Development and test of SWAT for modelling hydrological process in irrigation disticts with paddy rice

Georgakakos, Editor-in-Chief, with the assistance of Michael Brian Butts, Associate Editor Keywords: Model development SWAT Hydrological process Irrigation district Paddy rice s u m m a

Development and test of SWAT for modeling hydrological processes

in irrigation districts with paddy rice

Xianhong Xiea,b,⇑, Yuanlai Cuia

a

State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan 430072, China

b

College of Global Change and Earth System Science, Beijing Normal University, Beijing 100875, China

Article history:

Received 9 December 2008

Received in revised form 7 May 2010

Accepted 22 October 2010

This manuscript was handled by K.

Georgakakos, Editor-in-Chief, with the

assistance of Michael Brian Butts, Associate

Editor

Keywords:

Model development

SWAT

Hydrological process

Irrigation district

Paddy rice

s u m m a r y

The water movement in irrigation districts, especially for paddy rice cultivation, is characterized by com-plicated factors Soil and Water Assessment Tool (SWAT) is a popular tool for understanding the hydro-agronomic processes However, it fails to simulate the hydrological processes and crop yields in paddy rice areas In this study, we develop the SWAT model by incorporating new processes for irrigation and drainage The evapotranspiration process in paddy fields is simulated on the basis of water storage conditions, and a controlling irrigation scheme is introduced to manage the irrigation and drainage oper-ations The irrigation function of local water storages, such as ponds and reservoirs, is extended for these storages in order to provide water in a timely manner to paddy fields Moreover, an agronomic model is incorporated to estimate crop yields when available data sets are not satisfactory The model is tested in Zhanghe Irrigation District, China The simulated runoff matches well to the measurements and the results indicate the developed model is preferable to the original edition of SWAT The estimate of the paddy rice yield is acceptable and the dynamics of water balance components approximately characterize the state of water movements in paddy fields Therefore, the developed framework for SWAT is practical and capable of representing the hydrological processes in this irrigation district Further work is still needed to more broadly test the model in areas with paddy rice cultivation

Ó 2010 Elsevier B.V All rights reserved

1 Introduction

Paddy rice, as a major food crop in China, consumes large

amounts of water for agricultural irrigation It is important to

cre-ate a reasonable framework to evalucre-ate productivity and manage

water resources in irrigation districts where the hydrological cycle

depends not only on natural factors (e.g the evapotranspiration

and precipitation), but also on human activities (e.g the irrigation

and drainage operations) Especially in the paddy rice areas, the

different water bodies (e.g the ponds, reservoirs and paddy fields)

and constructions (e.g the irrigation canals) are highly distributed

Thus, the irrigation district is a human-nature composite

ecosys-tem (Wang and Yang, 2005), and a coupled hydro-agronomic

mod-el is needed to explore the hydrological processes and crop growth

conditions in this kind of area (Luo et al., 2008)

There are a number of sophisticated models able to address

these challenges, such as Soil–Water–Atmosphere–Plant (SWAP,

Van Dam et al., 1997; Kroes et al., 1999), MIKE SHE (Graham and

Butts, 2006), and Soil and Water Assessment Tool (SWAT,Arnold

et al., 1993) The SWAP model simulates vertical water flow, solute transport, heat flow in close interaction with crop growth in agri-cultural fields This permits water productivity analysis and esti-mation of the agricultural water use (Singh et al., 2006; Anuraga

et al., 2006; Utset et al., 2006, 2007; Mandare et al., 2008) How-ever, this model focuses on hydrological processes at the field scale, and it is not suitable for large scale simulations or the areas with great spatial variability Furthermore, the ponding boundary

at the ground surface is not considered in the model So it is better

at modeling upland areas rather than depression areas (e.g the paddy fields) As a fully physically-based hydrological model, MIKE SHE accounts for many hydrological processes and their interac-tions as well as water management practice (DHI, 2007) It is also widely used to simulate the hydrological water balance and investigate the effects of cropping practices in irrigation districts (Jayatilaka et al., 1998; Singh et al., 1999; Islam et al., 2006), and evaluate sustainable groundwater management options (

Demetri-ou and Punthakey, 1999) However, its performance depends highly on detailed information and abundant data sets from the area of interest, and scaling issues of parameters and variables are great challenges (Xiong and Guo, 2004)

While SWAT is a basin scale, physically-based continuous dis-tributed model developed to predict impacts of management on

0022-1694/$ - see front matter Ó 2010 Elsevier B.V All rights reserved.

Chengfu Road, Haidian District, Beijing, China Tel.: +86 10 58809071; fax: +86 10

82887918.

Contents lists available atScienceDirect Journal of Hydrology

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / j h y d r o l

water, sediment, and agricultural chemical yields in ungauged

watersheds (Neitsch et al., 2001) It allows for relatively complete

agricultural management practices (e.g planting, fertilization,

irri-gation and drainage) and spatial distributed characteristics (e.g

ponds, reservoirs) in irrigation areas So SWAT is considered as a

preferred tool for the agricultural watershed modeling in this

study

Recently, there have been a few studies concerning hydrological

processes based on SWAT in irrigation areas.Ritschard et al (1999)

used SWAT to estimate the irrigation water requirements and

monthly runoff in the Gulf Coast of the United States Their results

showed the capability of SWAT to deal with large scale problems

Bosch et al (2004)evaluated the SWAT model on a coastal plain

agricultural watershed, and suggested that a modification and

more extensive calibration may be necessary to increase the

accu-racy of the daily flow estimation.Behera and Panda (2006)

identi-fied the critical areas of an agricultural watershed and

recommended the best management practices using SWAT, and

their works revealed the robust performance of the model in

differ-ent simulation conditions Since data scarcity is a common

prob-lem in hydrological modeling, Immerzeel and Droogers (2008)

integrated remote sensing and observed monthly discharge to

cal-ibrate the SWAT model In addition,Luo et al (2008)assessed the

crop growth and soil water modules in SWAT2000 based on field

experiments in an irrigation district of the Yellow River Basin (in

China), and they proposed some improvements to soil water and

groundwater evaporation modules There are other studies

con-cerning the application of SWAT Comprehensive reviews on SWAT

model were given byGassman et al (2007), and Krysanova and

Arnold (2008)

When the current SWAT model is used in irrigation districts

with depressions or impounded areas, e.g the paddy fields,

how-ever, it may cause some bias because it is not able to completely

represent the characteristics of fields and address complicated

water management practice

In this study, we focus on the simulation of hydrological

pro-cesses in paddy rice areas and propose developments to the

cur-rent SWAT framework We first give an overview of SWAT

mainly on the hydrological cycle in paddy fields and comment on

its weaknesses In Section3, the developed components are

illus-trated, including the evapotranspiration, the processes of irrigation

and drainage, and the crop yield estimation In Section4, the

devel-oped model is examined in Zhanghe Irrigation District (ZID), in

China The runoff and crop yields are calibrated and validated and

the water balance in paddy fields is evaluated In Section5, a

discus-sion is presented and finally concludiscus-sions are given for this study

2 Model description

2.1 Structure of land phase processes

SWAT simulates major hydrological components and their

interactions as simply and yet realistically as possible (Arnold

et al., 1993) To realize this capability, a sequential structure

com-bined with thirteen routing commands is used to simulate

hydro-logical processes occurring within hydrologic response units

(HRUs) and subbasins and to route stream loadings through the

channel network in a watershed

The first and important loop run is a subbasin command in

which the land phase process of the hydrological cycle is

simu-lated, including surface and subsurface runoff generation, snow fall

and melt, vadose zone processes (i.e infiltration, evaporation,

lat-eral flows), crop growth and water quality transformation Paddy

fields in a subbasin are aggregated and treated as a pothole, like

an impounded or depression area

Fig 1shows the flowchart on the land phase process We should emphasize that the simulation scheme of the model distinguishes between different kinds of land cover (1) If an HRU is covered with water, only evaporation from the water body is simulated with the Priestly–Taylor equation and no water from this HRU will contrib-ute to stream flows In fact, the water movement in this kind of HRU can be better represented as processes in ponds, wetlands

or reservoirs (Neitsch et al., 2001) (2) If an HRU in a released state

is covered with general lands (pot_vol <e), then the surface runoff

is estimated by the curve number technique or the Green–Ampt method, and the actual evaporation from soil water is computed Here, paddy fields are treated as upland areas (3) If the HRU con-tains impounded potholes in which water is stored, then the sur-face runoff and the actual evaporation from soil profile are excluded, while the water routings in potholes, such as inflow, evaporation from water body and seepage, are taken into account

in the pothole procedure

Therefore, SWAT is a comprehensive and reasonable model that

is suitable for most of conditions in irrigation districts

2.2 Main components Three main components with respect to growth environment of the paddy rice are further described here

2.2.1 Water routing in potholes

A pothole, originally meaning a deep and round hole or a pit, is a depression that can receive a part of surface runoff from the related HRUs SWAT assumes the paddy rice could grow in this area Accordingly, the paddy field in this kind of HRU is assumed to be

a cone shape (Fig 2), and its surface area of water body is varied with the depth or the volume of water storages (Neitsch et al.,

2001)

The water balance components in potholes contain precipita-tion, irrigaprecipita-tion, surface runoff concentraprecipita-tion, evaporation from water body, seepage and outflow Since the pothole is character-ized with a cone shape, the volume of precipitation depends on the surface area of water body as well as precipitation intensity Irrigation water applied to a pothole is obtained from one of the five types of water source: a reach, a reservoir, a shallow aquifer,

a deep aquifer or an outside source Water can be removed from potholes to stream reaches through three different routes; over-flows, release operations and tile drainages

However, these representations are appropriate for general closed depression areas rather than real-world paddy fields (Fig 3) First, when an HRU containing potholes is impounded, the surface runoff from the non-pothole part is not considered (Fig 1) Even though the other pothole processes concerning paddy fields are rep-resented, this framework still results in underestimation of the sur-face runoff loading to main channels Second, the sursur-face area of the water body is a fluctuating value which varies with the volume of water stored in the impounded pothole In contrast, the actual paddy fields are characterized by a large number of plots or fields and sep-arated by low embankments that retain ponding water on the soil surface (Kang et al., 2006) Thus the areas of paddy fields remain approximately constant in the whole process This inconsistency can also underestimate the surface area of paddy fields which influ-ences the subsequent hydrological processes Furthermore, in a large irrigation area it is difficult to specify a reasonable value for the frac-tion of HRU area that drains into the related pothole

2.2.2 Water routing in ponds Ponds are water bodies located within a subbasin that received inflow from a fraction of the subbasin area It is assumed in SWAT that ponds are uniformly distributed in each HRU in a subbasin In addition to general components of water balance in ponds, the

consumptive water use item is also considered to estimate the

irri-gation for crops outside the watershed or removal of water for

ur-ban/industrial use However, the irrigation function from ponds for

local crop fields is not taken into account, which will limit model

applications to water management scenarios, such as the real-time

irrigation and drainage based on the local source of water (Guo,

1997)

2.2.3 Crop growth

SWAT incorporates a simplified version of the

Erosion-Produc-tivity Impact Calculator (EPIC) plant growth model In this model,

the phenological plant development is based on daily accumulated

heat units, potential biomass is based on a method developed by

Monteith (Monteith, 1977), a harvest index is used to calculate

yield, and plant growth can be inhibited by temperature, water,

nitrogen and phosphorus stress (Neitsch et al., 2001) When

applied to large areas, it still suffers from scarce data availability, such as fertilization and pesticide information

3 Model development 3.1 Evapotranspiration process in paddy fields

As the paddy field is a plot separated by low embankments that occupy only a very small proportion of the total area of the field, it

is reasonably to assume that the area of paddy fields is equal to the area of the HRU:

where SA is the field surface area (ha); AHRUis the area of the HRU whose land cover is the paddy rice (ha) Note that here the paddy field has a cuboid shape with a constant surface area rather than

Beginning for

an HRU

Initialization for variables

Is water body ?

Pot_vol < ε

and released ?

Soil water routing

ET0calculation

Pot_vol > ε?

Actual ET simulation ( E sand E canare calculated )

Groundwater routing

Is the pothole contained

Evaporation calculation

Is the pond Contained ?

Water balance calculation in ponds

Is the Irrigation specified?

Irrigation operation s

Consumptive water uses

Output treatment

Yes

No

Surface runoff generation

Yes

No

Next HRU simulation

Is water impounded ?

Update volume of water in pothole

Pot_vol > 0?

Pothole processes , including inflow , outflow, drainage , infiltration , evaporation and release /impounding operations , etc

No

Yes

Yes

No

No

Yes

Hydrological processes in potholes with the cultivation

of paddy rice

Yes No

No

Crop growth routing (E pis calculated ) Yes

Yes

No

Note: Pot_vol is the volume of water

stored in a pothole ; ε is a infinitesimal ;

ET0, E can, E p, and E s are potential evapotranspiraiton ; evaporation from free water in the canopy , plant transpiration , and evaporation from soil profile, respectively

?

Fig 1 Flowchart on the land phase simulation (only modules regarding agricultural hydrology are expressed).

Area of an HRU

Contributing area for a pothole

SA V

H

Cone shape of the Pothole

Note:SA is the surface area of the water

body, ha; V is its volume, m3

; H is the

depth , m; and the slp is the average slop

of a specified HRU

1

slp

Pothole

Fig 2 Schematic diagram of the area of an HRU (left) and its related pothole with the cone shape (right).

a round hole with a cone shape, which implies that the paddy rice

grows over the entire area of an HRU So the precipitation,

evapora-tion and transpiraevapora-tion can act on the entire land surface

Moreover, in China, a controlling irrigation scheme is a

gener-ally implemented practice in paddy areas in order to save

irriga-tion water and ensure considerable crop yields The water depth

in paddy fields is variable and even approaches zero sometimes

when the paddy field is in dry state Accordingly, two water

storage conditions are defined to calculate the actual

evapo-transpiration

(1) If the paddy field is in dry state (pot_vol <e) and the HRU is

not an impounded area or a drained area, then

(1) If the paddy field is in a wet state and the HRU is impounded,

then

where ETactis the actual amount of evapotranspiration occurring in

an HRU on a given day (mm, H2O); Ecanis the amount of evaporation

from free water in the canopy on a given day (mm, H2O); Epis the

amount of plant transpiration on a given day (mm, H2O); and Es, Epot

are the water evaporation from the soil profile and the water body

surface respectively (mm, H2O)

For the first condition, the equation defines a general state that

the land surface is exposed with no water stored in potholes or

fields, thus the evaporated water is from the soil water This kind

of HRU could be fallow fields, or paddy fields in dry-state periods

For example, in the final tillering stage, the fields should be kept at

a dry state via drainage operations in order to control useless tillers

of the paddy rice and improve the aerating and temperature

condi-tions (seeTable 1) This operation is so-called sun drying of the

paddy field (Li et al., 2003) For the second condition, the fields

are impounded and water is stored, so the evaporation (Epot) is

from the water body instead from the soil profile

3.2 Framework of irrigation and drainage controlling

A good controlling scheme for irrigation and drainage at the field scale should not only provide right moisture conditions to favor crop growth, but also save water and minimize water transfer Moreover, the local source of water in ponds or pools distributed in irrigation areas, could be conveniently used for crop irrigation to reduce water transfers from other sources that may be outside the agricultural watershed Therefore, controlling schemes for irrigation and drain-age and the utilization of local water are widely adopted practices for agricultural water management to save water transfers They also change the route of runoff in an agricultural watershed 3.2.1 Irrigation and drainage for paddy fields

In order to create favorable conditions with appropriate mois-ture, ventilation and temperature during the growth period, it is usual to design a scheme to regulate water depths through irriga-tion and drainage at different growth stages of the paddy rice.Guo (1997)introduced a technique with three critical depths, namely the minimum fitting depth (hmin), the maximum fitting depth (hmax) and the maximum ponding depth (Hp) As shown in the Fig 3, with water being depleted in fields (e.g evapotranspiration and seepage), the depth could reach a minimum fitting value, hmin, then the moisture conditions may threaten the paddy rice Subse-quently the irrigation operation is requested and implemented un-til the depth reaches a maximum fitting value, hmax On the other hand, if significant precipitation occurs during the stage, the water depth should be controlled under a maximum value Hpvia the drainage operation This technique is widely used as it is simple but effective for farmers Therefore, it is important to design a rea-sonable scheme for the three controlling depths (hmin hmax Hp)

in different growing stages for the paddy rice This is beyond the scope of this paper, and the reader is referred to Guo (1997), Anbumozhi et al (1998)andChi et al (2001)

As mentioned before, in the SWAT model, the daily water bal-ance equation can be updated as follows:

where ST is the water depth in fields (mm, H2O); P is the daily pre-cipitation (mm, H2O); IR is the irrigation depth (mm, H2O); DR is the drainage depth (mm, H2O); ET is the evapotranspiration (mm, H2O); and SP is the seepage (mm, H2O) The subscript i denotes day i The evapotranspiration of paddy fields is computed withEq (2) While the water lost by seepage through the bottom of paddy fields on a given day is calculated as a function of the water content of the soil profile beneath the pothole (Neitsch et al., 2001)

The irrigation depth or volume is represented as:

where hi,max and hi,min are the maximum and minimum fitting depth respectively (mm, H2O) Similarly, the drainage depth (DR)

is written as

Paddy rice

H p

h max

h min

Rainfall

Irrigation

Drainage Irrigation

canal

Seepage Root layer

E pot or E s

E can and E p

Q l

denotes the crop transpiration.

Table 1

Three critical depths of paddy fields in different growing stages in ZID with the intermittent irrigation technique.

DRi¼ STi Hi;p if STi>Hi;p ð5aÞ

where Hi,pis the maximum water depth at the day i (mm, H2O)

3.2.2 Irrigation water from ponds

Ponds are small reservoirs located in irrigation areas and they

allow farmers to capture rainfall and store surplus water from

other sources (Shahbaz et al., 2007a) that can then provide

irriga-tion water to crops when required In ZID, for example, thousands

of medium- and small-size ponds or reservoirs contribute

one-fourth of the amount of water in paddy fields, and they effectively

reduce the need for water transfers from the main Zhanghe

reser-voir (Shahbaz et al., 2007b)

For these reasons, the real-time irrigation function of ponds

should be extended in hydrological models The water balance

equation is now expressed as:

Vi¼ Vi1þ Vi;pcpþ Vi;flowin Vi;flowout Vi;evap Vi;seep Vi;use Vi;irr

ð6Þ

where Viand Vi1is the volume of water stored in ponds at the end

of the day i and i  1 (m3H2O); Vi,pcpis the volume of precipitation

falling on the water body during the day (m3 H2O); Vi,flowin and

Vi,flowoutare the volume of water entering and leaving the water

body during the day(m3H2O); Vi,evapis the evaporation volume of

water body during the day (m3H2O); Vi,seepis the volume of water

lost from the water body by seepage during the day (m3H2O); Vi,use

is the volume of water used for the urban or industrial requirement

during the day (m3H2O); and Vi,irris the volume of irrigation water

provided for local fields during the day (m3H2O)

The calculation of each item in Eq.(6)refers toNeitsch et al

(2001) With regard to the volume of irrigation water, there should

be water requirements from fields on the one hand and an enough

capacity to provide water from ponds on the other hand This is

specified as:

Vi;irr¼ 10  ðhi;max STiÞ  SA; if STi<hi;min and Vi

Vi;irr¼ Vi; if STi<hi;min and Vi6ðhi;max STiÞ  SA ð7bÞ

where Vi,irris the volume of irrigation water to local fields during

the day i (m3H2O); Viis the volume of water stored in ponds at

the end of the day i (m3H2O); hi,maxand hi,minare the maximum

and minimum fitting depth for crops at the day i respectively

(mm, H2O); STiis the ponding water depth in fields at the end of

the day i (mm, H2O); and SA is the area of paddy fields (ha) Clearly,

Eq (7)is consistent withEq (4)on the irrigation volumes

3.3 Simplified modeling of crop yields

The lack of available data at large scale is a common problem for

the crop growth simulation and it will limit the application of EPIC

model So it is more practical to search for a simplified method to

estimate the crop yields From previous research, crop growth and

yield generation are greatly influenced by the total volume of

evapotranspiration over the whole growth period.Li et al (2003)

found that the relative grain yield is dependent on the relative

evapotranspiration volume with a linear or non-linear relation A

lot of existing studies support such relations (Henry et al., 2007)

In this work, we utilize the linear function proposed byStewart

et al (1975), which can be described as

1 Ya

ETa

ETm

ð8Þ

where Yais the actual crop yield (kg/ha); Ymis the maximum ex-pected crop yield for a standard condition (no shortage of soil water for crop growth, kg/ha); ETais the actual crop evapotranspiration (mm, H2O); ETmis the crop evapotranspiration for standard condi-tions (mm, H2O); Kyis a yield response factor that describes the reduction in relative yield according to the reduction in ETmcaused

by soil water shortage (Allen et al., 1998)

From Eq.(8), we can see that the Stewart model does not con-sider influences from moisture stress and fertilization conditions

at different growth stages, but it is capable of predicting crop yields with only three parameters Even though other models may pro-vide accurate predictions allowing for more factors, they need many more parameters to be identified beforehand Therefore, the Stewart model is popular in agricultural water resources pro-gramming and economic analysis and it has been recommended

by FAO (Allen et al., 1998; Li et al., 2003; Tolk and Howell, 2008)

4 Model application 4.1 Demonstrational area and data 4.1.1 Description of the irrigation area

In this section, the developed SWAT model is applied in a subbasin of Zhanghe Irrigation District located in Hubei Province, China (Fig 4) The irrigation water for the area is mainly from Zhanghe reservoir through the two main canals In addition, there are thousands of medium-sized and small ponds providing water for irrigation and a complicated but effective irrigation canal sys-tem has been designed to transfer water from ponds and reservoirs

to the fields

The selected area covers 112 891 ha of which the paddy rice ac-counts for 41%, followed by upland crops (18%), forest (16%), bare land (10%), water (9%) and urban (6%) So paddy rice is the main crop in this area (Fig 5) The soil textures are mainly clay (82%) and loam (18%) soil Moreover, the study area is sloping, with ele-vations ranging from 450 m above sea level in the northwest to

20 m in the southeast About 80% of the irrigation area lies in the hilly region This area has a typical subtropical climate with an an-nual mean temperature of 17 °C In most years, there are between

246 and 270 frost-free days Average annual rainfall is 970 mm, although rainfall varies substantially from year to year depending upon the monsoon (Shahbaz et al., 2007a) Thus this area as one

of the large irrigation districts in China is very suitable for paddy rice

4.1.2 Data set

An application of SWAT to a basin needs general data, including topography, soil, land use, climate data and stream flow series A

90 m resolution Digital Elevation Model (DEM) was obtained from Chinese Academy of Sciences (CAS) The land use map with a res-olution of 14.25 m was derived from remote sensing data (Landsat ETM+) in the years of 2000 and 2001 and an unsupervised method was used to classify the land use types (Cai, 2007) Since the land use pattern in this area has not been changed significantly since

2000, it was reasonable to implement our calibration and valida-tion based on data sets of 2005 and 2006 The digital soil map accompanied by a database with soil properties was obtained from the local agriculture department of ZID Moreover, daily data sets for the radiation, wind speed, relative humidity, and air tempera-ture from January 1972 to December 2006 were obtained from Tuanlin experimental station (Fig 4) and they were mainly used

to calculate reference evapotranspiration The daily precipitation data sets (from January 1972 to December 2006) were available

from five stations and daily runoff series of the outlet were

ob-tained in the paddy growth period in the years of 2005 to 2006

In addition, the irrigation records including the time and the

quan-tity of irrigation were collected from the Management Bureau of

Zhanghe Reservoir

InFig 5, the land cover marked with water are ponds or

reser-voirs are simulated as pond processes because there are no distinct

differences for the two kinds of water objects in our study area and

both of them provide the paddy rice area with irrigation water The

characteristic parameters of ponds, such as the surface area and

the hydraulic conductivity, were specified from investigations of

typical ponds in this area Furthermore, crop yields, as the weight

of the paddy rice grain per field area, were collected from 116

typ-ical fields, as shown in Fig 8 In each field, we only picked six

square meters to measure the grain weight

Lastly but not less importantly, the three critical depths for

pad-dy fields should be specified beforehand according to the irrigation

scheme In ZID, the intermittent irrigation technique (IIT) as a

fa-vored scheme is widely implemented to regulate the water depth

in paddy fields For more information about this irrigation scheme, one can refer toMishraa et al (1990), Mao (1997), Anbumozhi

et al (1998), and Wang et al (2005) Here we just present the parameters shown inTable 1 The growing period of the paddy rice

is about 110 days and it can be divided into seven growth stages identified with different water controlling depths The seedling is transplanted on May 25, and the ripe paddy rice is harvested on August 29 It should be noted that there are two dry-state periods with no evaporation from water body, one is in the final tillering stage and the other is in the final ripening stage

4.2 Model evaluation criteria Here we apply three commonly used indicators to evaluate the efficiency and performance of the developed model The first one is the Pearson coefficient (R2) which is a good indicator to evaluate the correlation of observed and simulated results A value of 1 rep-resents perfect correlation, while a value of 0 indicates they are uncorrelated The second one is the relative error coefficient (RE)

Jingmen (City)

Wuhan (City)

Zhanghe Reservoir

Changhu Lake Hubei Province

China

Zhanghe Reservoir

Tuanlin Zhouji

Zhangchang

Shili

Sifang

Xin

Fu River

Outlet

Zhanghe Irrigation District

Fig 4 Location of the demonstrational area of ZID.

Fig 5 Elevation distribution (left) and land use map (right).

that represents the difference between observations and

simula-tions It is expressed as:

RE ¼

Pn

i¼1ðMi OiÞ

Pn

where n equals the total number of observations, Oiand Miare the

observed and simulated values, respectively, on time step i

The third indicator is the Nash–Sutcliffe model efficiency (Ens)

which is given by:

Ens¼ 1 

Pn

i¼1ðMi OiÞ2

Pn

where hOii denotes the mean value of the long-term observations,

and the others terms are defined above The values of Ensrange

be-tween 1 and 1 and the higher the value the more efficient the

cal-ibration A negative value indicates that the mean value of the

observations would have been a better predictor than the simulated

values (Immerzeel et al., 2008)

4.3 Model calibration and validation

The runoff and the crop yield data sets were used to test our

developed model by comparing the simulated and observed values

The crop yield only includes the paddy rice since it is the main crop

in this area (accounts for 70%) and the information for other crops

is difficult to collect

In the calibration process, a number of parameters in SWAT

model need to be adjusted either manually by users or by a

com-puterized optimization algorithm, until a ‘best fit’ parameter set

is found (Kang et al., 2006) The calibration tool incorporated in

AVSWAT (ArcView SWAT) allows users to perform global changes

on input parameters that are commonly modified during the

cali-bration process (Diluzio et al., 2001) Different scenarios should be

set in this tool to get an optimal result In fact, not all parameters

present the same degree of sensitivity in the modeling In the

parameter set, the initial SCS runoff curve number, CN2, the

avail-able water capacity of soil layers, SOL_AWC, and the soil

evapora-tion compensation factor, ESCO, are the most sensitive

parameters for the modeled runoff (Immerzeel et al., 2008; Shen

et al., 2008) We also found that acceptable results could be

ob-tained by mainly adjusting these three parameter values In this

study, therefore, it is convenient to calibrate the model manually

based on a trial-and-error method Certainly, other calibration

methods, including the calibration tool in AVSWAT, may be more

efficient (Neitsch et al., 2001; Muleta and Nicklow, 2005) In

addi-tion, the Stewart’s moisture stress yield reduction coefficient, Ky,

was also adjusted in the crop yield calibration process After

achieving a satisfactory simulation for the runoff and crop yields

in the calibration period, the same modeling environment is

ap-plied in the validation period

4.3.1 Runoff

The first 4 months (from January 1 to April 30, in 2005 and

2006) were used to warm up the model, and the subsequent

5 months (from May 1 to September 30) were used to calibrate

(in the year of 2005) or validate (in the year of 2006) the model

Acceptable results were obtained by manually adjusting

parameters and the performance of the model was assessed accord-ing to the three indicators As shown inTable 2, the Pearson coeffi-cient (R2) and Nash–Sutcliffe criterion (Ens) reach 0.79 and 0.68 in the calibration period Furthermore, the two are above 0.90 and 0.83 respectively in the validation period The absolute values of relative error coefficient (RE) are less than 20% in the two periods The validation period exhibits better agreement than the calibra-tion period probably because the model is more suitable for wetter conditions and the data quality is better in the validation period

In order to determine whether there were advantages in the developed model, we performed a comparative simulation based

on the original edition of SWAT The simulated conditions were identical except that the fraction of potholes in HRUs was set as 0.90 to represent the fraction of paddy fields As shown inFigs 6 and 7, the performance of the original SWAT is not as good as the developed model In particular, there is significant underesti-mation of the peak flow processes In contrast, the simulated daily runoff series from the developed model correspond fairly well with observed data, even though minor discrepancies still exist for the peak flow simulation, for example at the final tillering stage ( Ta-ble 1) when the drainage operation takes place in paddy fields The random and irregular drainage operations carried out by dif-ferent farmers are difficult to account for so we have had to subjec-tively define that the drainage operations for all the paddy fields were performed simultaneously So these minor discrepancies are unavoidable

4.3.2 Crop yields The 116 measured paddy fields were mainly distributed in eight subbasins (Fig 8) We aggregated these measurements to get effec-tive statistics at the subbasin level at which the crop yields are esti-mated in the model The results of calibration and simulation periods were expressed together due to the small number of the target subbasins (eight subbasins for both periods).Fig 9shows

a comparison of measurements and simulations for the paddy rice yields The Pearson coefficient (R2) is greater than 0.60, and the rel-ative error coefficient (RE) is less than 5% So the simulation results roughly agree with the measurement data

There are still some differences between the simulations and the observations Especially in the validation period, the simu-lated crop yields are relatively constant for the eight subbasins, which is not consistent with the observations These inconsisten-cies results from the limits of the Stewart model and the concept formulation of HRUs As described in Section 3.3, the Stewart model only considers the impact from the total amount of evapotranspiration and it fails to assess the influences of evapo-transpiration processes at multiple growth stages The crop growth and yields are dynamically dependent on the evapotrans-piration processes to some extent (Allen et al., 1998) Moreover, the lumped concept of the HRU, which is a combination of a un-ique land use and a soil type, could degrade the simulation of evapotranspiration and consequently spoil the crop yield estimation

4.4 Evaluation of water balance in paddy fields

It is important to test the water balance in paddy plots to eval-uate the effects of the developed model However, here we only

Table 2

Calibration and validation evaluations for daily runoff.

perform analysis on the simulated results rather than compare

them with the observations, since the distributed paddy plots are

aggregated to an HRU in which their specified locations are not

preserved in SWAT We have to pick water balance components

at the HRU level instead of the plot An HRU with land cover of

the paddy rice is randomly picked out from the model, and its

water balance components were derived from both of the

calibra-tion and validacalibra-tion periods

The water depth in paddy fields fluctuates in every day, while it

is restricted by the three critical depth (hmin hmax Hp) according

to the IIT scheme (Fig 10, left andTable 1) When it exceeds the maximum depth (Hp), a drainage operation is executed, for exam-ple on the June 10 in the calibration period (2005) Moreover, when

it reaches the minimum fitting depth (hmin), an irrigation operation

is performed, such as on the June 15 in the validation period (2006)

0 2 4 6 8 10

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep

Date

0

30

60

90

120

150

Precipitation Observation Simulation

(b)

0 2 4 6 8 10

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep

Date

0

30

60

90

120

150

Precipitation Observation Simulation

(a)

Fig 6 Comparison of the observed and the simulated daily runoff hydrographs for the calibration period from May 1 to September 30 in 2005 The (a) and (b) are results from the developed SWAT and the original SWAT respectively.

0 2 4 6 8 10 12 14

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep

Date

0

30

60

90

120

150

180

210 Precipitation (mm/d)

Precipitation Observation Simulation

(b)

0 2 4 6 8 10 12 14

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep

Date

0 30

60 90

120 150

180

210 Precipitation (mm/d)

Precipitation Observation Simulation

(a)

Fig 7 Comparison of the observed and the simulated daily runoff hydrographs for the validation period from May 1 to September 30 in 2006 The (a) and (b) are results from the developed SWAT and the original SWAT respectively.

Even though the evapotranspiration varies with the

tempera-ture and moistempera-ture conditions after the paddy seedlings have been

transplanted, they generally increase until the booting and heading stage and then decrease through to the harvest day (Fig 10, right)

It should be noted that there are some singular points on the trend-line at the final tillering stage in which no water is stored in the fields with the practice of sun drying.Fig 11shows monthly water balance components in paddy fields for the calibration and valida-tion periods The upper part refers to inflow components (In), including precipitation and irrigation, and the lower part denotes depletion components (Out), including evapotranspiration, surface and groundwater discharges The balance closures refer to the sum

of the net change, inflows plus depletions We can see that there is similar behavior in monthly water balance in the both periods The irrigation water, mostly from local ponds, is duly delivered to the fields when required Especially in May and June with a small quantity of precipitation, a significant amount of irrigation water from ponds and reservoirs is used to compensate the insufficient precipitation, and there is no outflow to the main channels In addition, the total inflow (505 mm) is approximately equal to the total depletion (510 mm) in the calibration period, while there is

a discrepancy in the validation period, 471 mm for the total inflow and 537 mm for the depletion In fact, this discrepancy will be off-set by the uptake of the soil water

5 Discussion The developed SWAT model is capable of simulating the main hydrological processes in irrigation areas First, the nature of runoff generation is accurately depicted The paddy field can capture a large amount of precipitation with a constant surface area in order

to keep water for the paddy rice growth In the study area, the ratio

of runoff to precipitation is around 20% (Table 2), which means the crop evapotranspiration and field storages account for most of the volume of precipitation Under these conditions, the simulated runoff processes show good agreement with the observed values and the developed model performs better than the original edition

of SWAT Second, the crop yields are simulated with a simple method The simulated crop yields approximately agree with those observed This method is preferable to the EPIC model when the data is poor Third, the dynamic variation of water depth in paddy fields is characterized well according to the three critical depths of the specified irrigation schemes.Kang et al (2006)also introduced

an outlet height for the pothole drainage, but the flexible operation

of the irrigation and drainage was not considered in their improve-ment Furthermore, the water balance components correspond well to the water requirement conditions of the paddy rice in dif-ferent growing stages Ponds and reservoirs, as local water sources, play an important role for the timely irrigation that compensates for the lack of water supply from canal transfers and precipitation (Shahbaz et al., 2007b) These functions have been adequately rep-resented in the model

1

12 15 11 9 8 7

10

6 5 17

3 4

2

Location for collecting paddy rice yields Subbasin

18

16

Fig 8 Measurement locations for the paddy rice yields.

7.0 7.5 8.0 8.5 9.0

7.0

7.5

8.0

8.5

9.0

Calibration Validation 1:1 line

Measurement (ton/ha)

Fig 9 Comparison of the measurement and the simulation of paddy rice yields.

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep 0

2 4 6 8

Date

ET_Calibration ET_Validation

1-May 31-May 30-Jun 30-Jul 29-Aug 28-Sep 0

20 40 60 80 100

120

Date

Depth_Validation Depth_Calibration h

max

H

p

Compared to the original edition of SWAT, the developed model

does not contain any special sensitive parameters, since no

addi-tional parameters were introduced except the response factor of

crop yields (Kyin Eq.(8)) When performing the parameter

sensi-tivity analysis on this model, we also found that the three

param-eters, runoff curve number, CN2, the available water capacity of soil

layer, SOL_AWC, and the soil evaporation compensation factor,

ESCO, greatly influence the processes of runoff and

evapotranspira-tion This is in line with the conclusions ofMuleta and Nicklow

(2005), Griensven et al (2006)andShen et al (2008) Moreover,

the response factor of crop yields in Stewart model is a dominant

parameter, which has been widely explored when the model is

used to simulated crop yields (Allen et al., 1998; Li et al., 2003)

Therefore the sensitivity analysis is not investigated further here

There are some aspects that deserve to be further research

First, the groundwater simulation system should be developed

Even though SWAT has its own module for the groundwater

simu-lation, the model itself is lumped and therefore distributed

param-eters such as the hydraulic conductivity distributions are not

represented and thus the spatial distributions of the groundwater

level and the recharge rates are difficult to characterize (Kim

et al., 2008) Perhaps combining the SWAT and physically based

ground water models such as MODFLOW is an effective approach

(Sophocleous and Perkins, 2000; Sophocleous, 2005; Kim et al.,

2008) Second, the Stewart model for crop yield estimation is

prac-tical but its precision is limited This is because the actual crop

yield depends on several inputs whereas the analysis here involves

only the water production function (Singh et al., 1999) As shown

inFig 9, the Pearson coefficient (R2) is just over 0.60 If the data

is easily collected, the EPIC model or other approaches, for example

the Jensen model (Jensen, 1968), may be a better alternative (Li

et al., 2003; Igbadun et al., 2007) Third, it is difficult to specify

rea-sonable parameter values for ponds in SWAT, such as the principal

pond volume, since the number of ponds is often overwhelming

and their shapes are too irregular to be characterized in irrigation

areas but these parameters are significant for modeling This can be

addressed by multi-period remote sensing to obtain reliable

esti-mation of parameters

Lastly, it should be noted that our developed SWAT model still

needs further verification and validation studies to construct a

comprehensive hydro-agronomic simulation tool In this study,

only a 2-year data set is used for model testing, one for calibration

and the other for validation In practice, this data set is not

ade-quate and satisfactory for model tests, especially for long-term

hydrological simulations So we are carrying out measurements

in Zhanghe Irrigation District and other irrigation areas

Neverthe-less, based on the qualitative assessments and the water balance

analysis, the results appear to provide a reasonable representation for the paddy fields in agricultural watersheds

6 Conclusion

As a physically-based, comprehensive hydro-agronomic model, SWAT is capable of accurately modeling hydrological processes and crop growth in agricultural watersheds But it fails to consider the complicated water management conditions in paddy areas In this study, we performed improvements on this model The paddy field is assumed to occupy the whole area of an HRU which is the basic computational unit in SWAT and the estimation of actual evapotranspiration of paddy fields depends on two kinds of water storage conditions Moreover, a scheme of controlling irrigation is introduced to this model with irrigation and drainage processes Specifically, three critical water depths are used to adjust the irri-gation and drainage operations in paddy fields Ponds and reser-voirs, as local sources of water storage objects, can provide in a timely manner water for paddy fields to compensate for canal water transfers In addition, a simplified model, the Stewart model,

is adopted to estimate crop yields

We take Zhanghe Irrigation District (in China) as a demonstra-tion area to test these developments The simulated runoff exhibits good agreement with the observed runoff in calibration and valida-tion periods except for the stages when the drainage is carried out These results also indicate that the developed model is preferable

to the original edition of SWAT for paddy rice areas The estimates

of rice yields are also acceptable in both of the periods Moreover, the water balance components, including the daily water depth, actual evapotranspiration and the monthly water balance closures, reasonably represent the actual conditions for paddy fields Consequently, the improved framework is flexible and practical and each of the components can be regarded as an improvement to SWAT in simulating the hydrological processes in irrigation dis-tricts where paddy rice is planted Ongoing work is oriented to-wards improving the groundwater simulation and further testing

of the performance of the model with more data sets from different paddy rice areas All these efforts will help to assess the influences from human activities in the agricultural watershed, such as irriga-tion schemes and crop planting distribuirriga-tion This new tool can also

be used to examine the productivity at different scales in agricul-tural water management

Acknowledgements This study was partially supported by grants from the National Natural Science Foundation of China (No 50879060/50839002)

-400 -300 -200 -100 0 100 200 300 400

May June July August

Month

Precipitation Irrigation Evapotransipiration Outflow Blance closure

-400

-300

-200

-100

0 100

200

300

400

May June July August

Month

Precipitation Irrigation Evapotransipiration Outflow Blance closure

Fig 11 Monthly water balance in paddy fields for the calibration (left) and the validation periods (right).