Hl

AOAa

b

c

d

e

f

g

h

i

a

ARRA

KRDRGS

1

ASttpi

0d

Field Crops Research 133 (2012) 37–47

Contents lists available at SciVerse ScienceDirect

Field Crops Research

jou rn al h om epage: www.elsev ier .com/ locate / fc r

igh-yielding, drought-tolerant, stable rice genotypes for the shallow rainfedowland drought-prone ecosystem

. Kumara,∗, S.B. Verulkarb, N.P. Mandalc, M. Variarc, V.D. Shuklac, J.L. Dwivedid, B.N. Singhe,

.N. Singhf, P. Swainf, A.K. Malla, S. Robing, R. Chandrababug, A. Jainh, S.M. Haefelea, H.P. Piephoi,

. Ramana

International Rice Research Institute (IRRI), DAPO Box 7777, Metro Manila, 4031 Los Banos, Laguna, PhilippinesIndira Gandhi Krishi Vishwavidyalaya (IGKV), Raipur, IndiaCentral Rainfed Upland Rice Research Station (CRURRS), Hazaribag, IndiaNarendra Dev University of Agriculture and Technology (NDUAT), Faizabad, IndiaBirsa Agricultural University (BAU), Ranchi, IndiaCentral Rice Research Institute (CRRI), Cuttack, IndiaTamil Nadu Agricultural University (TNAU), Coimbatore, IndiaBarwale Foundation (BF), Hyderabad, IndiaUniversitaet Hohenheim Bioinformatics Unit, 70593, Stuttgart, Germany

r t i c l e i n f o

rticle history:eceived 29 September 2011eceived in revised form 12 March 2012ccepted 13 March 2012

eywords:iceroughtainfed shallow lowlandenotypestable

a b s t r a c t

High and stable yield of rainfed lowland rice is important for sustainable rice production and food secu-rity. Many varieties grown on large holdings in rainfed areas provide good yield under normal wateravailability but suffer high losses in the event of drought. From a set of 129 genotypes tested in shallowrainfed drought-prone environments at three locations in eastern India from 2005 to 2007, a subset of39 genotypes that were tested for two or more years under favorable irrigated, moderate reproductive-stage drought stress, and severe reproductive-stage drought stress situations in 16 environments wasselected for a GGE biplot analysis to identify genotypes that provide stable yield across environments.IR74371-70-1-1 and IR74371-46-1-1 were identified as stable genotypes showing high yield under var-ied environments across different sites. IR36, IR64, and MTU1010, the three popular varieties grown onlarge holdings in rainfed areas but bred for irrigated ecosystem, as well as improved genotypes CB2-458, DGI237, R1027-2282-2-1, RR272-21, IR67469-R-1-1, and IR66873-R-11-1, and varieties PMK1 andPMK2 released for rainfed ecosystems performed well only in irrigated non-stress environments and

were not found promising in drought environments. Improved genotypes ARB6, ARB2, ARB5, ARB7,ARB8, RF5329, CB0-15-24, IR72667-16-1-B-B-3, IR74371-78-1-1, and IR55419-04, and drought-tolerantreleased varieties Tripuradhan, Annada, and Poornima performed well only in drought-stress environ-ments. The identification of improved genotypes with ability to provide stable high yield across variableenvironments and their release for cultivation by farmers will enable farmers to reap high yield and stableincome.

. Introduction

Rainfed rice accounts for around 45% of the world’s rice area.round 40 million ha of rainfed area is concentrated in South andoutheast Asia alone (Maclean et al., 2002). The rainfed rice ecosys-em is highly fragile. It encounters environments more complex

han other rainfed crops. Rainfed rice-growing areas are highlyrone to abiotic stresses such as drought or submergence depend-

ng upon the amount and distribution of rainfall and toposequence

∗ Corresponding author. Tel.: +63 2 580 5600; fax: +63 2 580 5699.E-mail address: a.kumar@cgiar.org (A. Kumar).

378-4290/$ – see front matter © 2012 Elsevier B.V. All rights reserved.oi:10.1016/j.fcr.2012.03.007

© 2012 Elsevier B.V. All rights reserved.

of the region. Each rice-growing area targeted by any individualbreeding program may still have a mixture of different types ofwater-stress environments in the same year or in different years.

Among the different stresses, drought is the single largest yield-reducing factor in rainfed areas of South and Southeast Asia, withproduction losses common on more than 23 million ha (Huke andHuke, 1997). In light of recent climate change, in the near future,water deficit is predicted to be a major challenge for sustainablerice production (Wassmann et al., 2009). The intensity and fre-

quency of drought are expected to become aggravated (Bates et al.,2008), resulting in decreased food production and food security andincreased vulnerability of the crop to drought (Bates et al., 2008).Among the different rainfed rice-growing areas, India and adjoining

3 ps Res

afohh(grDoctCavaodstor

lvirsrWficlpoTtobpdmbasTb

baetlGsa1tmistKSp1

8 A. Kumar et al. / Field Cro

reas of Nepal occupy the largest drought-prone area in the world,ollowed by northeastern Thailand and Laos. In India, out of the totalf 20.4 million ha of rainfed rice area, approximately 16.2 milliona lie in eastern India (Singh and Singh, 2000), of which 6.3 milliona of upland and 7.3 million ha of lowland area are drought-pronePandey and Bhandari, 2008). In India, from the beginning of thereen revolution era in rice in 1965 till 2009, on 14 occasions,ice production failed to achieve the expected production level.rought was the factor for lower production on 11 of these 14ccasions (DES, 2009). Severe drought witnessed in 2002 and 2009aused a significant reduction in rice as well as total food produc-ion in India. In the eastern Indian states of Jharkhand, Orissa, andhhattisgarh alone, rice production losses in severe drought yearsveraged about 40% of the total production, with an estimatedalue of US$ 888 million (Bhandari et al., 2007). Severe droughtlso has far-reaching effects on the production and productivityf subsequent-season crops grown after rice. Analysis of rice pro-uction in different years from 1950–1951 to 2009–2010 in Indiahows that severe to moderate drought stress not only reducedhe production of rice but also of wheat and pulses. The absencef moisture in the soil caused by late-season drought resulted in aeduction in total food crop production (DES, 2009).

Despite the direct link with development issues, there has beenittle success in developing drought-tolerant rice cultivars. A culti-ar with high mean yield but a low degree of fluctuations in yieldn diverse environments is considered stable and desirable for theainfed ecosystem. However, rice breeders have not been able toelect for such stable varieties because of high genotype by envi-onment (G × E) interactions for grain yield (Cooper et al., 1999a;

ade et al., 1999a). Earlier, a number of constraints such as dif-culty in defining the target population of environments (TPE),hoosing suitable test locations representative of the target popu-ation, and the absence of effective selection criteria from breedingopulations were cited as reasons that hindered the developmentf stable varieties for the rainfed ecosystem (Mackill et al., 1996).he absence of resources and platforms for precise multilocationesting of a large number of varieties in a defined target populationf environments is considered another hindrance to selecting sta-le varieties. In view of this, the emphasis of many rice breedingrograms shifted from developing broadly adaptable varieties toeveloping varieties for specific target environments but this haset with very little success. Over the years, breeders have gained a

etter understanding of the target population of environments. Inddition, in recent years, many national programs have developedystematic testing and evaluation systems for rainfed ecosystems.he development of stable varieties for rainfed ecosystems looks toe more feasible than the scenario a few years back.

Investigations of G × E interactions in rainfed lowland rice haveeen conducted in many studies across Asian countries (Coopernd Somrith, 1997; Cooper et al., 1999a; Ouk et al., 2007). Thesestimated the variance components attributable to G × E interac-ions using restricted maximum likelihood (REML) and the bestinear unbiased estimators (BLUPs) of genotype performances. If

× E interactions are significant, breeders need to know abouttable genotypes with relatively consistent performance across

range of environments. Stability may be static (Lin et al.,986; Becker and Leon, 1988) or dynamic. Stability is static ifhe genotype tends to maintain constant yield across environ-

ents and it is dynamic if a genotype’s performance respondsn a consistent fashion to changes in the environment. Severaltability statistics have been proposed to investigate G × E. Theraditional measures are the coefficient of variation (Francis and

annenberg, 1978), environmental variance (Lin et al., 1986),hukla’s stability variance (Shukla, 1972), and regression-basedarameters of the Finlay–Wilkinson model (Finlay and Wilkinson,963) and Eberhart–Russell model (Eberhart and Russell, 1966).

earch 133 (2012) 37–47

Comprehensive reviews of these measures have been reportedby Westcott (1987), Piepho (1998a), Piepho (1999), and Piephoand van Eeuwijk (2002). The assessment of the stability of grainyield under drought using the above-mentioned statistics has beenreported by several workers (Seboska et al., 2001; Tollenaar andLee, 2002; Okuyama et al., 2005).

Multivariate techniques are powerful tools in extractingpatterns from interactions. The commonly used multivariateapproaches for the analysis of G × E interactions are cluster analy-sis, principal component analysis, and pattern analysis. The majorobjective of the clustering procedure applied to G × E analysis isto cluster lines that have similar responses across environments,thereby reducing the number of comparisons (among lines). Wadeet al. (1999b) used cluster analysis in addition to REML-based mixedmodel analysis to identify genotype groups that vary in yield andrainfed environment groups that were different in terms of theirinfluence on genotypes. Pattern analysis is the combined use ofclassification and ordination methods to explore and explain thestructure of G × E interactions inherent in the data under study(William, 1976). Pattern analysis has served as a useful tool to inves-tigate patterns of G × E interactions of rice in rainfed environments(Wade et al., 1997, 1999b; Abamu and Alluri, 1998).

Biplots are an extensively used graphical technique to displayinteraction patterns and to visualize the interrelationships amonggenotypes, environments, and interactions between genotypes andenvironments and in identifying genotypes that are relatively sta-ble across environments or suitable in particular environments.The biplots are obtained by subjecting the two-way G × E datato singular value decomposition (SVD) and displaying the resultgraphically. Two types of biplot models have been extensively used:(i) AMMI (the additive main effects and multiplicative interaction)biplots and (ii) GGE (genotype + genotype × environment) biplots.The AMMI model combines the analysis of variance of the genotypeand environment main effects with the principal component analy-sis of the G × E interaction. The AMMI2 or the GE interaction biplot isbased on the SVD of a double-centered genotype-by-environmenttable (Gauch, 1992). Since G and E are removed prior to SVD, it dis-plays GE interaction only. The AMMI model has been shown to beeffective in discriminating genotypes that performed well in irri-gated and rainfed mega-environments and in identifying varietiesthat had stable performance across both conditions in wheat andmaize trials (Ozberk et al., 2005; Farshadfar and Sutka, 2006; Kayaet al., 2006; Admassu et al., 2008).

The biplots based on singular value decomposition ofenvironment-centered or within-environment standardized G × Edata were referred to as “GGE biplots” by Yan et al. (2000). Thesebiplots display both G (genotype) and GE (genotype–environment),which are the two sources of variation that are relevant to culti-var evaluation (Kang, 1993). The GGE biplot is based on the sitesregression (SREG) linear–bilinear model (Cornelius et al., 1996;Crossa and Cornelius, 1997; Crossa et al., 2002). The sites regres-sion model as a multiplicative model in the bilinear terms absorbsthe main effects of cultivars plus the cultivar × environment inter-action (GGE). Many studies have used GGE biplot analysis mainlyfor mega-environment evaluation, cultivar evaluation, and assess-ment of varietal stability (Navabi et al., 2006; Dehghani et al., 2006;Blanche et al., 2007; Otoo and Asiedu, 2008; Mohammadi et al.,2009, 2010).

In our study, a set of advanced breeding lines that were testedfor two or more years under a diverse set of conditions that rangedfrom favorable irrigated conditions to conditions with moderate tosevere reproductive-stage drought in the drought-prone easternIndian region was analyzed using GGE biplots with the objective ofidentifying (i) genotypes with stable performance across different

drought-stress levels and irrigated situations, and (ii) genotypesthat performed well in a given situation or environment.

ps Research 133 (2012) 37–47 39

2

2

serafrafIe((Sdherave

cpastsra

fit9bsiatpavfs

2e

setwamkatsss

rip

tion

of

shal

low

rain

fed

low

lan

d

exp

erim

enta

l sit

es.

Envi

ron

men

t

Site

des

crip

tion

Loca

tion

Yea

r

of

exp

eri-

men

tati

onTa

rget

envi

ron

men

t

Leve

l of s

tres

s

Seed

ing

dat

e

Tria

l mea

nyi

eld

(t

ha−1

)R

ain

fall

(mm

)

FZC

05N

aren

dra

Dev

Un

iver

sity

ofA

gric

ult

ure

and

Tech

nol

ogy

(ND

UA

T), F

aiza

bad

, Utt

ar

Prad

esh

26◦ 4

7′ N,

82◦ 1

2′ E20

05R

ain

fed

low

lan

dN

on-s

tres

s

20-J

un

e-05

7.0

575

(128

)FZ

C06

2006

Non

-str

ess

26-J

un

e-06

3.7

738

(130

)FZ

S06

2006

Seve

re

26-J

un

e-06

1.3

738

(130

)FZ

C07

2007

Non

-str

ess

30

Jun

e-07

4.1

960

(117

)FZ

M07

2007

Mod

erat

e

30-J

un

e-07

2.2

960

(117

)H

ZC05

Cen

tral

Rai

nfe

d

Up

lan

d

Ric

eR

esea

rch

Stat

ion

(CR

UR

RS)

,H

azar

ibag

, Jh

arkh

and

23◦ 5

9′ N,

82◦ 2

5′ E20

05R

ain

fed

shal

low

low

lan

dN

on-s

tres

s

25-J

un

e-05

5.0

824

(119

)H

ZS05

2005

Seve

re

28-J

un

e-05

2.0

824

(119

)H

ZS06

2006

Seve

re

05-J

uly

-06

1.0

956

(114

)H

ZC07

2007

Non

-str

ess

26-J

un

e-07

4.6

1265

(112

)H

ZM07

2007

Mod

erat

e

04-J

uly

-07

3.4

1265

(112

)R

PC05

Ind

ira

Gan

dh

i Kri

shi

Vis

hw

avid

yala

ya

(IG

KV

),R

aip

ur,

Ch

hat

tisg

arh

21◦ 1

4′ N,

81◦ 3

8′ E20

05R

ain

fed

shal

low

low

lan

dN

on-s

tres

s

26-J

un

e-05

3.8

1478

(118

)R

PS05

2005

Seve

re

17-J

uly

-05

0.8

1478

(118

)R

PC06

2006

Non

-str

ess

18-J

un

e-06

4.8

1047

(123

)R

PS06

2006

Seve

re

18-J

uly

-06

1.2

1047

(123

)R

PC07

2007

Non

-str

ess

14-J

un

e-07

3.6

1424

(115

)R

PM07

2007

Mod

erat

e

27-J

un

e-07

1.9

1424

(115

)

envi

ron

men

ts

use

d

in

anal

ysis

;

FZ, F

aiza

bad

;

HZ,

Haz

arib

ag;

RP,

Rai

pu

r;

C, C

ontr

ol

non

-str

ess;

S,

Seve

re

dro

ugh

t

stre

ss;

M, m

oder

ate

dro

ugh

t

stre

ss;

(), n

um

ber

of

rain

less

day

s

from

Jun

e

15

to

Nov

embe

r

30

of

each

A. Kumar et al. / Field Cro

. Material and methods

.1. Experimental sites, years of screen, and material screened

For the experimentation, three testing sites representing thehallow rainfed lowland ecosystem (Table 1) were selected tovaluate a set of advanced breeding lines of 105–120 days’ matu-ity duration along with 11 well-characterized released varietiesnd a drought tolerant land race under diverse conditions rangingrom irrigated control (non-stress) to moderate (MRS) to severeeproductive-stage (SRS) drought stress from 2005 to 2007. Thedvanced breeding lines, hereafter called improved genotypesor testing, were nominated by the International Rice Researchnstitute (IRRI), Philippines; the University of Agricultural Sci-nces (UAS), Bangalore; the Tamil Nadu Agricultural UniversityTNAU), Coimbatore; the Indira Gandhi Krishi VishwavidyalayaIGKV), Raipur; and the Central Rainfed Upland Rice Researchtation (CRURRS), Hazaribag. The released varieties includedrought-tolerant varieties such as Annada, Baranideep, Tripurad-an, Khiradhan, and Kallurundaikar; varieties adapted to irrigatedcosystems, such as IR36, IR64, and MTU1010; and varietieseleased in rainfed ecosystems, that is, Poornima, PMK 1, PMK 2,nd PM 1011. These released varieties served as reference lines foralidating the performance of advanced breeding lines in differentnvironments.

At all sites, the experiments were planted under irrigatedontrol and reproductive-stage drought stress. Experiments werelanted in alpha lattice designs with three replications. At Faizabadnd Hazaribag, sowing of the irrigated control and reproductive-tage drought stress experiments was conducted more or less onhe same dates. However, at Raipur, sowing of the reproductive-tage experiments was delayed by 2–4 weeks so that theeproductive stage of the crop synchronized with maximum prob-bility of drought appearance due to withdrawal of monsoon.

Twenty-one-day-old seedlings were transplanted in puddledelds at 20 cm × 20 cm spacing in 4 rows of 4 m length withotal plot size of 3.2 m2. Inorganic NPK fertilizer at the rate of0–60–40 kg ha−1 was applied. P and K were applied as a singleasal dose at transplanting, whereas N was applied in three equalplits, at transplanting and at 25 and 50–56 days after transplant-ng. In stress trials, N was applied in two splits, at transplantingnd at 25 days after transplanting. Weeds were controlled bywo hand weedings. Observations on days to 50% flowering (DTF),lant height (PH), grain yield per plot converted to ha−1 (GY),bove-ground biomass per 1 m linear length of row (BIO), and har-est index (HI) were recorded in all the trials. Daily rainfall dataor the cropping season (Table 1) were recorded at each of theites.

.2. Management of irrigated control and drought-stressxperiments

For each drought-stress experiment, an irrigated control (non-tress, NS) trial was planted at each of the locations. In non-stressxperiments, standing water was maintained at each site fromransplanting to 10 days before maturity by withholding the rain-ater or by providing supplementary irrigation through a pump as

nd when required. The reproductive-stage drought-stress experi-ents (stress, S) were irrigated like the non-stress experiments by

eeping standing water up to 28 days after transplanting. There-fter, the stress fields were drained to allow them to dry and stress

o appear. The stress experiments were not provided with anyupplemental irrigation after drainage even if the stress was veryevere. This has been reflected in the difference in mean yield oftress trials at different sites (Table 1). Ta

ble

1D

etai

led

des

c

Cod

ea

A

B

I C

J D

K

L E M

F N

G

O

H

P aC

ode

for

year

.

4 ps Research 133 (2012) 37–47

2

utoMGciwlTsc

fe(cw

Y

wmficmtrsinaate

z

ecs

v

c

w

tsv

aaiatia

acr

0 A. Kumar et al. / Field Cro

.3. Statistical analysis

From the set of 129 genotypes tested at three locationsnder relatively similar shallow rainfed drought-prone ecosys-ems, Raipur, Hazaribag, and Faizabad from 2005 to 2007, a subsetf 39 genotypes that were tested in two or more years under NS,RS, and SRS in 16 environments at three sites was selected for theGE analysis. A site-year-stress level cross combination is hereafteronsidered a trial/environment. The details of trials are presentedn Table 1. Data from each trial were analyzed using CROPSTAT soft-

are. Line means were estimated using the REML algorithm takingines as fixed and replicates and blocks within replicates as random.he means of the chosen lines obtained from the trial-wise analy-es were arranged on a 39 × 16 G × E table. The table has missingells as not all genotypes were measured in all trials.

The mixed-model analog of the SREG model incorporating aactor-analytic structure to model GGE (genotype plus genotype-by-nvironment) was used to get the best linear unbiased predictorsBLUPs) of all cell means of the G × E table, including the emptyells. The mean yield yij of ith genotype in the jth environment isritten as:

ij = � + ˇj +∑M

m=1�imωjm + �ij (1)

here � is the overall mean and ˇj is the effect of the jth environ-ent which is considered random and genotypes fixed �im is the

actor loading for the factor m associated with genotype g. It can benterpreted as the sensitivity of the ith genotype to environmentalhanges. ωjm is the mth factor (environmental) score for environ-ent j (Piepho, 1998b, 2006). The interaction is thus described as

he product of the sensitivity or the tendency of a treatment toespond to an environmental change (factor loadings) and a mea-ure of the characteristic of the environment (factor scores) and �ijs the error associated with yij. The factor scores ωjm are randomormally distributed with mean 0 and variance �2

ω = 1. The factor-nalytic (FA) model extracts a small number of latent factors thatccount for the inter correlations among the genotypes. The quan-ity to be estimated to complete a two way table of genotypes bynvironments is:

ij = � + ˇj +∑M

m=1�imωjm (2)

The model was fitted using the MIXED procedure in SAS (Littellt al., 2006) with a FA0 (2) structure (Meyer, 2009) in which the spe-ific variances are assumed to be absent. The variance covariancetructure of genotypes within an environment is:

ar(yij) = �2ˇ +

∑M

M=1�2

im�2m + �2

v (3)

ov(yij, yi′ j) = �2ˇ +

∑M

m=1�im�i′m�2

m (4)

here �2ˇ

is the variance associated with the random environmen-

al effect ˇ, �2ω is the variance of ωim and can be taken as 1 bycaling the loadings and scores appropriately and �2

� is the residualariance.

The analysis therefore amounts to an unweighted two stagenalysis (Moehring and Piepho, 2009). In the first stage the data wasnalyzed per trial which produced the adjusted means of genotype

in trial j. In the second stage we compute the marginal meanscross the environments based on the adjusted trial wise geno-ype means. In this case a separation of genotype × environmentnteraction and residual variance is not possible. The BLUPs of inter-∑M

ction effects aij = m=1�imωjm were then environment centered

s aij − ¯aj where ¯aj is the mean of environment j. The environmententered BLUPs were subjected to singular-value decompositionesulting in the genotype and environment scores for each of the

Fig. 1. Biplot of the PC1 vs. PC2 scores obtained from the yield data of 39 genotypesin 8 control and 8 drought stress environments.

two principal components (PC). The G × E-BLUPs were obtainedbased on a multiplicative FA model with two terms, and thereforeconstitute a rank-two approximation of the original data matrix.Singular value decomposition is therefore applied on the rank-twoapproximation of the G × E matrix. The percentages on the figuresthus indicate how the dimensions are split between the two axes.Here, with the use of mixed models there is no way to ascertainhow much of the interaction is captured by the biplot as in thefixed effects case. The SVD merely corresponds to a rotation of theBLUP solution so that the first axis maximizes the explained vari-ance. Symmetrical partitioning of singular values was used (Yanand Hunt, 2003). Plotting PC1 scores against the PC2 scores gen-erated the GGE biplots (Fig. 1). GGE biplots (Yan et al., 2000) wereused to get a graphical view of the SREG model analysis. The GGEbiplot methodology dissects the G and GE which are the sources ofvariation in the SREG model.

Comparisons were made between the stability assessmentsas per “mean vs. stability” view of the GGE biplot, the threecommonly used stability measures (i) Shukla stability variance,(ii) Eberhart–Russell model, (iii) Finlay–Wilkinson model andthe AMMI model. The Eberhart–Russell model, Finlay–Wilkinsonmodel and the AMMI model belong to the family of linear bilinearmodels. The linear–bilinear model for means is given by

yij = � + ˛i + ˇj +∑M

m=1�miωmj + ıij (5)

where � is the overall mean, ˛i is the effect of the ith genotypeand ˇj is the effect of the jth environment. The multiplicativeterm is the product of genotypic sensitivities � and hypotheti-

cal environmental variables ω and ıij is the residual interaction.This equation was named as the additive main effects and multi-plicative model (AMMI) by Gauch (1988). The regression modelsof Finlay–Wilkinson and Eberhart–Russell can be derived from the

ps Research 133 (2012) 37–47 41

lmso

y

m

y

m

y

witap

EGwesMfiitmtmatst

stm

y

woanlsos

3

tfagntRtrn

A. Kumar et al. / Field Cro

inear–bilinear model by setting the environmental effect ω as theean yield of the environment. ω then represents a biological mea-

ure for the environment. The model also known as the regressionn mean model is:

ij = � + ˛i +∑M

m=1�miωmj + ıij (6)

The environmental equivalent of model 4 is the site regressionodel which is:

ij = � + ˇj +∑M

m=1�miωmj + ıij (7)

Shukla’s stability variance model for the genotype–environmenteans is given by:

ij = � + ˛i + ˇj + εij (8)

here εij is the random residual comprising by genotype by trialnteraction and error. Each genotype has its own variance for theerm εij i.e. var(εij) = �2

iknown as the stability variance. The covari-

nce between observations in an environment is the same for allairs of observations cov(yij, yi′ j) = �2

ˇ(Piepho, 1998a).

The stability models of Shukla, Finlay–Wilkinson andberhart–Russell and the AMMI model were fitted to the 39 × 16

× E means in a mixed model framework where genotypesere considered as fixed effect and environments as random

ffect. Stability analysis is based on fitting appropriate covariancetructures for the random effects in models 4 and 6 using theIXED procedure of SAS (Littell et al., 2006). A UN(1) structure

tted to (εij) assigns a separate stability variance to each genotypen model 6. A factor-analytic (FA) structure involving one factorogether with equal or unequal specific variances for the treat-

ents fitted to the multiplicative effect in model 4 correspondso the Finlay–Wilkinson model-FA1(1) and Eberhart–Russell

odel-FA(1), respectively (Piepho, 1997, 1999). Together withn environment main effect var (�j) in model 3, it correspondso the mixed model version of AMMI. The correlations betweentability rankings from the GGE biplot, the stability parameters ofhe traditional stability models and of AMMI were obtained.

Stability of entries was also assessed using a model in which thetress level was incorporated as an explicit factor in the model andhe genotype means for each stress level is reported. The mixed

odel for the 39 × 16 G × E table of means is given by:

lij = � + sl + gi + sgli + flj + εlij (9)

here sl is the effect of the lth stress category, gi is the effectf the ith genotype, sgli is the interaction of the lth stress levelnd the ith genotype, flj is the main effect of the jth environmentested within in the lth stress level and εlij is the residual stress

evel × genotype × environment interaction. The main effects fortress level, genotype and environments as well as the interactionf genotype and stress level are considered fixed, and genotype-pecific variances (�2

s(i)) are estimated for the random residual εlij.

. Results

The experiments at all three sites were planted from June 18o July 18. Total rainfall received by the respective sites rangedrom 575 mm to 960 mm at Faizabad, 824 mm to 1265 mm at Haz-ribag, and 1047 mm to 1478 mm at Raipur. Although, Raipur ineneral received more total rainfall than the other two sites, theumber of rainless days at Raipur (115–123) was comparable withhat of Hazaribag (112–119) and Faizabad (117–130). Further, at

aipur, drought-stress experiments were planted 3–4 weeks laterhan the irrigated non-stress experiments so as to synchronize theeproductive phase of the crop with the period with minimum oro rainfall due to the withdrawal of monsoon rains in the second

Fig. 2. Polygon view of biplot showing the mega environments and the respectivehigh yielding genotypes for each environment.

to third week of September and to expose the experiments to thedesired level of stress.

Among 16 experiments included in the study, 8 were non-stressand 8 were stress experiments. Among the stress experiments, 5stress experiments, FZS07, HZS05, HZS06, RPS05, and RPS06, wereexposed to severe drought stress, leading to a yield reduction ofmore than 60% (Kumar et al., 2008; Verulkar et al., 2010) comparedwith the respective non-stress experiments in the same year ateach site. For HZS06, there was no counterpart non-stress experi-ment. However, with the mean trial yield of HZS06 being half thatof severe drought-stress trial HZS05, this could be considered asa severely drought-stressed experiment. The other three exper-iments (FZM07, HZM07, and RPM07) were moderately exposedto drought, with a yield reduction of 26–47% compared with therespective non-stress experiment (Kumar et al., 2008; Verulkaret al., 2010). The predicted means of 39 genotypes in 16 environ-ments from 2005 to 2007 are presented in the supplementary Table1. The results of the GGE analysis are presented in four sections.

3.1. Identification of ideal genotypes for each environment:Polygon view of the GGE biplot

The polygon was drawn joining cultivars that are located far-thest from the origin so that all other cultivars are contained inthe polygon. The cultivars at the corner of the polygon are called

vertex cultivars (Fig. 2). Vertex genotypes are 22 (IR 72667-16-1-B-B-3), 17 (IR 36), 34 (PMK 2), 14 (DGI 75), 3 (ARB 3), 6 (ARB 6),4 (ARB 4), 7 (ARB 7) and 37 (RF 5329). Perpendicular lines to thesides of the polygon divide the biplot into sectors. Each sector has

42 A. Kumar et al. / Field Crops Research 133 (2012) 37–47

F

aivaF43tic

3

etokdTtfbaaap

at(i

Table 2Comparison of ranking of genotypes based on the (i) environment – O (RPS06) axison the biplot and (ii) the predicted means from SREG model for environment O.

ENO Entry Predicted means(kg ha−1)

Biplotranking

Ranking basedon predictedmeans

1 Annada 2232 13 132 ARB 2 3168 2 33 ARB 3 2863 6 64 ARB 4 3173 3 25 ARB 5 3039 4 46 ARB 6 3221 1 17 ARB 7 2962 5 58 ARB 8 2814 7 79 Baranideep 1907 21 2010 CB 0-15-24 2499 9 911 CB 2-458 1558 27 2912 DGI 237 1416 30 3013 DGI 307 1884 17 2214 DGI 75 1924 16 1815 DSL 104-1 2069 14 1516 DSU 4-7 2016 15 1617 IR36 943 37 3918 IR55419-04 2014 16 1719 IR64 1075 34 3620 IR66873-R-11-1 1057 35 3721 IR67469 R-1-1 1314 31 3322 IR72667-16-1-B-B-3 1906 22 2123 IR74371-3-1-1 1857 20 2424 IR74371-46-1-1 1733 23 2525 IR74371-54-1-1 1628 25 2726 IR74371-70-1-1 1870 19 2327 IR74371-78-1-1 2253 11 1128 Kallurundaikar 2243 12 1229 Khiradan 1413 28 3130 MTU 1010 1587 26 2831 NDR 1098-6 1709 24 2632 PM 1011 1909 18 1933 PMK 1 1093 33 3534 PMK 2 953 36 3835 Poornima 2094 15 1436 R 1027-2282-2-1 1364 29 3237 RF 5329 2692 8 8

ig. 3. Comparison of cultivar performances in the selected environment ‘O’-RPS06.

vertex cultivar. The vertex cultivar is the highest yielding cultivarn the environments that share the sector with it. The sector withertex cultivar 17 (IR36) has environments C (FZC07), I (FZS06),nd J (FZM07); 34 (PMK 2) has B (FZC06), D (HZC05), E (HZC07),(RPC05) and H (RPC07); 3 (ARB 3) has G (RPC06); 6 (ARB 6) and

(ARB 4) have K (HZS05); M (HZM07), N (RPS05), and O (RPS06);7 (RF5329) has L (HZS06), and P (RPM07). It is inferred that cul-ivar DGI 75 is suited to non-stress environments. IR36 does welln Faizabad and PMK 2 yields well under non-stress situation. ARBultivars are good for stress environments.

.2. Performance of different genotypes in a given environment

To visualize the performance of different genotypes in a givennvironment, for example, “O’ (RPS06), which falls in the midst ofhe set of stress trials, a straight line passing through the biplotrigin and the marker of “O” was drawn to make the O-axis. A bro-en line from each cultivar perpendicular to the O-axis was thenrawn. A genotype’s rank is based on its projection onto the O-axis.he genotype’s rank increases the closer the projection is towardhe direction of the environment marker. The relative ranking of dif-erent genotypes in “O” based on the biplots (Fig. 3) and the rankingased on predicted means from model (2) are shown in Table 2. Thenalysis indicated that the ARB lines were more suitable for RPS06,

drought-stressed environment. Among the ARB lines, 6 (ARB6)nd 4 (ARB 4) with the highest biplot ranking and ranking based onredicted means are well suited to stress environments.

In the RPS06 environment, a line that passes through the origin

nd is perpendicular to the O-axis in Fig. 3 separates genotypeshat yielded above the mean (from 4 to 15) and below averagefrom 9 to 34). This applies to other drought-stress environmentsn the same cluster as O, namely, K (HZS05), L (HZS06), M (HZM07),

38 RR 272-21 1298 32 3439 Tripuradhan 2414 10 10

N (RPS05), and P (RPM07). The genotypes between 6 and 15 thatwould possibly yield above average under drought stress include6 (ARB 6), 2 (ARB 2), 5 (ARB 5), 7 (ARB 7), 8 (ARB 8), 37 (RF 5329),10 (CB 0-15-24), 39 (Tripuradhan), 1 (Annada), 28 (Kallurundkar),27 (IR74371-78-1-1), 35 (Poornima), 22 (IR72667-16-1-B-B-3), and18 (IR55419-04). Verulkar et al. (2010) reported that, across sitesand years under severe stress, ARB 8 produced the highest yieldunder stress (1.9 t ha−1), followed by IR55419-04, and Tripuradhanwith 1.8 t ha−1 and 1.7 t ha−1, respectively. Under control, they pro-duced only moderate yield of about 4.5 t ha−1. Entries from 14 to34 would possibly perform above average under non-stress con-ditions as projections from these entries would be closer to thedirection of the control trial markers. These are 9 (Baranideep), 16(DSU 4-7), 32 (PM 1011), 23 (IR74371-3-1-1), 13 (DGI 307), 14 (DGI75), 31 (NDR 1098-6), 24 (IR74371-46-1-1), 25 (IR74371-54-1-1),30 (MTU 1010), 11 (CB 2-458), 12 (DGI 237), 29 (Khiradhan), 36 (R1027-2282-2-1), 38 (RR 272-21), 21 (IR67469-R-1-1), 19 (IR64), 17(IR36), 20 (IR66873-R-11-1), 33 (PMK 1), and 34 (PMK 2).

3.3. Relative adaptation of a given genotype in differentenvironments

To know the environment that is most suited for a cultivar, forexample, genotype 18 (IR55419-04), a line that passes through theorigin and the marker of 18 is drawn, which can be called the gen18axis (Fig. 4). To this line are drawn straight perpendiculars from the

A. Kumar et al. / Field Crops Research 133 (2012) 37–47 43

Fr

eg

tTiIraom

TCt

ig. 4. Comparison of performance of genotype 18 (IR 55419-04) in different envi-onments.

nvironment markers. The rank of the environment increases as itets closer to the environment marker.

The ranking of environments for the performance of geno-ype 18 based on the biplots and predicted yields are shown inable 3. The coefficient of determination between the biplotrank-ng of the environments and the environment-standardized yield ofR55419-04 is 59% (Fig. 4), which is sufficient to distinguish envi-

onments that are suitable and not suitable for IR55419-04 (Yannd Kang, 2003). The broken line that passes through the biplotrigin and is perpendicular to the gen18 axis separates environ-ents where IR55419-04 performed above average (P to I) from

able 3omparison of ranking of sites based on the (i) genotype 18 (IR55419-04) axis onhe biplot and (ii) the predicted means from SREG model for genotype 18.

ENO ENVT Predicted mean(kg ha−1)

Biplot ranking Ranking basedon predictedmeans

1 FZC05 7869 16 12 FZC06 4158 9 23 FZC07 4948 8 34 HZC05 5647 14 45 HZC07 5247 11 56 RPC05 4504 10 67 RPC06 5505 15 78 RPC07 4351 13 89 FZS06 1848 3 910 FZM07 3036 1 1011 HZS05 3207 12 1112 HZS06 1672 2 1213 HZM07 4233 7 1314 RPS05 1703 5 1415 RPS06 1979 6 1516 RPM07 2725 4 16

Fig. 5. Average environment coordinate (AEC) showing the mean yield and stability.

environments where IR55419-04 yielded below average (G to D)(Fig. 4), indicating that the GGE biplot can group environmentsin which a particular genotype performs above or below aver-age. From Fig. 4, it is clear that IR55419-04 performed better indrought environments. These included two of the three moder-ately stressed environments, P (RPM07) and M (HZM07), and allfive severely stressed environments, I (FZS06), J (FZM07), L (HZS06),N (RPS05),and O (RPS06). This analysis reconfirms the finding thatIR55419-04 is well suited to both moderate and severe drought-stress environments.

3.4. Mean performance and stability of the genotypes: Averageenvironment coordinate (AEC) view

To visualize the mean performance and stability of genotypes,a mean-environment represented by a small circle, defined by themean PC1 and PC2 scores, was defined. A line that passes throughthe biplot origin and the mean environment was drawn and wascalled the average environment axis (Fig. 5). The arrow on the axisof the AEC abscissa points in the direction of higher mean perfor-mance of the genotypes and therefore ranks the genotypes withrespect to their means. The AEC abscissa approximates the geno-types’ contributions to G and the AEC ordinate approximates thegenotypes’ contributions to GE, which is a measure of their stability(Yan et al., 2007).

An ideal genotype would be one that has both high mean yieldand high stability. The position of an ‘ideal’ genotype is closer

to the direction of the mean environment and has a zero pro-jection onto the perpendicular AEC ordinate. The genotype meanand stability rankings are presented in Table 4. Based on stabilityranking, DGI75, IR55419-04, DGI307, DSU4-7, Poornima, PM1011,

44 A. Kumar et al. / Field Crops Research 133 (2012) 37–47

Table 4Mean and stability ranking of the genotypes based on the average environmentcoordinate (AEC).

ENO Entry Mean(kg ha−1)

Meanranking

Stabilityranking

Stabilityvariance(Shuklamodel)

1 Annada 3703 37 9 0.3982 ARB 2 3995 11 33 0.8903 ARB 3 4159 1 28 0.7224 ARB 4 3952 17 32 0.8005 ARB 5 3946 18 31 0.6636 ARB 6 4054 5 34 1.0977 ARB 7 3878 25 29 0.6078 ARB 8 3856 28 25 0.6099 Baranideep 3818 31 6 0.16910 CB 0-15-24 3832 29 18 0.35611 CB 2-458 3883 24 17 0.30012 DGI 237 3708 36 21 0.47613 DGI 307 4084 3 2 0.29614 DGI 75 4135 2 1 0.34715 DSL 104-1 4008 10 7 0.14316 DSU 4-7 3983 15 2 0.17717 IR36 3770 33 31 0.32418 IR55419-04 3769 34 1 0.16619 IR64 3925 20 27 0.31320 IR66873-R-11-1 3992 12 27 0.37721 IR67469 R-1-1 3983 14 22 0.64322 IR72667-16-1-B-B-3 3613 39 8 0.40023 IR74371-3-1-1 4047 6 5 0.20224 IR74371-46-1-1 4029 8 10 0.26925 IR74371-54-1-1 3982 16 14 0.22026 IR74371-70-1-1 4019 9 4 0.22127 IR74371-78-1-1 3945 19 13 0.12328 Kallurundaikar 3858 27 11 0.22129 Khiradan 3829 30 19 0.14730 MTU 1010 3921 22 15 0.18731 NDR 1098-6 3895 23 12 0.11132 PM 1011 3924 21 3 0.11533 PMK 1 4045 7 26 0.65234 PMK 2 4065 4 30 0.76835 Poornima 3664 38 2 0.60336 R 1027-2282-2-1 3987 13 20 0.20137 RF 5329 3802 32 23 0.322

I1D7pe4ss

flseauth

gvtatF

Table 5Means and variance components from model 9 where stress-level is an explicitfactor in the model.

Entry Mean (kg ha−1) Variance comp.

Control Stress Stress× environment

Pr Z

ANNADA 4015 2090 144402 0.032ARB 2 4301 2131 928715 0.031ARB 3 4805 2139 848374 0.031ARB 4 4224 2090 800047 0.026ARB 5 4302 2162 652331 0.026ARB 6 4442 2027 1244406 0.051ARB 7 4380 2362 540176 0.029ARB 8 4351 2531 445167 0.024Baranideep 4463 2045 110402 0.029CB 0-15-24 4235 2135 235042 0.050CB 2-458 4833 1847 327558 0.007DGI 237 4292 1672 500622 0.028DGI 307 5030 2154 332518 0.006DGI 75 5225 2040 314932 0.027DSL 104-1 4824 1778 131270 0.027DSU 4-7 4941 1802 190340 0.006IR36 4286 1124 331255 0.006IR55419-04 4529 2203 130239 0.006IR64 4922 1366 213223 0.006IR66873-R-11-1 4967 1131 141239 0.021IR67469 R-1-1 4905 1475 690185 0.020IR72667-16-1-B-B-3 4283 2171 317342 0.020IR74371-3-1-1 5049 1971 193285 0.025IR74371-46-1-1 5018 1966 246051 0.025IR74371-54-1-1 4995 2053 216450 0.025IR74371-70-1-1 5064 2197 222301 0.020IR74371-78-1-1 4816 2037 151145 0.026Kallurundaikar 4598 2053 200143 0.025Khiradan 4494 1315 152678 0.025MTU 1010 4904 1719 167237 0.005NDR 1098-6 4589 1740 108846 0.005PM 1011 4656 1780 117202 0.025PMK 1 5078 1229 451427 0.025PMK 2 5377 1216 327150 0.018POORNIMA 4108 1895 499682 0.043R 1027-2282-2-1 4873 1528 151290 0.018

70-1-1, IR74371-3-1-1, and IR74371-46-1-1 are both reasonably

38 RR 272-21 3870 26 24 0.39739 Tripuradhan 3740 35 16 0.577

R74371-70-1-1, IR74371-3-1-1, Baranideep, DSL104-1, IR72667-6-1-B-B-3, Annada and IR74371-46-1-1 were the top stable lines.GI 75, DGI 307, IR74371-3-1-1, IR74371-46-1-1 and IR74371-0-1-1showed high mean ranking and were identified as the besterforming lines in terms of both mean yield and stability acrossnvironments in the shallow rainfed ecosystem. Further, IR74371-6-1-1 and IR74371-70-1-1 possessed good grain quality andhowed high farmers preference in farmers’ participatory varietalelection trials (data not reported).

The genotype specific REML variance components and meansrom model (9) are presented in Table 5. Based on this model, theines that performed considerably well under both non-stress andtress were IR74371-70-1-1, DGI 75 and DGI 307. The best perform-rs under stress were ARB 8, ARB 7 and Tripuradhan. IR 74371-3-1-1nd IR 74371-46-1-1 yielded well under control and fairly wellnder stress. The stability rankings from model 9 were based onhe overall means of entries across stress levels. The entry with theighest overall mean was considered stable.

Correlation coefficients between the stability rankings for theenotypes obtained from the GGE model, the Shukla stabilityariances, the factor loadings of the Finlay–Wilkinson model,he Eberhart–Russell model, the AMMI model and model (9)

re presented in Table 6. Table 6 also includes the correla-ions obtained after mean centering the factor loadings from theinlay–Wilkinson, Eberhart–Russell, and AMMI models After mean

RF 5329 4216 2008 274771 0.018RR 272-21 4459 1579 456293 0.018TRIPURADHAN 4424 2289 474116 0.024

centering the factor loadings obtained using the AMMI-1 model(data not shown), the rank correlation between them and the sta-bility ranking obtained by the GGE model was high (r = 0.85).

The rank correlation coefficient between the stability ranksobtained using Shukla’s model and the GGE model was also high(r = 0.68) which implies that the genotype’s contributions to the GEinteraction approximated by the AEC ordinate can be more directlyassessed by Shukla’s stability variance model (Gauch et al., 2008).If MINQUE is used as an estimation method, Shukla’s stability vari-ance model has a correlation of 1 with the ecovalence (Lin et al.,1986), which is a genotype’s contribution to the interaction sum ofsquares.

4. Discussion

IR36, IR64, and MTU 1010, the three prominent varieties of110–115 days’ duration and grown on a large area in the shallowrainfed ecosystem of eastern India, are suitable for non-stress situ-ations. MTU 1010 has earlier been reported to possess a moderatedrought tolerance (Jackson, 2008). In our study, MTU1010 is shownto possess moderate stability across stress levels (stability rank15) but to be more stable than IR64 (stability rank 27). IR74371-

good yielding and stable. IR74371-70-1-1 yielded about 5.1 t ha−1

in irrigated conditions and 2.2 t ha−1 under stress. IR74371-3-1-1yielded 5.0 t ha−1 in irrigated conditions and 1.9 t ha−1 under stress.

A. Kumar et al. / Field Crops Research 133 (2012) 37–47 45

Table 6Table of correlations between the stability rankings of entries produced by the GGE model and the parameters of Shukla, AMMI, regression models and the model thatincludes stress level as an explicit factor.

GGERANK Shukla FW ER AMMI1 FW-c ER-c AMMI1-C model9

GGERANK 1 0.68 0.21 0.06 −0.35 0.04 0.12 0.84 0.40<0.0001 0.21 0.72 0.03 0.81 0.49 <0.0001 0.01

Shukla 1 0.04 −0.14 −0.58 0.31 0.24 0.79 0.140.83 0.40 <0.0001 0.06 0.13 <0.0001 0.38

betaFW 1 0.97 −0.44 −0.16 0.12 −0.25 0.12<0.0001 0.01 0.32 0.47 0.13 0.48

betaER 1 −0.63 −0.12 0.08 −0.06 0.09<0.0001 0.45 0.62 0.73 0.59

AMMI1 1 0.11 −0.14 0.53 0.110.52 0.41 0.00 0.50

betaFW c 1 −0.93 0.004 0.17<0.0001 0.98 0.30

betaER c 1 0.07 0.170.67 0.29

AMMI1 C 1 0.100.56

F ean ce

AadbgwsHnp

bMtbtteoltciw

gtowepa

pstttftidhbs

model9

W, Finlay–Wilkinson model; ER, Eberhart–Russell model; Extension-c indicates m

nother breeding line, IR74371-46-1-1, had a high mean yield ranknd showed moderate stability. Baranideep, a variety released forrought-prone areas of eastern Uttar Pradesh province, was sta-le but a moderate yielder. In comparison, other drought-tolerantenotypes such as IR55419-04, ARB 6, and ARB 4 did exceptionallyell in stress conditions but yielded below average in non-stress

ituations at Faizabad (A-FZC05, B-FZC06, C-FZC07), Hazaribag (D-ZC06, E-HZC07), and Raipur (F-RPC05, G-RPC06, H-RPC07) and areot considered stable even though they yielded well in drought-rone environments.

Rainfed lowland rice ecosystems in Asia have been reported toe highly heterogeneous (Cooper et al., 1999b; Wade et al., 1999b;ackill et al., 1996; Fukai et al., 1997). Extremely high environmen-

al variability within a relatively small area and the continual fluxetween anaerobic flooded and aerobic drained situations morehan once in a season within a paddy field have been cited ashe two most important features causing heterogeneity (Coopert al., 1999c). High heterogeneity, coupled with a high frequencyf abiotic stress occurrence such as drought, and submergence andow fertilizer application in fear of crop loss due to abiotic stresseshat leads to further low nutrient availability to plants under stressause yield to remain low in rainfed ecosystems. This has led to lim-ted progress in rainfed environments for increased yield compared

ith irrigated ecosystems.Targeted breeding efforts that take into consideration the

enetic and environmental bases of both broad and specific adap-ation have been suggested as an approach to break the bottleneckf slow yield progress (Cooper et al., 1999b). In many countriesith a large area under rainfed rice cultivation, every year, vari-

ties are released for a specific environment based on their superiorerformance in a specific region. However, gains in yield with thispproach have not been significant.

Rainfed ecosystems are highly heterogeneous, with topogra-hy extending from shallow to mid to deep lowland. Most earliertudies with an aim to find stable varieties for the rainfed ecosys-em tested genotypes without giving much consideration to theopography of testing and investigated G × E interactions to iden-ify stable genotypes for the broader rainfed ecosystem that rangedrom shallow lowland to mid lowland to deep lowland situa-ions. Testing under a large heterogeneous environment resultedn high G × E interactions with no stable variety identified across

ifferent environments. In our study, 105–115 days’ duration,igh-yielding drought-tolerant genotypes derived from crossesetween drought-tolerant donors and high-yielding but drought-usceptible varieties and selected for yield under both non-stress

1

ntered �s from the respective models.

and stress situations were evaluated in well-characterizedcomparatively uniform shallow lowland drought-prone environ-ments vis-à-vis the earlier practice of testing together genotypeswith a different maturity group in shallow to mid lowland vari-able environments. Variable but comparatively uniform shallowlowland environments together with high yield potential andadaptability of the drought-tolerant genotypes to field conditionsvarying from anaerobic to aerobic situations under which they wereselected earlier during the breeding process led to the identifica-tion of stable genotypes. High-yielding varieties IR36, IR64, andMTU1010, bred for the irrigated ecosystem, showed a lower meanyield ranking as well as lower stability ranking because of their poorperformance in stress environments. In contrast, drought-tolerantgenotypes Baranideep, DSU4-7, IR55419-04, and PM 1011 showeda high stability ranking because of their consistent performance inboth non-stress and stress environments even though they werenot very high yielding. Among the identified stable lines, DGI 75,DGI 307, and DSL 104-1 were tall (>115 cm) and showed low farm-ers’ preference (data not reported) compared with IR74371-70-1-1,IR74371-3-1-1, and IR74371-46-1-1. Because of their stable per-formance in our study, IR74371-70-1-1 and its sister lines werefurther evaluated in the shallow lowland drought-prone ecosystemin Nepal and Bangladesh and performed exceedingly well. IR74371-70-1-1 has been released as variety Sahbhagidhan for cultivationin Jharkhand and Orissa provinces of India, IR74371-70-1-1 and itssister line IR74371-46-1-1 have been released as varieties Sukhad-han 3 and Sukhadhan 2 in Nepal, and IR74371-70-1-1 is releasedas variety BRRI dhan56 in Bangladesh.

5. Conclusions

IR74371-70-1-1 and IR74371-46-1-1 were identified as geno-types with comparatively stable performance and adaptable planttypes for shallow rainfed lowland drought-prone ecosystems. BothIR74371-70-1-1 and IR74371-46-1-1 possessed good grain qualitytraits and received high farmers’ preference. These lines are derivedfrom a cross of drought-tolerant breeding line IR55419-04 withdrought-susceptible variety Way Rarem and they were advancedthrough selection under both non-stress and stress conditions dur-ing early segregating and advanced generations. Selection underboth situations enabled breeders to select such lines that provide

high yield under normal conditions and good yield under droughtconditions through better adaptation to fluctuating anaerobicto aerobic situations, leading to better maintenance of repro-ductive activities under drought-stress situations. The popular

4 ps Res

vps

A

P

A

t

R

A

A

B

BB

B

C

C

C

C

C

C

C

D

D

E

F

F

F

F

G

6 A. Kumar et al. / Field Cro

arieties IR36, IR64, and MTU 1010 bred for the irrigated ecosystemerformed better only under non-stress situations but they are notuitable for drought-prone shallow rainfed ecosystems.

cknowledgments

This study was supported in part by the Generation Challengerogram (GCP) and the Bill & Melinda Gates Foundation, USA.

ppendix A. Supplementary data

Supplementary data associated with this article can be found, inhe online version, at doi:10.1016/j.fcr.2012.03.007.

eferences

bamu, F.J., Alluri, K., 1998. AMMI analysis of rained lowland rice (Oryzasativa) trialsin Nigeria. Plant Breed 117, 395–397.

dmassu, S., Nigussie, M., Zelleke, H., 2008. Genotype–environment interaction andstability analysis for grain yield of maize (Zea mays L.) in Ethiopia. Asian J. PlantSci. 7, 163–169.

ates, B.C., Kundzewicz, Z.W., Wu, S., Palutikof, J.P., 2008. Climate change and water.In: Technical Paper of the Intergovernmental Panel on Climate Change, IPCC,Secretariat, Geneva, p. 210. http://www.ipcc.ch/ipccreports/tp-climate-change-water.htm.

ecker, H.C., Leon, J., 1988. Stability analysis in plant breeding. Plant Breed 101, 1–23.handari, H., Pandey, S., Sharan, R., Naik, D., Hirway, I., Taunk, S.K., Sastri, A.S.R.A.S.,

2007. Economic costs of drought and rice farmers’ drought-coping mechanismsin eastern India. In: Pandey, S., Bhandari, H., Hardy, B. (Eds.), Economic Costsof Drought and Rice Farmers’ Coping Mechanisms: A Cross-Country Compara-tive Analysis. International Rice Research Institute, Los Banos, Philippines, pp.43–111.

lanche, S., Myers, G., Kang, M.S., 2007. GGE biplots and traditional stability mea-sures for interpreting genotype by environment interactions. J. Crop Imp. 20,123–135.

ooper, M., Somrith, B., 1997. Implications of genotype-by-environment interac-tions for yield adaptation of rainfed lowland rice: influence of flowering date onyield variation. In: Fukai, S., Cooper, M., Salisbury, J. (Eds.), Breeding Strategies forRainfed Lowland Rice in Drought-prone Environments. Proceedings of an Inter-national Workshop held at Ubon Ratchathani, Thailand, 5–8 November 1996.Australian Centre for International Agricultural Research, Canberra, Australia,pp. 104–114.

ooper, M., Rajatasereekul, S., Immark Fukai, S., Basnayake, J., 1999a. Rainfed low-land rice breeding strategies for northeast Thailandm I. Genotypic variationand genotype × environment interaction for grain yield. Field Crops Res. 64,131–151.

ooper, M., Rajatasereekul, S., Somrith, B., Sriwisut, S., Immark, S., Boonwite, C.,Suwanwongse, A., Ruangsook, S., Hanviriyapant, P., Romyen, P., Porn-uraisanit,P., Skulkhu, E., Fukai, S., Basnayake, L., Podlich, D.W., 1999b. Rainfed lowlandrice breeding strategies for northeast Thailand. 2. Comparison of intrastationand interstation selection. Field Crops Res. 64, 153–176.

ooper, M., Fukai, S., Wade, L.J., 1999c. How can breeding contribute to more produc-tive and sustainable rainfed lowland rice systems? Field Crops Res. 64, 199–209.

ornelius, P.L., Crossa, J., Seyedsadr, M.S., 1996. Statistical tests and estimatorsof multiplicative models for genotype-by-environment interaction. In: Kang,M.S., Gauch, H.G. (Eds.), Genotype-by-Environment Interaction. CRC Press, BocaRaton, FL, pp. 199–234.

rossa, J., Cornelius, P.L., 1997. Sites regression and shifted multiplicative modelclustering of cultivar trials sites under heterogeneity of variances. Crop Sci. 37,406–415.

rossa, J., Cornelius, P.L., Yan, W., 2002. Biplots of linear–bilinear models for studyingcrossover genotype–environment interaction. Crop Sci. 42, 619–633.

ehghani, H., Ebadi, A., Yousefi, A., 2006. Biplot analysis of genotype by environmentinteraction for barley yield in Iran. Agron. J. 98, 388–393.

ES, Directorate of Economics and Statistics, 2009. Area, production and yield ofprincipal crops. Department of Agriculture and Cooperation, Ministry of Agri-culture, Government of India. http://www.dacnet.nic.in.

berhart, S.A., Russell, W.A., 1966. Stability parameters for comparing varieties. CropSci. 6, 36–40.

arshadfar, E., Sutka, J., 2006. Biplot analysis of genotype environment interactionin durum wheat using the AMMI model. Acta Agron. Hung. 54, 459–467.

inlay, K.W., Wilkinson, G.N., 1963. The analysis of adaptation in a plant-breedingprogram. Aust. J. Agric. Res. 14, 742–754.

rancis, T.R., Kannenberg, L.W., 1978. Yield stability studies in short-season maize.1. A descriptive method for grouping genotypes. Can. J. Plant Sci. 58, 1029–1034.

ukai, S., Cooper, M., Salisbury, J. (Eds.), 1997. Breeding strategies for rainfed lowland

rice in drought prone environments. Proceedings of an International Workshop,Ubon Ratchathani, Thailand, 5–8 November 1996. ACIAR Proceedings no. 77,Canberra, ACT, p. 248.

auch, H.G., 1988. Model selection and validation for yield trials with interaction.Biometrics 44, 705–715.

earch 133 (2012) 37–47

Gauch, H.G., 1992. Statistical Analysis of Regional Yield Trials: AMMI Analysis ofFactorial Designs. Elsevier, Amsterdam, p. 278.

Gauch, H.G., Piepho, H.P., Annicchiarico, P., 2008. Statistical analysis of yield trialsby AMMI and GGE: further considerations. Crop Sci. 48, 866–889.

Huke, R.E., Huke, E.H., 1997. Rice Area by Type of Culture: South, Southeast, and EastAsia. IRRI, Los Banos, Philippines.

Jackson, M.T., 2008. Integrating and mobilizing rice knowledge to improve andstabilize crop productivity to achieve household food security in diverse andless favorable rainfed areas of Asia (IRRI Ref. No.: DPPC2003-42). In: TechnicalAssistance Completion Report submitted to the Asian Development Bank (ADB).

Kang, M.S., 1993. Simultaneous selection for yield and stability in crop performancetrials: consequences for growers. Agron. J. 85, 754–757.

Kaya, Y., Akcura, M., Ayranci, R., Taner, S., 2006. Pattern analysis of multi-environment trials in bread wheat. Commun. Biometry Crop Sci. 1, 63–71.

Kumar, A., Bernier, J., Verulkar, S., Lafitte, H.R., Atlin, G.N., 2008. Breeding for droughttolerance: direct selection for yield, response to selection and use of drought-tolerant donors in upland and lowland-adapted populations. Field Crops Res.107, 221–231.

Lin, C.S., Binns, M.R., Lefkovitch, L.P., 1986. Stability analysis: where do we stand?Crop Sci. 26, 894–900.

Littell, R.C., Milliken, G.A., Stroup, W.W., Wolfinger, R.D., Schabenberger, O., 2006.SAS for Mixed Models, 2nd ed. SAS Institute Inc., Cary, NC.

Mackill, D.J., Coffman, W.R., Garrity, D.P., 1996. Rainfed Lowland Rice Improvement.International Rice Research Institute, Manila, Philippines, p. 242.

Maclean, J.L., Dawe, D.C., Hardy, B., Hettel, G.P. (eds.), 2002. Rice almanac. Los Banos(Philippines): International Rice Research Institute, Bouaké (Côte d’Ivoire): WestAfrica Rice Development Association, Cali (Colombia): International Center forTropical Agriculture, Rome (Italy): Food and Agriculture Organization.

Meyer, K., 2009. Factor-analytic models for genotype × environment type problemsand structured covariance matrices. Genet. Select. Evol. 41, 21.

Mohammadi, R., Amri, A., Ansari, Y., 2009. Biplot analysis of rainfed multi-environment trials in Iran. Agron. J. 101, 789–796.

Mohammadi, R., Haghparast, R., Amri, A., Ceccarelli, S., 2010. Yield stability of rainfeddurum wheat and GGE biplot analysis of multi-environment trials. Crop PastureSci. 61, 92–101.

Moehring, J., Piepho, H.P., 2009. Comparison of weighting in two-stage analysis ofplant breeding trials. Crop Sci. 49, 1977–1988.

Navabi, A., Yang, R.C., Helm, J., Spaner, D.M., 2006. Can spring wheat-growing megaenvironments in the Northern Great Plains be dissected for representative loca-tions or niche-adapted genotypes? Crop Sci. 46, 1107–1116.

Okuyama, L.A., Federizzi, L.C., Neto, J.F.B., 2005. Grain yield stability of wheat geno-types under irrigated and non-irrigated conditions. Braz. Arch. Biol. Technol. 48,697–704.

Otoo, E., Asiedu, R., 2008. GGE biplot analysis of Dioscorearotundata cultivar DENTEin Ghana. Afr. Agric. Res. 3, 115–125.

Ouk, M., Basnayake, J., Tsubo, M., Fukai, S., Fischer, K.S., Kang, S., Men, S., Thun, V.,Cooper, M., 2007. Genotype-by-environment interactions for grain yield asso-ciated with water availability at flowering in rainfed lowland rice. Field CropsRes. 101, 145–154.

Ozberk, I., Ozberk, F., Braun, H.J., 2005. AMMI stability of some internationallyderived durum wheat varieties in the south eastern of Anatolia. Pakistan Biol.Sci. 8, 118–122.

Pandey, S., Bhandari, H., 2008. Drought: economic costs and research implications.In: Serraj, R., Bennett, J., Hardy, B. (Eds.), Drought Frontiers in Rice: Crop Improve-ment for Increased Rainfed Production. World Scientific Publishing, Singapore,pp. 3–17.

Piepho, H.P., 1997. Analyzing genotype–environment data by mixed models withmultiplicative effects. Biometrics 53, 761–766.

Piepho, H.P., 1998a. Methods for comparing the yield stability of cropping systems:a review. J. Agron. Crop Sci. 180, 193–213.

Piepho, H.P., 1998b. Empirical best linear unbiased prediction in cultivar trialsusing factor analytic variance covariance structures. Theor. Appl. Genet. 97,195–201.

Piepho, H.P., 1999. Stability analysis using the SAS system. Agron. J. 91, 154–160.Piepho, H.P., 2006. Erfassung der phänotypischen Stabilität in züchterischen Ver-

suchen, pp. 93–98. In: Tagungsband der 56. Jahrestagung der Vereinigungder Pflanzenzüchter und Saatgutkaufleute Österreichs, 22–24 November 2005.Raumberg-Gumpenstein, Austria.

Piepho, H.P., van Eeuwijk, F.A., 2002. Stability analysis in crop performance valu-ation. In: Manjit, S.K. (Ed.), Crop Improvement: Challenges in the Twenty FirstCentury. Food Products Press, New York.

Seboska, G., Nigussie, M., and Bogale, G., 2001. Stability of drought tolerant maizegenotypes in the drought stressed areas of Ethiopia. Seventh Eastern and SouthAfrican Regional Maize Conference, Nazreth Research Center, P.O. Box 436,Nazreth, Ethiopia.

Shukla, G.K., 1972. Some aspects of partitioning genotype–environmental compo-nents of variability. Heredity 29, 237–245.

Singh, V.P., Singh, R.K. (Eds.), 2000. Rainfed Rice: A Source Book for Best Practicesand Strategies in Eastern India. International Rice Research Institute, Los Banos,Philippines, p. 292.

Tollenaar, M., Lee, E.A., 2002. Yield potential, yield stability and stress tolerance in

maize. Field Crops Res. 75, 161–169.

Verulkar, S.B., Mandal, N.P., Dwivedi, J.L., Singh, B.N., Sinha, P.K., Mahato, R.N., Swain,P., Dongre, P., Payasi, D., Singh, O.N., Bose, L.K., Robin, S., Babu, R.C., Senthil,S., Jain, A., Shashidhar, H.E., Hittalmani, S., Vera Cruz, C., Paris, T., Hijmans, R.,Raman, A., Haefele, S., Serraj, R., Atlin, G., Kumar, A., 2010. Breeding resilient

ps Res

W

W

W

W

A. Kumar et al. / Field Cro

and productive rice genotypes adapted to drought-prone rainfed ecosystems ofIndia. Field Crops Res. 117, 197–208.

ade, L.J., Fukai, S., Samson, B.K., Ali, A., Mazid, M.A., 1999a. Rainfed lowland rice:physical environment and cultivar requirements. Field Crops Res. 64, 3–12.

ade, J.L., McLaren, C.G., Criseno, L., Amarante, S.T., Warawgi, A.K., Kumar, R.,Bhamri, M.C., Singh, O.N., Ahmed, H.U., Rajatasereeku, S., Pomurai, S.P., Boon-wite, C., Harnpichivitaya, D., Sarkarung, S., 1997. Genotype-by-environmentinteractions: RLRRC experience. In: Fukai, S., Cooper, M., Salisbury, J. (Eds.),Breeding Strategies for Rainfed Lowland Rice in Drought-Prone Environments.International Rice Research Institute, Philippines, pp. 115–125.

ade, L.J., McLaren, C.G., Quintana, L., Harnpichitvitaya, D., Rajatasereekul, S.,

Sarawgi, A.K., Kumar, A., Ahmed, H.U., Sarwoto Singh, A.K., Rodriguez, R., Sio-pongco, J., Sarkarung, S., 1999b. Genotype by environment interactions acrossdiverse rainfed lowland rice environments. Field Crops Res. 64, 35–50.

assmann, R., Jagadish, S.V.K., Sumfleth, K., Pathak, H., Howell, G., Ismail, A., Ser-raj, R., Redona, E., Singh, R.K., Heuer, S., 2009. Regional vulnerability of climate

earch 133 (2012) 37–47 47

change impacts on Asian rice production and scope for adaptation. Adv. Agron.102, 91–133.

Westcott, B., 1987. A method for assessing the yield stability of crop genotypes. J.Agric. Sci., Camb. 108, 267–274.

William, W.T., 1976. Pattern Analysis in Agricultural Science. Elsevier Scientific Pub-lishing Company, Amsterdam.

Yan, W., Hunt, L.A., 2003. Biplot analysis of multi-environment trial data. In: Kang,M.S. (Ed.), Quantitative Genetics, Genomics, and Plant Breeding. CAB, Interna-tional, Wallingford, Oxon, UK, pp. 289–303.

Yan, W., Hunt, L.A., Sheng, Q., Szlavnics, Z., 2000. Cultivar evaluation andmega-environment investigation based on the GGE biplot. Crop Sci. 40,

597–605.

Yan, W., Kang, M.S., 2003. GGE Biplot Analysis: A Graphical Tool for Breeders, Geneti-cists and Agronomists. CRC Press, Boca Raton, FL.

Yan, W., Kang, M.S., Ma, B., Woods, S., Cornelius, P.L., 2007. GGE biplot vs. AMMIanalysis of genotype-by-environment data. Crop Sci. 47, 643–653.