15 days and

leaf area index were calculated according to the corn leaf shape. For every plot, five corn were

selected to measure the height and leaf area.

Daily ET calculation

In this study, the average daily ET of seed corn at seedling stage, jointing stage, booting

stage, tasseling stage and filling stage was calculated using the water balance equation:

𝐸𝑇𝑎 = 𝑃0 + 𝐼 − 𝑊 − 𝐷 − 𝑅 (2)

𝐸𝑇(𝑚𝑚 𝑑𝑎𝑦−1) = 𝐸𝑇𝑎

𝑑 (3)

Where ETa is the cumulative crop evapotranspiration for a growth stage (mm); W is the soil

water depletion in the measured soil depth during the growth stage (mm); R is the surface runoff

during the stage (mm); D is the bottom water flux, i.e. deep percolation losses during the stage

(mm); P0 is the effective rainfall during the stage (mm) and here the rainfall less than 3 mm were

considered as invalid; I is the irrigation amount during the stage (mm); and ET id the average

daily evapotranspiration during the stage (mm). Soil water depletion (W) was calculated as

change in soil water storage in the 0–100 cm soil profile from the beginning and ending of a

certain growth stage. Surface runoff (R) was assumed to be zero because flooding was not adopted.

7

Bottom water flux (D) was estimated according to Darcy’s equation (Azevedo et al., 2003; Ma et

al., 2011), and it was positive when downward and negative when upward. Finally, the daily ET

was estimated according to equation (3).

All the terms on the right hand side of equation (2) apart from the deep percolation (D) are

known or can be measured with ease. In the study, D was estimated using Darcy’s law as:

𝑞 = −𝐾𝑠𝜕𝜑

𝜕𝐿 (4)

𝐷 = 𝑞∆𝑡 (5)

Where q is the vertical water flux (mm day−1); t is the number of days of the certain growth stage

(day); Ks is the unsaturated hydraulic conductivity f (mm day−1). The unsaturated hydraulic

conductivity (mm day−1) is related to the saturated hydraulic conductivity using can be described

by the following relations:

𝐾𝑠 = 𝐾𝑠𝑎𝑡𝑆𝑒𝜆 [1 − (1 − 𝑆𝑒𝑛

𝑛−1⁄ )𝑛−1/𝑛

]2

(6)

𝑆𝑒 = 𝜃−𝜃𝑟𝑒𝑠

𝜃𝑠𝑎𝑡−𝜃𝑟𝑒𝑠 (7)

Where Ksat is the saturated conductivity (0.921cm day-1); λ is a shape parameter (0.5); Se is the

relative saturation; is the saturated water content (0.385 cm3 cm−3); is the residual water

content in the very dry range (0.05 cm3 cm−3) and n = 1.5486.

Artificial neural networks

The artificial neural network (ANN) algorithm selected is the back-propagation neural

network (BP). The BP is a supervised learning technique used for training the network and is

enough robust for the simulation of any non-linear system, The neural network structure in this

sat res

8

study possessed three layers learning network consisting of an input layer, a hidden layer and an

output layer as shown in Figure 1.

In Figure 1, the three layers are labelled as L−1, Land L+1, respectively, with the

interconnection weights Wij and Wjk between the neurons from adjacent layers. The activation

value ajLat jth neuron in Lth layer can be expressed as:

𝑎𝑗𝐿 = ∑ 𝑊𝑖𝑗

𝑛𝑖 = 1 𝑃𝑖

𝐿−1 + 𝑏𝑗𝐿 , 𝑖 = 1,2,3, … , 𝑛, 𝑗 = 1,2,3, … , 𝑚 (8)

Where Wij is the interconnection weight between the jth neuron in Lth layer and the ith neuron in

(L−1) layer, 𝑃iL−1is the output of the ith neuron in the (L−1)th layer, 𝑏j

Lis the bias of the jth

neuron in the Lth layer.

The activation value of a neuron was used to obtain its output value through a transfer

function. The network transfer function adopted was the sigmoid which is one of the most

commonly used transfer functions. This function takes the input and squashes the output into the

range 0–1. The function value of each neuron in the output layer was obtained by the input effect

propagating layer by layer. The network is trained by a back-propagation algorithm and conjugate

gradient learning algorithms, which adjusts the weights and biases to minimize the error.

Evaluation criterion of model

In this study, three statistical parameters were used to quantify the deviation in simulated

results of ANN. The three statistical parameters is root mean squared error (RMSE), relative error

(RE) and coefficient of determination (R2) which were defined respectively

𝑅𝑀𝑆𝐸 = √1

𝑁∑ (𝑃𝑖 − 𝑂𝑖)2𝑁

𝑖 = 1 (9)

𝑅𝐸 = 1

𝑁(∑ |

𝑃𝑖−𝑂𝑖

𝑂𝑖|𝑁

𝑖 = 1 (10)

9

𝑅 = ∑(𝑃𝑖−�̅�)(𝑂𝑖−�̅�)

√∑(𝑃𝑖−�̅�) ∑(𝑂𝑖−�̅�) (11)

Where Pi and Oi are the simulated and measured daily ET, respectively,�̅�and �̅�the average values

of the data arrays of Pi and Oi, respectively, and N is the observed sample number.

RESULTS

Development of Artificial neural network model

Input and output variables

The most important step in the ANN development process is to determine input variables.

ET depends on external environmental factors and internal plant factors that influence water

demand. Internal factors that affect ET refer to the crop growth stages. Weather conditions

(including solar radiation, temperature, sunlight, wind speed and humidity, etc.) and soil

conditions (including soil moisture, soil texture, structure, soil salinity and groundwater table) are

external factors. Therefore, the study of crop water demand process must consider the impact of

each of these factors on the process of ET. Relationship ET and influencing factors can be

expressed as:

ET = f0 (M，S，B) (12)

Where ET is crop evapotranspiration; M, S, B are weather conditions, soil conditions and crop

characteristics respectively.

For crop characteristics, in the study, the main consideration is plant height and leaf area

index at seedling stage, jointing stage, booting stage, tasseling stage and filling stage. For soil

conditions, soil moisture and total salt content have a significant impact on ET, and water and salt

stress at different growth stages have different effects on ET in saline water irrigation. Soil

10

moisture content, total salt content of 1m soil depths at the beginning and ending of every growth

stages were respectively averaged as the soil moisture content and total salt content of its

corresponding stages. The impact of weather conditions on ET is the combined effects of solar

radiation, atmospheric temperature, humidity and wind speed, which may be measured by ET0.

Much input can make the network complex and ET0 can be treated as one of inputs including the

climate information. Therefore, in this study, the average soil moisture, the total salt content, plant

height, leaf area index, and ET0 in seed corn’s five different stages are identified as the main

factors on ET, simultaneously as five input variables of an artificial neural network. The output

of the ANN was seeding corn daily ET calculated by the water balance equation and Darcy’s law.

ET0 was calculated with FAO-PM method, the FAO-PM equation may be written as:

𝐸𝑇0 = 0.408∆(𝑅𝑛−𝐺)+𝛾

900

𝑇+274𝑈2(𝑒𝑎−𝑒𝑑)

∆+𝛾(1+0.34𝑈2) (13)

Where ET0 is crop reference evapotranspiration (mm day−1); Rn is the net radiation at the crop

surface (MJm-2 day-1), G is the soil heat flux density (MJm-2 day-1),T is the air temperature at 2 m

height (°C), U2is the wind speed at 2 m height (m s-1), ed is the vapour pressure of the air at

saturation (kPa), ea is the actual vapour pressure (kPa), D is the slope of the vapour pressure curve

(kPa°C-1) and g is the psychrometric constant (kPa°C-1). All these factor were recorded with a

weather tower in the experimental station. Mean daily ET0 calculated by P-M method at seedling

stage, jointing stage, booting stage, tasseling stage and filling stage were regarded as input

variables of the ANN. The ranges of inputs variables are listed in Table III.

Training and testing result of ANN model

The aim of this study is to develop an ANN model for crop ET at different growth stages.

In this study, the input factor is the average soil moisture, the total salt content, plant height, leaf

area index, and ET0 for the various growth stages of the crop, the output is daily ET. Total 234

samples were trained and simulated in Matlab7.0.To facilitate training and testing of the neural

11

networks, 2/3 of the collected data were used for training purpose and the remaining1/3 of data

were tested of the trained ANN. The numbers of training epochs given to the net were a minimum

of 500 and a maximum of 5000.

The number of neurons in the hidden layer was determined by trial and error. The five

stages of data were imported into artificial neural network model in sequence. For training and

testing results from ANNs, the RMSE, RE, and R with different node numbers of the hidden

layers were compared to select the optimal node numbers of the hidden layers (Table IV). For the

training data，when the node number was more or less than five, the RMSE and RE decreased

and R increased. However，for the testing data, the ANN with the node of five produced the

lowest RMSE and RE and the highest R2, and when the node number was more than five, the

RMSE and RE increased and R2 decreased. More neurons in the hidden layer make the data

overfitted and increase the ANN training time, so the optimal node numbers of the hidden layers

was 5.

After training of ANN, the RMSE range of the training sample was from 0.0009 to 1.27mm

day-1, and the average RMSE and RE were 0.41 mm day-1, 19.6%, respectively. The testing

sample’s RMSE is between 0.016 mm day-1 and 1.25 mm day-1, the average RMSE and RE were

0.52 mm day-1, 25.6%, respectively. The correlation coefficient R of the training sample and

testing sample was 0.87and 0.79, respectively (Figure 2). The results show that differences

between simulated and measured values is small and the model has a high reliability and accuracy.

Simulation of response of daily ET under water and salt stress at different growth stages

The developed ANN was used to simulate response of daily ET to soil moisture and salinity

stresses at the different growth stages. In the simulation, soil moisture content was set alternatively

at 30, 25 and 20%, while the total salt content was varied between 0.1 and 2 g kg-1(step of 0.1 g

kg-1) .Plant height, leaf area index and ET0 data are the average value for every growth stages

from the experiment and are listed in Table V.

The simulated results are shown in Figure 3 and reveal that after a threshold salinity level,

the daily ET decreased as the total salt content increase for a given soil moisture (Figure 3). The

12

threshold salinity concentrations were 1.0, 0.6, 1.5, 1.4, 0.5 g kg-1 at seedling stage, jointing stage,

booting stage, tasseling stage and filling stage, respectively. Prior to the threshold being reached,

the daily ET remains largely unaffected by increasing soil salinity. However, when the total salt

content exceeds the threshold, namely daily ET was affected by salt stress, daily ET begin to

linearly decrease with the increase of total salt content. Furthermore, for the same total salt content,

daily ET increased with soil moisture content increase. As a result, increasing soil moisture can

compensate for the adverse effects of salt stress on crop’s ET.

Furthermore, daily ET appears more sensitive to salt stress at the seedling, jointing and

filling stage of seed corn, but less so for the booting and tasseling stages. For example, when the

total salt content increased from 0.1 to 2 g kg-1, daily ET for the seedling stage decreased from

4.14 to 0.99 mm day-1at the 30% soil moisture condition, i.e. a reduction of 63%. Conversely, the

corresponding reduction for the booting stage was much less at 43%. As noted earlier, the greatest

reduction in ET occurred after the threshold salinity level for all the growth stages. Thus, although

the recorded reduction in the ET for the booting stage was 43% due at the 30% moisture content

level, the rate of reduction before the threshold salinity of 0.7 g kg-1 was a mere 4%, which became

34% after the threshold. The threshold salinity levels for the other growth stages differ from 0.7

g kg-1 but the rapid decline in the ET after the threshold is reached is a feature common to all the

stages as revealed by Figure 3. In general, moisture stresses appear to accentuate the effect of

salinity on the ET decline for all the growth stages.

Water consumption under water and salt stress at total growth stages

While Figures 3a–e show the effect of salt and water stresses on the respective growth

stages of seed corn, Figure 4 has been produced to show the cumulative effects of these stressors

over the total growth period. As expected, there was declining water consumption with increasing

salinity for all the three soil moisture conditions simulated in the study. As was the case with the

different growth stages, wetter soil conditions appear to temper the ET reduction due to salinity.

This is to be expected in long-term irrigation with high salinity brackish water, which will cause

salt to accumulate in the root zone and the problem will be accentuated for deficit irrigation when

13

there is insufficient water for leaching the salts out of the root zone. This high soil salt content

will bring about osmotic stress, water consumption have been seriously affected, this is mainly

due to the osmotic potential that will hinder root water uptake, leading to less transpiration and

ultimately less yields (Shani, 2007; Ben., 2001).

DISCUSSIONS

It is observed that the ANN was capable to predict the response of crop water consumption to soil

water and salt content with the simple architectures as one hidden layer and small numbers of

neurons. In this study, The ANN were trained with the average soil moisture, the total salt content,

plant height, leaf area index, and ET0 as input and daily ET calculated by the water balance

equation and Darcy’s law as output. The training sample and testing sample’s the average RMSE

and RE were 0.41 mm day-1, 19.6%; 0.52 mm day-1, 25.6 %, respectively. R of the training sample

and testing sample was 0.87and 0.79, respectively. Thus, the ANN was capable of predicting the

response of crop water consumption to soil water and salt content and the model has a high

reliability and accuracy.

The study results showed that the response of daily ET to salt stress is accentuated once a

growth stage dependent threshold concentration was reached. The threshold were 1.0, 0.6, 1.5,

1.4, 0.5 g kg-1 at seedling stage, jointing stage, booting stage, tasseling stage and filling stage,

respectively. Further daily ET is more sensitive to salt stress at the seedling, jointing and filling

stage, whereas the seed corn investigated appear to with stand the salt stress at the booting and

tasseling stages. The result is similar with response of crop yields under water and salt stress.

Mass (Mass et al., 1977) think that a linear relationship with decrease of crop yields caused by

salt stress and soil solution electrical conductance (ECe). Salt tolerance of crops has a threshold

(ECt). When electrical conductance (ECe) exceeds the threshold, crop yields begin to linearly

proportional decrease. Response daily ET to salt stress also presence threshold. When the total

salt content in soil does not reach a certain threshold, daily ET maintained at a potential level,

14

essentially unchanged, but when the total salt content exceeds the threshold, daily ET begin to

linearly proportional decrease with the total salt content increase. In addition, sensitive degree of

ET under water and salt stress in different growth period is different. It is similar with the result

of study of Katerji (Katerji et al., 1992) and Kong et al. (2001). Katerji found crops’ salt

sensitivity exist significant differences in different growth period. Kong et al. considered the basic

phenomenon that crops’ salt sensitivity changes along with the growth stage. Furthermore, it is

worth noting that daily ET increased with soil moisture content increase, increasing soil moisture

was able to compensate for the adverse effects of salt stress on daily ET. This is similar to previous

results (Letey et al., 1985; Bresler et al., 1987) which showed that increased moisture content can

temper the negative effects of salt stress on crop growth and yields.

CONCLUSIONS

In this study, the artificial neural network model was developed to calculate daily ET and simulate

response of crop to soil moisture and salinity in seeding corn of different growth stages. The

model has a high reliability and precision with RMSEs of 0.41 and 0.52 mm day-1, REs of 19.6%

and 25.6%, and Rs of 0.87 and 0.79 for training sample and testing sample, respectively. The

simulated results showed that daily ET decreased with the total salt content increase. Furthermore,

response daily ET to salt stress presence different threshold at each growth stage. When the total

salt content in soil does not reach a certain threshold, Reduce the amount of daily ET is small,

essentially unchanged. But when the total salt content exceeds the threshold, namely daily ET

was affected by salt stress, daily ET begin to linearly proportional decrease with the total salt

content increase. When the soil total salt content is constant, daily ET increased with soil moisture

content increase.

In summary, ANN can be used to estimate daily ET under water and salt stress with easy

measurable soil and crop parameters. The study provides a new way to calculate daily ET, the

accuracy and stability of the model to meet the needs of agricultural production. Furthermore

15

simulation of response of daily ET under water and salt stress at different growth stages contribute

to research ET and water consumption.

ACKNOWLEDGEMENTS

This study was supported by the National Natural Science Foundation of China (91425302,

51679236, 51639009). The contributions of the editor and anonymous reviewers whose

comments and suggestions significantly improved this article are also appreciated.

REFERENCES

Adeloye AJ. 2009.The relative utility of multiple regression and ANN models for rapidly

predicting the capacity of water supply reservoirs. Env. Modelling & Software, 24(10),

1233 – 1240.

Adeloye AJ, De Munari A. 2006. ‘Neural networks based generalised models for reservoir

capacity-yield-reliability behaviour’. Journal of Hydrology, 326, 215-230.

Adeloye AJ, Rustum R, Kariyama ID. 2012. Neural computing of the reference crop

evapotranspiration. Environmental Modelling & Software, 29(1), 61 – 73

Ai MH, Hoque MR, Hassan AA, et alnote: give names and initials of all authors, 2007. Effect

of deficit irrigation on yield, water productivity, and economic returns of wheat.

Agricultural Water Management.92, 151-161.

Allen RG, Pereira LS, Howell TA. 2011a. Evapotranspiration information reporting: I. Factors

governing measurement accuracy. Agricultural Water Management. 98(6), 899-920.

Allen RG, Pereira LS, Howell TA. 2011b. Evapotranspiration information reporting: II.

Recommended documentation. Agricultural Water Management. 98(6), 921-929.

Alvarez R. 2009.Predicting average regional yield and production of wheat in the Argentine

16

Pampas by an artificial neural network approach. European Journal of Agronomy. 30, 70-

77.

Azevedo PV, da Silva BB, da Silva VPR. 2003. Water requirements of irrigated mango orchards

in northeast Brazil. Agricultural Water Management. 58, 241-54.

Eshel Bresler E, Hoffman GJ. 1986. Irrigation management for soil salinity control. Theories and

Tests Soil Science Society American Journal. 50, 1552-1560.

Burt CM, Isbell B, Burt L. 2003. Long-term salinity buildup ondrip/micro irrigated trees in

California. IA Technical Conference. San Diego, USA. 12–15.

Dai X, Shi H, Li Y, Ouyang Z, Huo Z. 2009. Artificial neural network models for estimating

regional reference evapotranspiration based on climate factors. Hydrological Processes. 23,

442–450.

Dai X, Huo Z, Wang H. 2011. Simulation for response of crop yield to soil moisture and salinity

with artificial neural network. Field Crops Research. 121, 441-449.

El-Shafie A, Taha MR, Noureldin A. 2007. A neuro-fuzzy model for inflow forecasting of the

Nile river at Aswan high dam. Water Resources Management. 21(3),533–556.

El-Shafie A, Abdin AE, Noureldin A, note: give names and initials of all authors., 2009.

Enhancing inflow forecasting model at Aswan high damutilizing radial basis neural

network and upstream monitoring stations measurements. Water Resources Management.

23, 2289–2315.

Geerts S, Raes D. 2009.Deficit irrigation as an on-farm strategy to maximize crop water

productivity in dry areas. Agricultural Water Management. 96, 1275–1284.

Hsu K, Gupta HV, Sorooshian S. 1995. Artificial neural network modeling of the rainfall-runoff

process. Water Resources Research. 31, 2517-2530.

Huo Z, Feng S, Kang S, note: give names and initials of all authors. 2012. Artificial neural

network models for reference evapotranspiration in an arid area of northwest China.

Journal of Arid Environments. 82, 81-90

Imrie CE, Durucan S, Korre A. 2000. River flow prediction using artificial neural networks:

Generalization beyond the calibration range. Journal of Hydrology. 233, 138-153.

17

Jain SK, Nayak P, Sudheer K. 2008. Models for estimating evapotranspiration using artificial

neural networks, and their physical interpretation. Hydrological Processes. 22, 2225–2234.

Jain SK, Das A, Srivastava DK. 1999.Application of ANN for reservoir inflow prediction and

operation. Water Resources Planning and Management. 125, 263–271.

Jiang J, Huo Z, Feng S, note: give names and initials of all authors. 2012. Effect of irrigation

amount and water salinity on water consumption and water productivity of spring wheat in

Northwest China. Field Crops Research.137, 78-88.

Jiang J, Feng S, Wang Y, note: give names and initials of all authors. Year. Effect on water-salt

distribution and evapotranspiration of spring maize under different water quantities and

qualities. Scientia Agricultura Sinica, 2010, 43(11):2270－2279. (in Chinese with English

abstract).

Katerji N, Hoorn J, Hamdy A, note: give names and initials of all authors. 1992. Effect of

salinity on water stress, growth and field of Broad beans. Agricultural Water Management.

21, 237-249.

Kaul M, Hill RL, Walthall C. 2004. Artificial neural networks for corn and soybean yield

prediction. Agricultural Systems. 85, 1-18.

Kite GW, Droogers P. 2000.Comparing evapotranspiration estimates from satellites，hydrological

models and field data. Journal of Hydrology. 229, 3-18．

Kong D, Shi H, Chen Y. 2001. Preliminary experiment study on effect of water-salt stress on

sunflower. China agricultural soil and water conservation engineering professional

committee of academic papers. 169-172.

Kumar M, Raghuwanshi N, Singh R, Wallender W, Pruitt W. 2002. Estimating

Evapotranspiration using Artificial Neural Network. Journal of Irrigation and Drainage

Engineering. 128(4), 224–233.

Letey J, Dinar A, Kanpp KC. 1985. Crop-water production model for salineirrigation waters. Soil

Science Society American Journal. 49, 1005-1009.

Liu X, Kang S, Li F. 2009. Simulation of artificial neural network model for trunk sap flow of

Pyruspyrifolia and its comparison with multiple-linear regression. Agricultural Water

18

Management. 96, 939–945.

Ma Y, Feng S, Huo Z, note: give names and initials of all authors. 2011. Application of the

SWAP model to simulate the field water cycle under deficit irrigation in Beijing, China.

Mathematical and Computer Modelling, 54, 1044–1052.

Maier HR, Dandy GC. 1999.Empirical comparison of various methods for training feed-forward

neural networks for salinity forecasting. Water Resources Research. 35, 2591–2596

Mass EV, Hoffman GT. 1997. Crop salt tolerance-current assessment. Journal of the Irrigation

and Drainage Division. 103, 115-134.

Ould ABA, Inoue M, Moritani S. 2010. Effect of saline water irrigation and manure application

on the available water content, soil salinity, and growth of wheat. Agricultural Water

Management. 97, 165-170.

Rustum R, Adeloye AJ. 2007. Replacing outliers and missing values from activated sludge data

using Kohonen Self Organizing Map. J. Env. Engrng, ASCE, 133(9), 909-916.

Shahbaz M, Mahnoosh M. 2011. Evaluating the potentials of deficit irrigation as an adaptive

response to climate and environmental demand. Environmental Science & Policy. 14,1139

－1150.

Shani U, Cudley L. 2001. Field studies of crop response to water and salt stress. Soil Science

Society of American Journal. 65, 1522-1528.

Shigeoki M, Tahei Y, Henintsoa A, Mitsuhiro I, Hirotaka S. 2013. Effect of combined water and

salinity stress factors on evapotranspiration of Sedum kamtschaticum Fischer in relation to

green roof irrigation. Urban Forestry & Urban Greening. 12, 338–343.

Shuttleworth WJ. 1993. Evaporation, In: Maidment D.R, ed., Chapter 4, Handbook of Hydrology,

McGraw Hill. Inc., New York, USA.

Sudheer KP, Gosain AK, Ramasastri KS. 2003. Estimating actual evapotranspiration from limited

climatic data using neural computing technique. Journal of Irrigation and Drainage

Engineering. 129,214–218.

Sulaiman M, El-Shafie A, Karim O, Basri H. 2011. Improved water level forecasting performance

by using optimal steepness coefficients in an artificial neural network. Water Resources

19

Management. 25, 2525–2541.

Tokar AS, Johnson PA. 1999. Rainfall-runoff modeling using artificial neural networks. Journal

of Hydrologic Engineering. 4, 232–239.

Trajkovic S, Todorovic B, Stankovic M. 2003. Forecasting of reference evapotranspiration by

artificial neural networks. Journal of Irrigation and Drainage Engineering. 129, 454–457.

Zanetti S, Sousa E, Oliveira V, note: give names and initials of all authors. 2007. Estimating

Evapotranspiration Using Artificial Neural Network and Minimum Climatological Data.

Journal of Irrigation and Drainage Engineering. 133(2), 83–89.

20

Figure captions

Figure 1. Structure of the neural network used in this study with five input variables, one hidden layer,

and one output variable forestimation of daily ET

Figure 2. Simulated daily ET and the measured values for training samples and testing samples

Figure 3. ET changes with soil saline for different soil moisture at different stages: seedling stage; b)

jointing stage; c) booting stage; d) tasseling stage; e) filling stage

Figure 4. Water consumption of seed corn under water and salt stress over full growth period

21

Table I. Basic physical and chemical properties of tested soil

Soil depth

(cm)

Sand (%) Silt (%) Clay (%)

Organic

matter(g kg-1)

Textural class of

international

system

0-20 59.465 28.586 11.962.0 1.48 Sandy loam

＞20-40 58.33 29.475 11.21 1.50 Sandy loam

＞40-120 43.354 42.63 14.02 1.52 Clay loam

(Note: three significant figures imply already an accuracy of better than one promille,

which you cannot achieve in practice. Please check the whole text and the Tables and

Graphs for not more than three significant figures)

22

Table II. Detailed irrigation schedule of each treatment

Treatment

Irrigation water quota (mm) Irrigation

norm (mm) 2012 2013

6-6 6-30 7-21 8-13 8-30 6-5 6-30 7-20 8-10 8-29 2012 2013

w1(s1、s2、s3、s4) 120 120 120 120 105 120 120 105 105 105 585 555

w2(s1、s2、s3) 80 80 80 80 70 80 80 70 70 70 390 370

w3(s1、s2、s3) 60 60 60 60 52.5 60 60 52.5 52.5 52.5 292.53 277.58

Note: w1, w2, w3 denote 1, 2/3, 1/2 of crop water requirement (ETc); s1, s2, s3, s4 denote irrigation water concentration of 0.71, 3, 6,9 g L−1.w1s1,w1w2, w1s3 with

four replicates each, the other with two replicates each.

23

Table III. Soil moisture, total salt content, plant height, leaf area index, and ET0 ranges in

seeding corn different growth stages from the two year field experiment

Growth stages

seedling jointing booting tasseling filling

Soil moisture (%) Lowest 17.9 19.8 20.8 21.8 21.0

Highest 26.5 29.3 30.8 32.0 33.4

average 21.5 24.2 25.9 27.7 29.0

Total salt content

(g kg-1)

Lowest 0.57 0.61 0.60 0.49 0.56

Highest 1.88 1.97 1.96 1.91 2.20

average 1.30 1.17 1.24 1.31 1.35

plant height (cm) Lowest 12.9 58.3 93.8 98.0 98.1

Highest 28 109.4 160.1 162.7 162.1

average 21.9 82.4 138.6 138.8 136.0

leaf area index Lowest 0.09 0.82 1.31 1.25 1.25

Highest 0.33 2.05 3.22 3.12 3.02

average 0.20 1.40 2.24 2.16 1.99

ET0 (mm day-1) 2013/2012 4.04/4.33 3.45/3.78 3.01/4.28 3.64/4.28 2.64/2.78

24

Tables IV. Error changes with the nodes of the hidden layer for ANN

Nodes 3 4 5 6 7 8 9 10 11 12

Training data

RMSE(mm day-1) 0.64 0.60 0.41 0.45 0.49 0.46 0.44 0.40 0.38 0.36

RE(%) 0.30 0.28 0.20 0.25 0.22 0.21 0.20 0.18 0.18 0.17

R 0.79 0.81 0.87 0.85 0.88 0.89 0.90 0.91 0.93 0.93

Testing data

RMSE(mm day-1) 0.74 0.68 0.52 0.79 0.82 0.83 0.90 0.87 0.95 0.94

RE(%) 0.40 0.35 0.25 0.37 0.39 0.38 0.41 0.41 0.43 0.41

R 0.68 0.69 0.79 0.62 0.61 0.61 0.49 0.63 0.52 0.56

25

Table V. Plant height, leaf area index of corn and ET0 at different growth stages

seedling jointing booting tasseling filling

Plant height（cm） 25 85 130 150 150

leaf area index 0.2 1.4 2.2 2.2 1.8

ET0（mm day-1） 4.50 4.20 4.20 4.20 3

26

L-1 Wij L Wjk L+1

Moisture content

Total salt content

Plant height Daily ET

Leaf area index

ET0

Input Layer Hidden Layer Output Layer

Figure 1

Figure 2

01

23

45

67

0 1 2 3 4 5 6 7

Mod

eled

val

ue（

mm

day

-1）

Observed value（mm day-1）

01

23

45

67

0 1 2 3 4 5 6 7

Mod

eled

val

ue（

mm

day

-1）

Observed value（mm day-1）

27

a b

c d

e

Figure 3

01

23

45

6

0 0.5 1 1.5 2

ET（

mm

day

-1）

Total salt content（g kg-1）

01

23

45

6

0 0.5 1 1.5 2

ET（

mm

day

-1）

Total salt content（g kg-1）

01

23

45

6

0 0.5 1 1.5 2

ET（

mm

day

-1）

Total salt content（g kg-1）

01

23

45

6

0 0.5 1 1.5 2

ET（

mm

day

-1）

Total salt content（g kg-1）

01

23

45

6

0 0.5 1 1.5 2

ET（

mm

day

-1)

Total salt content（g kg-1）

moisture content 20%

moisture content 25%

moisture content 30%

28

Figure 4

0

100

200

300

400

500

600

700

0 0.5 1 1.5 2 2.5

wat

er c

on

sum

pti

on（

mm）

The total salt content（g kg-1）

moisture content 20%moisture content 25%moisture content 30%