norms for multivariate diagnosis of nutrient imbalance in the east african highland bananas (musa...

This article was downloaded by: [North Carolina State University]On: 07 September 2012, At: 16:34Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Plant NutritionPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/lpla20

NORMS FOR MULTIVARIATE DIAGNOSISOF NUTRIENT IMBALANCE IN THE EASTAFRICAN HIGHLAND BANANAS (MUSA SPP.AAA)Lydia Wairegi a & Piet van Asten ba Centre for Agriculture and Biosciences International (CABI), AfricaRegional Centre, Nairobi, Kenyab International Institute of Tropical Agriculture, Kampala, Uganda

Version of record first published: 04 Jul 2011

To cite this article: Lydia Wairegi & Piet van Asten (2011): NORMS FOR MULTIVARIATE DIAGNOSIS OFNUTRIENT IMBALANCE IN THE EAST AFRICAN HIGHLAND BANANAS (MUSA SPP. AAA), Journal of PlantNutrition, 34:10, 1453-1472

To link to this article: http://dx.doi.org/10.1080/01904167.2011.585203

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representationthat the contents will be complete or accurate or up to date. The accuracy of anyinstructions, formulae, and drug doses should be independently verified with primarysources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand, or costs or damages whatsoever or howsoever caused arising directly orindirectly in connection with or arising out of the use of this material.

http://www.tandfonline.com/loi/lpla20

http://dx.doi.org/10.1080/01904167.2011.585203

http://www.tandfonline.com/page/terms-and-conditions

Journal of Plant Nutrition, 34:1453–1472, 2011Copyright C© Taylor & Francis Group, LLCISSN: 0190-4167 print / 1532-4087 onlineDOI: 10.1080/01904167.2011.585203

NORMS FOR MULTIVARIATE DIAGNOSIS OF NUTRIENT IMBALANCE

IN THE EAST AFRICAN HIGHLAND BANANAS (MUSA SPP. AAA)

Lydia Wairegi1 and Piet van Asten2

1Centre for Agriculture and Biosciences International (CABI), Africa Regional Centre,Nairobi, Kenya2International Institute of Tropical Agriculture, Kampala, Uganda

� Despite low yields and soil fertility problems, fertilizer use in the East African Highland banana(AAA-EA) production is absent. High fertilizer costs increase the need for site-specific fertilizer recom-mendations that address deficiencies. This study aimed to derive and compare norms for AAA-EAbananas, using Compositional Nutrient Diagnosis (CND), Diagnosis and Recommendation Inte-grated System (DRIS), and a DRIS that includes a filling value (DRIS-Rd), and study nutrientinteractions. Data on foliar nitrogen, phosphorus, potassium, calcium, and magnesium concentra-tions, and plant performance were obtained from 300 plots in Uganda. CND indices were closelyrelated to DRIS and DRIS-Rd indices (R2 > 0.965). Four nutrient interactions were common inboth low and high bunch weight subpopulations. Although the three approaches can be used to de-termine nutrient imbalances in AAA-EA bananas, we recommend CND for ease of use. Diagnosisof nutrient deficiencies should be based on methods that identify plant nutritional imbalances.

Keywords: compositional nutrient diagnosis (CND), diagnosis and recommendationintegrated system (DRIS), foliar nutrient concentration, nutrient imbalance, nutrientinteractions

INTRODUCTION

Banana is an important food crop in the world (Samson, 1992) and iswidely grown in the Caribbean, Asia, Africa and Latin America (FAO, 2009).In East Africa, the highland bananas (AAA-EA) are a primary food and cashcrop for over 30 million people. In Uganda, bananas are the primary staplecrop (Edmeades, 2006). However, actual production (<30 tonnes hectare−1

year−1) is low compared to potential yield (>70 tonnes hectare−1 year−1)(van Asten et al., 2005). Poor soil fertility is a major production constraint

Received 1 July 2009; accepted 25 July 2009.Address correspondence to Lydia Wairegi, Centre for Agriculture and Biosciences International

(CABI), Africa Regional Centre, P.O. Box 633-00621, Nairobi, Kenya. E-mail: [email protected]

1453

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012

1454 L. Wairegi and P. van Asten

(Wairegi et al., 2007). Despite declining banana yields, soil nutrient miningdue to increased banana commercialization, and the existence of blanketfertilizer recommendations, on-farm fertilizer use in Uganda is extremelylow (<2 kg hectare−1 of arable land in 2002, Camara and Heinmann, 2006).Farmers reported high fertilizer prices as the major bottleneck to its adoption(van Asten et al., 2010). The recent (2007–2008) sharp rise (50–100%) infertilizer prices in Uganda puts further pressure on the profitability andadoptability of fertilizer, with consequent negative impact for the long termfood security in the densely populated East African highlands. There is astrong need to develop more profitable fertilizer recommendations thattarget primary nutrient deficiencies.

Plant nutrient deficiencies are more directly diagnosed using foliar anal-ysis as opposed to soil analysis (Hallmark and Beverly, 1991) since the corre-lation between soil and plant nutrient status is often poor (Hanson, 1987).Since nutrient uptake and distribution are affected by interactions within theplant, multi-nutrient approaches have been derived. Two common methodsused to diagnose nutritional imbalances are the Diagnosis and Recommen-dation Integrated System (DRIS) (Walworth and Sumner, 1987) and Com-positional Nutrient Diagnosis (CND) (Parent and Dafir, 1992).

DRIS is based on dual ratio functions while CND is based on row-centeredlog ratios where each nutrient is adjusted to the geometric mean of all nutri-ents and to a filling value (Rd). Hallmark et al. (1987) proposed a modifiedDRIS which includes ratios of nutrient concentrations to dry matter (e.g.,N/DM, where N and DM denote nitrogen and dry matter, respectively)for identifying limiting nutrients. Khiari et al. (2001a) used Rd instead ofDM (e.g., N/Rd) because the dry matter content includes nutrient concen-trations already defined in the tissue simplex. Modified DRIS can identifysituations where nutrients are not limiting unlike DRIS where at least onenutrient will always have a negative index (Hallmark et al., 1987).

Holland (1966) suggested use of Principal Component Analysis (PCA) toexplain the effects of fertilizer application on foliar nutrient concentrations.Since CND is related to PCA (Parent et al., 1994b), PCA has been appliedto CND derived row-centered log ratios to explore nutrient relationshipsin studies by Parent et al. (1993), Parent et al. (1994b) and Raghupathi etal. (2002). PCA has also been conducted on DRIS indices (Raghupathi etal., 2005; Parent et al., 1994b) to further explore plant nutrient interactions.PCA conducted on row-centered log ratios were easier to interpret comparedwith PCA conducted on DRIS indices (Parent et al., 1994b)

The DRIS method is reportedly inferior CND in diagnosing imbalancesas it assumes additivity of dual ratios (Parent and Dafir, 1992). Furthermore,DRIS does not directly take into account higher order interactions (Parentand Dafir, 1992). Compared with DRIS, CND appeared to be more sensitivefor early detection of N stress in sweet corn (Khiari et al., 2001a) and forprojecting nutrient imbalances in turmeric (Kumar et al., 2003). In studies

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012

Nutrient Norms for Highland Bananas 1455

on tomatoes (Parent et al., 1993) and potatoes (Parent et al., 1994a), DRISand CND provided similar results.

There have been several studies on nutrient imbalances in bananas.Angeles et al. (1993) calculated DRIS norms primarily based on Cavendishcultivars. Wortmann et al. (1994) developed DRIS norms for AAA-EA cul-tivars, but limited the study to the Kagera region in Northwest Tanzania.Raghupathi et al. (2002) derived CND norms for Ney Poovan and Robustabanana cultivars grown in India. DRIS norms for a crop sometimes differbetween regions (Walworth and Sumner, 1987). CND norms specific to theEast African Highland cooking banana do not exist. Hence, studies on nu-trient imbalances in East African bananas have neither covered a wide rangeof agro-ecological conditions, nor have norms been developed using CND.

The objectives of this study were (i) to investigate how foliar nutrientconcentrations relate to yield, (ii) to derive and compare CND and DRISnorms for East African highland bananas, and (iii) to investigate the signif-icance and direction of nutrient interactions using data derived a very widerange of agro-ecologies in Uganda. The norms will be compared to those de-veloped in previous studies. Nutrient interactions will be investigated usingPrincipal Component Analysis (PCA).

MATERIALS AND METHODS

Study Area

The study was carried out in 300 plots, in mature plantations (5–50years old), in Central, South, Southwest and East Uganda. The area liesapproximately between latitudes 1◦30′ North and 1◦ 00′ South and longitude29◦ 52′ and 34◦ 30′ East at an altitude of 1120–1700 m above sea level. Thesoils are mainly Acrisols and Ferralsols according to the FAO classification.The parent rocks underlying the soils sampled in this study have beendescribed Archaean Gneissic-Granulitic-Complex, Proterozoic sedimentsand Cenozoic volcanic outcrops (Schluter, 2008). The East African Mete-orological Department (1963) generalized the rainfall pattern as bimodalwith rains occurring in March-May and September-November. Annualrainfall (mm year−1) at the study sites estimated using the local climateestimator (LocClim, FAO, Rome, Italy ranged between 782 and 1797 mmyear−1. LocClim rainfall estimates were assumed to be reliable because themean annual rainfall predicted by LocClim for specific sites in Central andSouthwest was not significantly different from actual observations reportedfor these sites in previous studies (e.g., Smithson et al., 2004; Okech et al.,2004). In addition, farmers and local agricultural staff observed that theactual rainfall amount and distribution during the study period was normal.LocClim has been used in previous studies to estimate rainfall (e.g., Heatonet al., 2004; Mokany et al., 2006).

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


Of these plots, 28, 34, 24 and 9 plots in Central, South, Southwestand East, respectively, were applied nitrogen (N), phosphorus (P), andpotassium (K) fertilizer averaging 71, 8 and 32 kg hectare−1 year−1, respec-tively, supplied by an extension program (Wairegi et al., 2007) while the rest(25, 64, 42 and 71 plots in Central, South, Southwest and East, respectively)were not applied fertilizer. Since all plots were fully managed by farmers,other management practices (e.g., use of animal manure, mulch) were thosecommonly used by farmers. Banana production in all plots was rain-fed.

Data Collection

The data used in this study was collected over a period of 12 months in2006–2007. In order to have a large dataset, data from a monitoring study(176 plots) and a multiple visit survey (124 plots) were combined. In theformer, 30 randomly selected mats were monitored for the year while inthe latter data was collected from 20 randomly selected mats with bunch-bearing plants. During the study period, field visits were made at 3–5 monthintervals in the monitoring study and at 4–6 month interval in the survey. Inthe monitoring study, farmers received weighing balances to record bunchweights of marked mats. In both studies, data on girth of stem at base and1m, number of hands, number of fingers in the bottom row of the secondlowest hand were collected from fruiting plants and used to estimate bunchweights using the general allometric regression derived by Wairegi et al.(2009), where data on bunch weights were missing.

Foliar sub-samples of 10 by 20 cm were collected from both sides of themidrib in the midpoint of the lamina from the third most fully expanded leafof a flowering plant (Lahav, 1995) and composited for each plot. Compositesamples consisted of three to seven plants per plot. Since plots were notuniform in size, efforts were made to collect more samples in larger plotsthan in smaller plots. Analyses were as described in Okalebo et al. (1993).The samples were oven dried at 72◦C for 48–96 hours, ground to < 2 mmparticle size, digested in a sulfuric and selenium acid mixture. The sampleswere analyzed colorimetrically for N and P while K was analyzed using aflame photometer and calcium (Ca), and magnesium (Mg) were analyzedusing an atomic absorption spectrophotometer.

Analytical Approach

The production data was collected during a 12-month period over awide range of agro-ecological conditions across the entire country. Bananaproduction is continuous; i.e. plants in farmers’ fields are at different stagesof development at any given point in time. Therefore, the data presented inthis study are assumed to be representative for the study regions as it took intoconsideration temporal and spatial variation in banana production systems.This was further emphasized by the fact that banana production in Ugandais rainfed and the rainfall amount and distribution during the study period

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


was close to the long term average. The intra-annual rainfall variation hasimpact on production variation (bunch weight and frequency of harvesting)(Birabwa et al., 2010) and this variation was captured by monitoring plantsfor 12 consecutive months and collecting data in 2–3 spaced visits in the caseof the survey. The inter-annual rainfall variation was not captured, but a widerange of annual rainfall conditions was covered (782–1797 mm year−1) by thewide geographic spread of the study sites. Although Walworth and Sumner(1987) proposed that survey approach is desirable for determination ofnorms, estimating banana production through single visit surveys is difficultbecause production is continuous and most farmers do not use any formalway of monitoring production. However, our study used a novel techniquefor reliably quantifying bunch weights during non-destructive farm surveys.Since our study avoids temporal and spatial bias, we believe that the datacollected in this study can be used to derive norms for bananas in Uganda,with potential use of these results in the other East African highland bananaproduction area (i.e., Burundi, Rwanda, East Democratic Republic of Congo,Northwest Tanzania, and West Kenya)

The selection of the high bunch weight subpopulation and calculationof CND norms was based on methods outlined by Khiari et al. (2001b). Planttissue composition forms a d-dimensional nutrient arrangement; i.e. simplex(Sd) made of d + 1 nutrient proportions including d nutrients and a fillingvalue (Rd) defined as (Parent and Dafir, 1992):

Sd = [(N, P, K, · · · · · · , Rd): N > 0, P > 0, K > 0, · · · Rd > O, N + P + K

+ · · · + Rd = 100] (1)

where 100 is the dry matter concentration (%); N, P, K,. . . . are nutrientproportions (%); and Rd is the filling value computed as:

Rd = 100 − (N + P + K + . . . .) (2)

A geometric mean (G) computed as:

G = (N × P × K × · · · × Rd)1

d+1 (3)

is used to divide nutrient proportions to derive row-centered log ratios asfollows:

VN = ln(

NG

), VP = ln

(PG

), VK = ln

(KG

), · · · , VRd = ln

(RdG

)(4)

and

VN + VP + VK + · · · , + VRd = 0 (5)

where Vx is the CND row-centered log ratio expression for nutrient X.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


The data was next partitioned into low and high bunch weight subpop-ulations using the procedure described by Khiari et al. (2001b). The pro-cedure identifies the optimum partition between the two subpopulations.The variance ratio must be low when comparing the variance of a nutrientexpression for lower yields with that of the remainder of the population.

The observations are ranked in a decreasing bunch weight order andthe Cate-Nelson procedure was next used to iterate a partition of the databetween the two subpopulations (Khiari et al., 2001b). In the first partition,the two highest bunch weight values formed one group, and the remainder ofthe yield values formed another group; thereafter, the three highest bunchweight values formed one group, and the remainder of the yield valuesformed the other. This process was repeated until the two lowest bunchweight values formed one group, and the remainder of the bunch weightvalues formed the other. At each iteration, the numbers of observations weren1 and n2 for the first and second subpopulation, respectively. For the twosubpopulations obtained at each iteration, the variances of the row-centeredlog ratios were computed. The variance ratio of component X was thencomputed as:

f i(Vx) = Variance of Vx of n1 observationsVariance of Vx of n2 observations

(6)

where fi(Vx) is the ratio function between two subpopulations for nutrientX at the ith iteration. The first variance ratio computed from the two highestbunch weight was put on the same line as the highest bunch weight.

The cumulative function, which is the sum of the variance ratios at theith iteration from the top, was then computed as follows:

Fci (Vx) =

n1−1∑i=1

f i(Vx)

n−3∑i=1

f i(Vx)× 100 (7)

The relationship between the cumulative function Fci (Vx) and bunch weight

(Y) was defined as follows:

Fci (Vx) = aY 3 + bY 2 + cY + h (8)

where h is the intercept and a, b, and c are parameter coefficients. Theoptimum partition, which is the inflection point, was obtained by equatingthe second derivative of the previous equation to zero. The bunch weightcut off value was –b/3a. The highest bunch weight cut-off value among the

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


d + 1 nutrient computations was then selected as the lowest bunch weightin the high bunch weight subpopulation.

From the high bunch weight subpopulation, CND norms, which arethe means and standard deviations of row-centered log ratios, denoted asV∗

N, V∗P,V∗

K,. . . . V∗Rd and SD∗

N, SD∗P, SD∗

K,. . . . SD∗Rd, respectively, were then

calculated. The CND indices, denoted as IN, IP, IK,. . . . IRd were then calculatedfrom the row-centered log ratios as follows:

IN = VN − V∗N

SD∗N

, IP = VP − V∗P

SD∗P

, IK = VK − V∗K

SD∗K

, . . . . . . IRd = VRd − V∗Rd

SD∗Rd

(9)The nutrient imbalance index of a diagnosed specimen, which is its CND r2,was computed as:

r2 = I2N + I2

P + I2K + · · · · · · + I2

Rd(10)

The mean, variance and coefficient of variation (CV) for each possiblenutrient pair ratio were calculated for both low and high bunch weightsubpopulation, according to Beaufils (1973). For each nutrient pair, themean and CV of the ratio that maximized the variance ratio between thelow- and high-bunch weight group were selected as norms for that pairof nutrients, as described by Walworth and Sumner (1987) and Letzsch(1985). Increase in variance ratio results in increase in diagnostic sensitivityas it increases chances of discriminating between the two subpopulations(Walworth and Sumner, 1987).

DRIS indices were calculated, for each plot, using two steps. First, for eachpair of nutrients, observations were related to norms using standardizationand index equations (Beaufils, 1973) as shown in the example below:

f(N/P) =(

N/Pn/p

− 1) (

1000CVn/p

)if N/P > n/p (11)

f(N/P) =(

1 − n/pN/P

) (1000CVn/p

)if N/P < n/p (12)

where lower case type refers to the reference-population nutrient ratio, up-per case type refers to the sample being evaluated, f is the functional rela-tionship between the sample and reference-population values, and CV is thecoefficient of variation of the nutrient ratio in the reference population.

Next, values from the standardization equations were used to calculateindices as shown in the examples below.

IN1 = −f (P/N) + f (N/K) − f (Ca/N) − f(Mg/N

)4

(13)

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


and

IN2 = −f (P/N) + f (N/K) − f (Ca/N) − f(Mg/N

) − f (R5/N)5

(14)

where IN1 and IN2 are N indices for DRIS and DRIS where the Rd was included(DRIS-Rd), respectively.

Negative indices denoted relative deficiencies while positive indices de-noted excesses. For any sample, the nutrient indices sum to zero. The mea-sure of the total nutritional imbalance, the nutrient imbalance index (NII),was calculated using the absolute values of the indices generated for thesample as shown in the examples below:

NII1 = |IN1| + |IP1| + |IK1| + |ICa1| + ∣∣Img1∣∣ (15)

for DRIS or

NII2 = |IN2| + |IP2| + |IK2| + |ICa2| + ∣∣Img2∣∣ + ∣∣IR5

∣∣ (16)

for DRIS-Rd. The greater the sum, the more the imbalance among nutrients(Snyder and Kretschmer, 1987).

The relationships between CND and both DRIS and DRIS-Rd were ex-plored using regressions. The coefficient of determination (R2) depicted thecloseness of the relationship. Coefficient confidence intervals (95%) wereused to compare all the regressions relating CND and DRIS indices, as wellas regressions relating CND and DRIS-Rd. Regressions were assumed to besimilar if the coefficients did not differ significantly based on confidenceintervals (Akram-Ghaderi and Soltani, 2007). Graphs were used to explorefurther these relationships.

Since CND is more compatible with PCA compared to DRIS (Parentet al., 1994b), PCA were performed on row-centered logratioed nutrientvalues for the low and high bunch weight subpopulation. The principalcomponents (PCs) were varimax-rotated to obtain maximum relationshipsamong standardized variables (Vandamme et al., 1978). The PC loadingshaving values greater than the selection criteria (SC) were given significance.The SC was calculated as follows (Ovalles and Collins, 1988):

SC = 0.5/(PC eigenvalue)0.5 (17)

Selection of high bunch weight population was carried out using Mi-crosoft Office Excel 2003. Calculation of norms and indices, regression andPC analyses were carried out using SPSS for Windows, release 11.0.0, stan-dard version (SPSS Inc., Chicago, IL, USA).

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


RESULTS

The S5, i.e., six-dimensional (d + 1) simplex comprised the five nutri-ents (N, P, K, Ca, and Mg) and the filling value R5. Bunch weight rangedfrom 5 to 36 kg while foliar N, P, K, Ca, Mg, and R5 ranged between1.35–3.89, 0.10–0.38, 1.54–5.83, 0.23–1.74, 0.25–0.85 and 89.38–94.51%, re-spectively. The relationship between the concentrations of N, P, K, Ca, Mg,R5 and bunch weight showed triangular patterns (Figure 1). With increasingnutrient concentrations, there was a gradual increase in the maximum ob-served bunch weights, until a peak was reached. Further increases in nutrientconcentrations subsequently led to a gradual decrease in maximum bunchweight.

The relationship between cumulative variance function and bunchweight was cubic (data not presented) and the inflection point (-b/3a)

FIGURE 1 Relationships between foliar nutrient concentrations and bunch weight. Rd denotes thefilling value and is the sum of all nutrients with the exception of N, P, K, Ca and Mg.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


TABLE 1 Cumulative variance function [Fic(Vx)] for row-centered ratios and bunch weight (kg) at

point of inflection

Bunch weight at inflectionFi

c(VX) = ax3 + bx2 + cx + h R2 point = −b/(3a)

VN 0.007x3 − 0.261x2 − 1.891x + 100 0.96 12.2VP 0.010x3 − 0.440x2 + 0.757x + 100 0.96 14.5VK 0.009x3 − 0.384x2 + 0.129x + 100 0.98 14.4VCa 0.012x3 − 0.490x2 + 1.546x + 100 0.98 15.1VMg 0.009x3 − 0.422x2 + 1.061x + 100 0.98 15.3VR5 0.004x3 − 0.122x2 − 3.245x + 100 0.97 9.9

was highest for VMg (15.3 kg) and lowest for VR5 (9.9 kg) (Table 1). Hence,15.3 kg bunch weight was used to partition the low bunch weight subpopu-lation from the high bunch weight subpopulation. The high bunch weightsubpopulation comprised of 45.2% of all observations.

The mean bunch weight in the high bunch weight subpopulation wassignificantly (P ≤ 0.001) higher than in the low bunch weight subpopulation(Table 2). The N, P, K, Ca and Mg concentrations averaged 2.79, 0.21, 3.79,0.91 and 0.44%, respectively, in the high bunch weight subpopulation and2.66, 0.21, 3.73, 0.99 and 0.45%, respectively, in the low bunch weight sub-population. The high bunch weight subpopulation had significantly higherN (P ≤ 0.05) and lower Ca (P ≤ 0.01) compared with the low bunch weightsubpopulation. Average concentrations of other nutrients did not differ sig-nificantly (P ≤ 0.05) between the two subpopulations.

The CND norms, i.e. means and standard deviations, of VN, VP, VK, VCa,VMg andVR5, for the high bunch weight subpopulation, are presented inTable 3. The variances of the low and high bunch weight subpopulationsdiffered significantly (P ≤ 0.001) for VN. The variances for other nutrients

TABLE 2 Means and standard deviations (SD) of bunch weight (kg) and nutrient concentration (%)for low and high bunch weight subpopulations

Low bunch weight High bunch weightsubpopulation subpopulation

Mean SD Mean SD

Bunch weight 11 3 21∗∗∗ 5N 2.66 0.52 2.79∗ 0.45P 0.21 0.05 0.21 0.04K 3.73 0.78 3.79 0.62Ca 0.99 0.27 0.91∗∗ 0.26Mg 0.45 0.10 0.44 0.11

∗, ∗∗ and ∗∗∗ denote that means between low and high bunch weight subpopulation are significantlydifferent at P ≤ 0.05, 0.01 and 0.001, respectively.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


TABLE 3 The means and standard deviations (SD) of row-centered log ratios for low and high bunchweight subpopulation

Low bunch weight High bunch weightsubpopulation subpopulationa

Mean SD Mean SD

VN 0.237 0.195 0.302∗∗ 0.153VP −2.294 0.208 −2.312 0.176VK 0.574 0.211 0.607 0.168VCa −0.766 0.241 −0.859 0.289VMg −1.551 0.177 −1.546 0.199VRd 3.801 0.102 3.809 0.087

∗∗The variance ratio between low- and high-yielding subpopulations significant at P ≤ 0.01.

aThe means and SD for high bunch weight subpopulation are the Compositionl Nutrient Diagnosis(CND) norms for d = 5 nutrients.

did not differ significantly (P ≤ 0.05) between the two subpopulations. Thesenorms were used to estimate nutrient indices for N, P, K, Ca, Mg and R5 andCND r2 values using Equations (9) and (10).

The DRIS norms, i.e. means and CVs of the selected nutrient ratios, forhigh bunch weight subpopulations are presented in Table 4. The variances ofthe nutrient ratios of the low and high bunch weight subpopulation differed

TABLE 4 The mean and coefficient of variation CV (%) of nutrient ratios, for low and high bunchweight populations for ratios that maximized the variance between the low and high bunchsubpopulation

Low bunch weight High bunch weightsubpopulation subpopulationa

Mean CV (%) Mean CV (%)

P/N 0.085 41.32 0.0758∗∗∗ 27.49N/K 0.755 34.41 0.755∗ 26.18Ca/N 0.390 36.11 0.333∗ 33.30Mg/N 0.172 22.90 0.163 27.05P/K 0.0587 26.95 0.0548∗ 23.36Ca/P 4.891 32.34 4.565 32.72P/Mg 0.504 36.71 0.491 30.61Ca/K 0.281 38.21 0.249 41.27Mg/K 0.127 39.93 0.122 37.20Ca/Mg 2.281 29.20 2.101 30.34R5N 36.086 21.67 33.864∗∗∗ 17.70P/R5 0.00232 24.33 0.00223∗ 18.04R5/K 25.871 24.46 24.792 20.63Ca/R5 0.0108 27.42 0.00989 28.39R5/Mg 215.842 21.39 217.107 20.67

∗, ∗∗ and ∗∗∗ denote within row significant differences in variances between low and high bunch weightpopulations of P ≤ 0.05, 0.01 and 0.001 respectively.

aThe means and CV(%) of ratios for high bunch weight subpopulation are the DRIS norms.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


significantly (P ≤ 0.05) for P/N, N/K, P/K, Ca/N, R5/N and P/R5. For allother nutrient ratios, the variances did not vary significantly between thetwo subpopulations. These norms were used to calculate nutrient indicesand Nutrient Imbalance Indices both conventional (using Equations 13 and15) and expanded DRIS (using Equations 14 and 16).

All regressions relating CND to DRIS or DRIS-Rd were significant (P ≤0.001) (data not presented). The R2 ranged 0.965–0.979 for relationshipsbetween CND and DRIS and 0.972–0.991 for relationships between CNDand DRIS-Rd, for nutrient indices. These regressions differed significantly(95% confidence intervals) with respect to the regression coefficients, forboth DRIS and DRIS-Rd. When all nutrient indices for DRIS and DRIS-Rd were plotted against the corresponding CND indices in one graph, thedistribution of observations for the different nutrients as well as the two DRISapproaches superimposed (as they showed similar trends). Therefore, therelationship for N is presented in this paper to represent all relationships(Figure 2). The regression lines relating nutrient imbalance indices for thetwo DRIS approaches (NII1 and NII2) to CND r2 and suggested “power”relationships with R2 of 0.767 (NII1) and 0.849. (NII2) (Figure 3).

The two significant PCs identified for each PCA conducted for low andhigh bunch weight subpopulations had eigenvalues adding up to 3.641 and

FIGURE 2 Relationship between CND and Diagnosis and Recommendation Integrated System (DRISand DRIS-Rd) nutrient indices for N. The dotted lines depict the y = 0 and x = 0 relationships.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


FIGURE 3 Relationship between CND r2 and Diagnosis and Recommendation Integrated System (DRISand DRIS-Rd) nutrient imbalance indices (NII1 and NII2). NII1 and NII2 are the nutrient imbalanceindices for DRIS and DRIS-Rd, respectively. The dotted and continuous lines are regression lines for NII1and NII2, respectively, with R2 of 0.767 and 0.849, respectively.

3.3841, respectively explaining 72.81 and 67.66% of the total variance, re-spectively (Table 5). The first PC of the low bunch weight subpopulation waspositively correlated for N and Mg and negatively for P and K while that forthe high bunch weight subpopulation correlated positively for N and P and

TABLE 5 Loadings or correlations between the row-centered log ratios and the first two principalcomponents for the low and the high bunch weight subpopulation, extracted from varimax normalizedmatrix

Low bunch weight High bunch weightsubpopulation subpopulation

PC1 PC2 PC1 PC2

VN 0.635∗ −0.656∗ −0.218 0.781∗VP −0.769∗ 0.033 0.834∗ −0.043VK −0.780∗ −0.289 0.714∗ 0.346VCa 0.272 0.903∗ −0.437∗ −0.795∗VMg 0.797∗ 0.040 −0.755∗ 0.067Eigen value 2.310 1.330 2.015 1.368Selection criteria 0.329 0.434 0.352 0.427% variance 46.21 26.61 40.30 27.36

∗Significance loading.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


negatively for Ca and Mg. For both subpopulations N and Ca had an inverserelationship in the second PC.

DISCUSSION

Relationships between Nutrient Concentrations

and Bunch Weight

The triangular distribution patterns observed between foliar nutrientconcentrations and bunch weights (Figure 1) suggest that the latter are in-fluenced by several (interacting) factors and not by single nutrients only.Similar conclusions were made in Quesnel et al. (2006), Walworth andSumner (1987) and Vizycayno-Soto and Cote (2004) where such relation-ships were observed between foliar nutrients and production. The upperboundaries of the data clouds expressing the maximum bunch weight at agiven nutrient concentration show a typical positive relation between yieldand nutrient content at low nutrient concentrations and a negative relationat very high nutrient concentrations. Increase in concentration of a defi-cient nutrient leads to increases in yield up to a maximum beyond whichyield declines (Havlin et al., 2004). The decline in yield due to excessivenutrient concentration is due to toxicity or reduction in concentrations ofother nutrients below their critical ranges (Havlin et al., 2004).

The fact that partitioning data at the highest inflection point of15.3 kg (Table 1) placed 45.2% of the observations in the high bunch weightpopulation suggested that the Cate-Nelson approach, as used by Khiariet al. (2001a, 2001b, 2001c), was applicable to this study. In other studies,the approach has not been always applicable for partitioning data. For ex-ample, in studies on nopal (Magallanes-Quintanar et al., 2004) and Aloe vera(Garcıa-Hernandez et al., 2006), the highest inflection points were beyondthe range of observations.

The fact that the high bunch weight subpopulation had significantlyhigher N (P ≤ 0.05) (Table 2) is in agreement with observations byWalworth and Sumner (1987) on maize data collected worldwide. They con-cluded that that low yields are more frequently associated with low N thanwith high N because farmers are more likely to apply inadequate nutrientsconcentrations than to apply excessive amounts. The significantly higher Ca(P ≤ 0.01) in the low bunch weight subpopulation compared with the highbunch weight subpopulation may be due to a negative interaction betweenN and Ca in the plant. Still, these differences in nutrient concentrations,between the two subpopulations, justify use of the high bunch weight sub-population to derive norms instead of using all the observations. Using thehigh yield subpopulation to derive norms assumes that the distribution ofobservations at low yields is skewed (Walworth and Sumner, 1987).

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


CND and DRIS Norms

The mean nutrient concentrations observed in the high bunch weightsubpopulation (Table 2) did not fully tally with those observed in otherstudies. Compared with this study, Wortmann et al. (1994) had higher N(3.15 vs. 2.79%), P (0.25 vs. 0.21%) and Ca (1.13 vs. 0.91%), and lowerK (3.04 vs. 3.84%) for AAA-EA cultivars, while for Robusta (AAA) and NeyPoovan (AB) dessert bananas, Raghupathi et al. (2002) had lower P (0.17%),and Angeles et al. (1993) had higher K (4.49%). Furthermore, the normsderived by Wortmann et al. (1994), were based on data collected in oneeco-region (i.e. Kagera in Northwest Tanzania) and may therefore not beapplicable outside this region. On the other hand, the norms derived in thisstudy are based on data from a wide range of ecological regions and maytherefore have a wider application.

Previous studies have suggested that norms may differ between genomegroups. For example, DRIS norms derived for Gros Michel (an AAA triploid),Ney Poovan (AB) and a cultivar thought to be Pisang Awak (ABB) seemedto differ (Bosch et al., 1996). Similarly, P. J. A. van Asten and S. Gaidashova(unpublished data, 2009) observed that foliar nutrient concentrations forPisang Awak (ABB) were consistently lower for N (15%) and Ca (36%) whencompared to Intuntu (AAA-EA) in 12 sites across Rwanda. Also, K and Caconcentrations differed significantly (P ≤ 0.05) between Robusta and NeyPoovan (Raghupathi et al., 2002). These differences further highlight theimportance of the norms calculated in this study for diagnosing nutrientimbalances in the East African Highland banana as they are specific to thegenome group AAA.

Although the norm for K derived in this study seems to differ thosecalculated by Wortmann et al. (1994) and Angeles et al. (1993), it seems tobe in agreement with observations made in other banana studies in Uganda.Based on their study, Smithson et al. (2001) suggested that the norm forK in banana is lower than the 4.49 suggested by Angeles et al. (1993) andhigher than the 3.04 suggested by Wortmann et al. (1994) Other bananastudies also seem to agree with this suggestion. For example, in a nematodetrial in Uganda, in 1997 plots with mulch had had significantly higher Kconcentrations, averaging 3.7, compared with plots that were not mulched,which averaged 2.2 and 2.3 (McIntyre et al., 2000). Production was alsosignificantly higher in the mulched plots compared to those without mulch.In another trial where application of K was reported to increase annual yieldfrom 14.4 to 18.9 t ha−1 year−1, foliar K concentrations in plots where K wasapplied averaged 3.3% (Smithson et al., 2004). Hence, it is likely that ournorm for K could be more appropriate for diagnosing nutrient imbalancesin the East African Highland banana production systems than norms derivedin previous studies.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


Furthermore, the fact that the low and high bunch weight subpopulationdiffered in variances of CND calculated Vn (Table 3), as well as DRIS calcu-lated P/N, N/K, Ca/N, P/K, R5/N and P/R5 (Table 4) implied that nutrientimbalances contributed to the differences in bunch weight between the twopopulations. Since Walworth and Sumner (1987) suggested that, for DRIS,discrimination between the two subpopulations is maximized when the ratioof variances between the two subpopulations is maximized, it is likely thatthis suggestion may also be true for CND. We therefore conclude that theCND and DRIS norms calculated in this study are reliable for diagnosingnutrient imbalances in the East African highland bananas.

The fact that CND and DRIS norms developed in this study were closelyrelated is in line with findings in earlier studies on tomato (Parent et al.,1993), potato (Parent et al., 1994a) and sweet corn (Khiari et al., 2001a) Theregressions relating CND to DRIS and DRIS-Rd had high R2 (0.965–0.991)for all nutrients and showed similar trends (Figure 2). Interestingly, the factthat regressions relating CND to DRIS or DRIS-Rd differed significantly (95%confidence intervals) with respect to regression coefficients suggested thatthe order of nutrient imbalances calculated for an observation may differbetween CND and DRIS as reported in the study by Kumar et al. (2003).For example (data not presented), the order of imbalances in one plot incentral region was P (−1.24) <K (−0.28) < Ca (0.17) <N (0.33) <Mg (1.31)for CND, P (−13.01) < Ca (−1.32) <K (−0.47) <N (2.80) <Mg (11.99) forDRIS, and P (−12.20) < Ca (−0.45) <K (0.16) <N (4.15) <Mg (14.0) forDRIS-Rd. Although the order differed between CND and DRIS or DRIS-Rd,all methods identified P as being most deficient, Mg as least deficient and Nas the second least deficient. This suggests that the differences between themethods may be minimal. This is further supported by the fact that all threeapproaches did not seem to differ in categorizing of observations as eitherdeficient or excess (Figure 2).

The closeness in relationships between CND to DRIS or DRIS-Rd

(R2 = 0.965–0.991) suggests that both DRIS and DRIS-Rd methods give sim-ilar diagnosis. On the other hand, the fact that the R2 relationship betweenDRIS-Rd and CND r2 had a slightly higher adjusted R2 (0.849) compared tothe relationship between DRIS and CND r2 (0.767) (Figure 3) suggests thatthe DRIS-Rd may be superior. Nonetheless, we did not find strong evidenceto suggest that the inclusion of the filling value in DRIS would be requiredwhen identifying nutrient deficiencies with the aim to develop new fertilizerrecommendations for East African highland bananas.

Nutrient Interactions

The positive relationship between P and K and the negative relationshipbetween these nutrients and Mg as well as the negative relationship betweenN and Ca in the two subpopulations (Table 5) suggest the positive P-K

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


interaction and negative interactions: N-Ca, P-Mg and K-Mg interaction.In addition, the positive N-Mg interaction and the negative N-P and N-Kinteractions were only specific to the low bunch weight subpopulation whilethe positive Ca-Mg interaction and negative P-Ca and K-Ca interactions werespecific to the high bunch weight subpopulation. These results partially agreewith findings in dessert bananas by Raghupathi et al. (2002) where the firstPC of the high yield subpopulation was positively correlation for N, P andK and negatively for Ca and Mg. The observation that some of the nutrientinteractions were subpopulation specific is in agreement with studies in Aloevera (Garcıa-Hernandez et al., 2006) and yellow pepper (Garcıa-Hernandezet al., 2004).

Although the underlying mechanisms of nutrient interactions are notalways well understood, some of the interactions have been explored. Forexample, ammonium (NH4

+)/nitrate(NO3+) increase in growth solution

contributes to decrease foliar K+ and Ca2+ (Grattan and Grieve, 1999). Theantagonism between P and Ca has been attributed to the reduction in theactivity of P due to ionic strength effects and the low solubility of the Ca-Pminerals (Grattan and Grieve, 1999).

The fact that all nutrients are either positively or negatively correlatedto at least one other nutrient suggests that these nutrients do not influencebanana production in isolation. This implies that nutrient uptake and distri-bution in the East African highland banana is influenced interactions whichis in agreement with observations made on dessert bananas (Raghupathiet al., 2002), Aloe vera (Garcıa-Hernandez et al., 2006) yellow pepper (Garcıa-Hernandez et al., 2004) and other crops. Since in both subpopulations, allnutrients except Ca in the low bunch weight subpopulation and N in thehigh bunch weight were correlated to more than one other nutrient, themultivariate approach (i.e., CND) would have been expected to be superiorto the dual ratio approach (i.e. DRIS). Interestingly, this study showed thatthe differences in approach between the CND and DRIS did not lead to largedifferences in identified deficiencies. Nonetheless, CND may in practice stillbe the preferred approach, since norms and indices proved much easier tocompute in CND compared with DRIS.

CONCLUSIONS

Bunch weights are determined by several interacting factors and not bysingle nutrients only. Both CND and DRIS norms derived in this study provedto be reliable tools for diagnosing nutrient imbalances in the East AfricanHighland bananas. Inclusion of a filling value (Rd) in DRIS improved the cor-relation with the CND approach, but overall differences between the CND,DRIS and DRIS-Rd methods were small. Comparison with earlier nutrientstudies on bananas revealed that norms developed using data from specific

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


eco-regions and cultivars are not necessarily applicable to other regions andcultivars. Nutrient interactions showed that the multivariate approaches ofCND and DRIS are better in diagnosis of nutrient imbalances in bananaproduction than single nutrient approaches. Given the wide range of agro-ecologies sampled in this study, we believe that the norms presented will pro-vide a good indication for nutrient deficiencies for most of the East Africanhighland region. Identification of nutrient imbalances will help farmers toprioritize investments in nutrient inputs that target the primary nutrientdeficiencies. The development of site/region-specific fertilizer recommen-dations instead of blanket recommendations will help farmers to increasefarm production and profitability. This will improve household food securityand income in these smallholder systems.

ACKNOWLEDGMENTS

The authors are grateful to the host farmers on whose plots the data wascollected. We also thank Symon Wandiembe for suggestions and supportwith some statistical aspects of this paper, NARO for laboratory analysis of soiland foliar samples, and Godfrey Taulya for support with quality maintenanceof laboratory results. The authors are also grateful to Professors M. Tenywaand M. Bekunda of Makerere University, Uganda, for supporting the firstauthor’s PhD research of which this paper is a component. This research wasfunded by USAID through Agricultural Productivity Enhancement Program(APEP).

REFERENCES

Akram-Ghaderi, F., and A. Soltani. 2007. Leaf area relationships to plant vegetative characteristics incotton (Gossypium hirsutum L.) grown in a temperate sub-humid environment. International Journalof Plant Production 1: 63–71.

Angeles, D. E., M. E. Sumner, and E. Lahav. 1993. Preliminary DRIS norms for banana. Journal of PlantNutrition 16: 1059–1070.

Beaufils, E. R. 1973. Diagnosis and Recommendation Integrated System (DRIS). Soil Science BulletinNo. 1. Durban, South Africa: University of Natal.

Birabwa, R., P. J. A. van Asten, I. N. Alou, and G. Taulya. 2010. Got matooke for Christmas. Acta Horticul-turae 879: 113–121.

Bosch, C., A. Lorkeers, M. R. Ndile, and E. Sentozi. 1996. Diagnostic survey: Constraints to bananaproductivity in Bukoba and Muleba Districts, Kagera region, Tanzania. Working Paper No. 8. Tan-zania/Netherlands Farming Systems Research Project, Lake Zone. Ari Muruku, Bukoba, Tanzania.

Camara, O., and E. Heinmann. 2006. Overview of the fertilizer situation in Africa. Background paperprepared for the African fertilizer summit. Abuja, Nigeria, 9–13 June, 2006.

East African Meteorological Department. 1963. Climatic seasons of East Africa. East African Meteorolog-ical Department Report No. 8.

Edmeades, S. 2006. Varieties, attributes and marketed surplus of a subsistence crop: Bananas in Uganda.Paper presented at the International Association of Agricultural Economists conference, Gold Coast,Australia, 12–18 August 2006.

FAO. 2009. FAOSTAT. Available at: http://faostat.fao.org/site/567/default.aspx (Accessed 26 February2009).

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


Garcıa-Hernandez, J. L., R. D. Valdez-Cepeda, B. Murillo-Amador, F. A. Beltran-Morales, F. H. Ruiz-Espinoza, I. Orona-Castillo, A. Flores-Hernandez, and E.Troyo-Dieguez. 2006. Preliminary compo-sitional nutrient diagnosis norms in Aloe vera L. grown on calcareous soil in an arid environment.Environmental and Experimental Botany 58: 244–252.

Garcıa-Hernandez, J. L., R. D. Valdez-Cepeda, B. Murillo-Amador, A. Nieto-Garibay, L. F. Beltran-Morales,R. Magallanes-Quintanar, and E. Troyo-Dieguez. 2004. Compositional nutrient diagnosis and mainnutrient interactions in yellow pepper grown on desert calcareous soils. Journal of Plant Nutritionand Soil Science 167: 509–515.

Grattan, S. R., and C. M. Grieve. 1999. Salinity-mineral nutrient relations in horticultural crops. ScientiaHorticulturae 78: 127–157.

Hallmark, W. B., and R. B. Beverly. 1991. Review – An update in the use of the Diagnosis and Recom-mendation Integrated System. Journal of Fertilizer Issues 8: 74–88.

Hallmark, W. B., J. L. Walworth, M. E. Sumner, C. J. deMooy, J. Pesek, and K. P. Shao. 1987. Separatinglimiting from non-limiting nutrients. Journal of Plant Nutrition 10: 1381–1390.

Hanson, E. J. 1987. Fertilizer use efficiency: Integrating soil tests and tissue analysis to manage thenutrition of highbush blueberries. Journal of Plant Nutrition 10: 1419–1427.

Havlin, J. L., J. D. Beaton, S. L. Tisdale, and W. L. Nelson. 2004. Soil Fertility and Fertilizers: An Introductionto Nutrient Management. Singapore: Pearson Education.

Heaton E., T. Voigt, and S. P. Long. 2004. A quantitative review comparing the yields of two candidateC4 perennial biomass crops in relation to nitrogen, temperature and water. Biomass and Bioenergy27:21–30.

Holland, D. A. 1966. The interpretation of leaf analyses. Journal of Horticultural Science and Biotechnology41: 311–329.

Khiari, L., L. E. Parent, and N. Tremblay. 2001a. Critical compositional nutrient indexes for sweet cornat early growth stage. Agronomy Journal 93: 809–814.

Khiari, L., L. E. Parent, and N. Tremblay. 2001b. Selecting the high-yielding subpopulation for diagnosingnutrient imbalance in crops. Agronomy Journal 93: 802–808.

Khiari, L., L. E. Parent, and N. Tremblay. 2001c. The phosphorus compositional nutrient diagnosis rangefor potato. Agronomy Journal 93: 815–819.

Kumar, P. S. S., S. A. Geetha, P. Savithri, R. Jagadeeswaran, and P. P. Mahendran. 2003. Diagnosis ofnutrient imbalances and derivation of new RPZI (Reference Population Zero Index) values usingDRIS/MDRIS and CND approaches in the leaves of turmeric (Curcuma longa). Journal of AppliedHorticulture 5: 7–10.

Lahav, E. 1995. Banana nutrition. In: Bananas and Plantains, ed. S. Gowen, pp. 258–316. London: Chap-man and Hall.

Letzsch, W. S. 1985. Computer program for selection of norms for use in the diagnosis and recom-mendation integrated system (DRIS). Communications in Soil Science and Plant Analysis 16: 339–347.

Magallanes-Quintanar, R., R. D. Valdez-Cepeda, F. Blanco-Macıas, M. Marquez-Madrid, R. R. Ruız-Garduno, O. Perez-Veyna, J. L. Garcıa-Hernandez, B. Murillo-Amador, J. D. Lopez-Martınez, andE. Martınez-Rubın de Celis. 2004. Compositional nutrient diagnosis in nopal (Opuntia ficus-indica).Journal of the Professional Association for Cactus Development 6: 78–89.

McIntyre, B. D., P. R. Speijer, S. J. Riha, and F. Kizito. 2000. Effects of mulching on biomass, nutrients,and soil water in banana inoculated with nematodes. Agronomy Journal 92: 1081–1085.

Mokany, K., R. J. Raison, and A. S. Prokushkin. 2006. Critical analysis of root:shoot ratios in terrestrialbiomes. Global Change Biology 12: 84–96.

Okalebo, J. R., K. W. Gathua, and P. L. Woomer. 1993. Laboratory Methods of Soil and Plant Analysis:A Working Manual. Nairobi: Tropical Soil Biology and Fertility institute.

Okech, S. H., P. J. A. van Asten, C. S. Gold, and H. Ssali. 2004. Effects of potassium deficiency, droughtand weevils on banana yield and economic performance in Mbarara, Uganda. Journal of AgriculturalScience 9: 511–519.

Ovalles, F. A., and M. E. Collins. 1988. Variability of northwest Florida soils by principal componentanalysis. Soil Science Society American Journal 52: 1430–1435.

Parent, L. E., A. N. Cambouris, and A. Muhawenimana. 1994a. Multivariate diagnosis of nutrient imbal-ance in potato crop. Soil Science Society of America Journal 58: 1432–1438.

Parent, L. E., and M. Dafir. 1992. A theoretical concept of compositional nutrient diagnosis. Journal of theAmerican Society for Horticultural Science 117: 239–242.

Dow

nloa

ded

by [

Nor

th C

arol

ina

Stat

e U

nive

rsity

] at

16:

34 0

7 Se

ptem

ber

2012


Parent, L. E., D. Isfan, N. Tremblay, and A. Karam. 1994b. Multivariate nutrient diagnosis of the carrotcrop. Journal of the American Society for Horticultural Science 119: 420–426.

Parent, L. E, A. Karam, and S. A. Visser. 1993. Compositional nutrient diagnosis of the greenhousetomato. Hortscience 28: 1041–1042.

Quesnel, P. O., B. Cote, J. W. Fyles, and A. D. Munson. 2006. Optimum nutrient concentrations and CNDscores of mature white spruce determined using a boundary line approach and spatial variation oftree growth and nutrition. Journal of Plant Nutrition 29: 1999–2018.

Raghupathi, H. B., B. M. C. Reddy, and K. Srinivas. 2002. Multivariate diagnosis of nutrient imbalance inbanana. Commununications in Soil Science and Plant Analysis 33: 2131–2143.

Raghupathi, H. B., Y. T. N. Reddy, M. Kurian Reju, and B. S. Bhargava. 2005. Diagnosis of nutrientimbalance in mango by DRIS and PCA approaches. Journal of Plant Nutrition 27: 1131–1148.

Samson, J. A. 1992. Tropical Fruits. London: Longman.Schluter, T. 2008. Geological Atlas of Africa: With Notes on Stratigraphy, Tectonics, Economic Geology, Geohazards,

Geosites and Geoscientific Education of Each Country. Berlin: Springer.Smithson, P. C., B. D. McIntyre, C. S. Gold, H. Ssali, and I. N. Kashaija. 2001. Nitrogen and potassium

fertilizer vs. nematode and weevil effects on yield and foliar nutrient status of banana in Uganda.Nutrient Cycling in Agroecosystems 59: 239–250.

Smithson, P. C., B. D. McIntyre, C. S. Gold, H. Ssali, G. Night, and S. Okech. 2004. Potassium andmagnesium fertilizers on banana in Uganda: Yields, weevil damage, foliar nutrient status and DRISanalysis. Nutrient Cycling in Agroecosystems 69: 43–49.

Snyder, G. H., and A. E. Kretschmer. 1987. A DRIS analysis for bahiagrass pastures. Soil and Crop ScienceSociety of Florida Proceedings 47: 56–59.

van Asten, P. J. A., C. S. Gold, J. Wendt, D. De Waele, S. H. O. Okech, H. Ssali, and W. K. Tushemereirwe.2005. The contribution of soil quality to yield and its relation with other banana yield loss factors inUganda. In: Proceedings of a Workshop Held on Farmer Participatory Testing of IPM Options for SustainableBanana Production in Eastern Africa, Seeta, Uganda, 8–9 December 2003, eds. G. Blomme, C. S. Gold, andE. Karamura, pp. 100–115. Montpellier, France: International Plant Genetic Resources Institute.

van Asten, P. J. A., L. W. I. Wairegi, F. Bagamba, and C. Drew. 2010. Factors driving fertilizer adoption inbanana systems in Uganda. Acta Horticulturae 879: 465–477.

Vandamme, D., I. Scheys, and D. Lamberts. 1978. The optimum fertility conditions for maximizing yieldand quality of tomatoes in the open field. Scientia Horticulturae 8: 199–212.

Vizycayno-Soto, G., and B. Cote. 2004. Boundary-line approach to determine standards of nutritionfor mature trees from spatial variation of growth and foliar nutrient concentrations in naturalenvironments. Communications in Soil Science and Plant Analysis 35: 2965–2985.

Wairegi, L. W. I., P. J. A. van Asten, C. Kiwanuka, M. Tenywa, and M. Bekunda. 2007. Assessment ofsoil management practices in East African highland cooking banana (Musa spp. AAA-EA) systemsin Uganda. Paper presented at the African Network of for Soil Biology and Fertility InternationalSymposium. Arusha, Tanzania 17–21 September 2007.

Wairegi, L. W. I., P. J. A. van Asten, M. Tenywa, and M. Bekunda. 2009. Quantifying bunch weights ofthe East African Highland bananas (Musa spp. AAA-EA) using non-destructive field observations.Scientia Horticulturae 121:63–72.

Walworth, J. L., and M. E. Sumner. 1987. The diagnosis and recommendations integrated system (DRIS).Advances in Soil Sciences 6: 149–188.

Wortmann C. S., C. H. Bosch, and L. Mukandala. 1994. Foliar nutrient analysis in bananas grown in thehighlands of East Africa. Journal of Agronomy and Crop Science 172: 223–226.D

ownl

oade

d by

[N

orth

Car

olin

a St

ate

Uni

vers

ity]

at 1

6:34

07

Sept

embe

r 20

12

norms for multivariate diagnosis of nutrient imbalance in the east african highland bananas (musa...

Documents