You are on page 1of 9

Crop Protection 67 (2015) 26e34

Contents lists available at ScienceDirect

Crop Protection
journal homepage: www.elsevier.com/locate/cropro

Standard area diagrams for assessment of powdery mildew severity


on tomato leaves and leaflets
Daniel A. Costa Lage a, 1, Waldir A. Marouelli b, Henrique da S. S. Duarte c,
-Filho a, *
Adalberto C. Cafe
a
Departamento de Fitopatologia, Universidade de Brasília, 70910-900, Brasília, DF, Brazil
b
Embrapa Hortaliças, 70359-970, Brasília, DF, Brazil
c
Departamento de Proteça~o de Plantas, Universidade Federal do Parana, 80035-050, Curitiba, PR, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Two standard area diagrams (SADs) to assess the severity of powdery mildew (Leveillula taurica) on
Received 30 May 2014 tomato leaves and leaflets were developed and evaluated. The SADs are composed of two sets of images
Received in revised form of leaves and leaflets with six distinct percentages of diseased areas (1, 5, 10, 20, 40 and 60%). Diagrams
19 September 2014
were evaluated by 16 raters divided into two groups, based on their former experience in assessing plant
Accepted 20 September 2014
Available online
disease intensities, using the two sets of images, first without SADs and then using the proposed SADs.
Lin's concordance correlation analysis of estimated vs. actual disease severities showed that precision
and accuracy were clearly improved when raters were aided by the proposed diagrams, for both leaflets
Keywords:
Solanum lycopersicum
and leaves. Based on coefficients of determinations (R2) and on the intra-class correlations (r), the es-
Tomato powdery mildew timates of severity were consistently more reliable when using SADs. Furthermore, when the scales were
Oidiopsis haplophylli adopted, the performance of the inexperienced group improved to the same level of that of the expe-
Disease quantification rienced group. The diagrams improved subjects' ability to accurately, precisely and reliably estimate
Epidemiology tomato powdery mildew severity, and as such can be used to assess severity for studies in epidemiology,
quantitative host resistance and in the evaluation of management practices in this important
pathosystem.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction However, their use alone or in combination with one another


creates an ideal environment for severe epidemics of powdery
Tomato (Solanum lycopersicum L.) is widely cultivated mildew (Reis et al., 2007), due to the fact that the presence of leaf
throughout the world, but the crop is challenged by a large wetness actually reduces infection by the Erysiphaceae, the causal
number of plant diseases, which strongly compromise yield (Jones agents of powdery mildew (Sutton and Jones, 1979). Powdery
et al., 1991). The use of chemicals has been the primary manage- mildews constitute an important group of fungal phytopathogens
ment practice to reduce disease risk, but other management that are adapted to drier environments and are widely distributed
practices are also available (Drinkwater et al., 1995). A basic factor throughout the world.
that affects disease intensity in most foliar pathosystems is leaf Tomato is host to more than one species that causes powdery
wetness (Rotem and Palti, 1969), and among the agronomic mildew, including the epiphytic anamorphic forms Oidium lyco-
practices aimed at reducing it are the choice of furrow or drip persici Cook & Massee emend. and Oidium neolycopersici Kiss (Kiss
irrigation, dry season cropping and cultivation in protected envi- v.) Arnaud (anamorph,
et al., 2001). However, Leveillula taurica (Le
ronments (Diver et al., 1999). These measures hinder infection Oidiopsis haplophylli Rulamort) is the most prevalent and
efficiency of the majority of foliar pathogens and have been economically damaging powdery mildew agent to tomato (Correll
increasingly adopted by tomato growers (e.g. Cabral et al., 2013). et al., 1988). L. taurica is an important pathogen on the crop in
dry conditions, when air relative humidity (RH) is between 50 and
* Corresponding author. Tel.: þ55 61 31072917; fax: þ55 61 31072931. 70%, and temperatures are between 20 and 25  C (Guzman-Plazola
E-mail addresses: cafefilh@unb.br, adalberto.cafe@pesquisador.cnpq.br et al., 2003; Palti, 1988).
-Filho).
(A.C. Cafe The symptoms caused by L. taurica are distinct from those
1
Present address: Bayer CropScience e SHIS QI 07 Bloco G Loja 102-B e Caixa
usually associated with other members of the Erysiphaceae that
Postal 7048, 71645-970, Brasília, DF, Brazil.

http://dx.doi.org/10.1016/j.cropro.2014.09.014
0261-2194/© 2014 Elsevier Ltd. All rights reserved.
D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34 27

cause powdery mildew. In the tomato plant, symptoms are initi- should be identified and the disease severity in the field must be
ated by small chlorotic lesions that turn yellowish and become ascertained. Furthermore, the SAD must be quantitatively eval-
visible on the adaxial surface of the leaf, which corresponds with uated before being recommended as a tool for improving disease
the colonization of the fungus on the abaxial surface and profuse severity estimates.
sporulation (Reis et al., 2005). Under favorable conditions, the However, despite the economic importance of the tomato crop
inoculum generated in successive cycles of the disease, results in and the high losses incurred by powdery mildew, so far no SAD to
rapid increase of the affected leaf area. Old lesions become evaluate Leveillula on tomato has been proposed. Both the
necrotic and are generally surrounded by chlorotic leaf tissue complexity of the symptoms, which evolves over time and com-
(Palti, 1988). Losses are further compounded when fruit yield of prises leaf chlorosis and necrosis, as well as the complex foliar
severely affected plants is lost by sun scald (Jones and Thomson, morphology of the tomato plant, may partially explain why no such
1987). Under favorable conditions, the estimated fruit yield los- scale exists for this pathosystem. None of the most up-to-date
ses due to Leveillula powdery mildew may reach 31% (Jones and studies that discuss epidemiology and control of tomato powdery
Thomson, 1987). In dry regions or in the greenhouse, the path- mildew (Guzma n-Plazola et al., 2011; Kasselaki et al., 2006;
ogen presents a constant threat to growers, and a pressing need Konstantinidou-Doltsinis et al., 2006) mention the use of dia-
for epidemiological and management studies to reduce disease grams as a tool for evaluating disease severity.
risk. The goal of this study was to develop and evaluate a set of SADs
So that quantitative studies on the epidemiology and man- encompassing the full range of disease severities as found in the
agement of powdery mildew of tomato may achieve the highest field, as an aid to raters estimating tomato powdery mildew,
possible quality in terms of accuracy and precision, it is essential bearing in mind the tomato plant's leaf anatomy, as well as the
to have accurate, precise and reliable disease estimates. The ac- disease pathogenesis and the symptoms of chlorosis and necrosis.
curacy of an estimate has been defined as how close the estimate
is to the true value (Nutter and Schultz, 1995), while precision is 2. Materials and methods
defined by the variability of the estimates (Yadav et al., 2013),
and reliability is associated to the extent to which the same es- 2.1. Development of the SAD sets to assess the severity of powdery
timate obtained under different conditions yields similar results mildew on tomato leaves and leaflets
(Nutter et al., 1991). In order to compare estimated and actual
values, Madden et al. (2007) discussed the concept of agreement, Two hundred naturally field-infected tomato leaflets and leaves,
defined as the product of precision and accuracy. Several that encompassed the widest possible range of disease severities
methods exist for determining and improving the precision, ac- were collected, during two tomato cropping seasons in Mid-West
curacy and reliability of disease assessments (Bock et al., 2010; Brazil. Images of each leaflet and leaf were captured digitally
Madden et al., 2007; Nutter et al., 1993), including the use of against a black background using a Hewlett Packard Photosmart
standard area diagrams (SADs). SADs are illustrations of diseased scanner (Model C3180, Sa~o Paulo, Brazil) with an image resolution
plant parts, within a range of severity values that are used as of 300 dpi, and individually processed with the aid of Quant soft-
references to which to compare samples when estimating dis- ware version 1.0.2 (Vale et al., 2003). The necrotic and chlorotic
ease severity. For this, representative symptoms of the disease diseased areas were measured separately and the proportion of

Fig. 1. Standard area diagrams for powdery mildew (Leveillula taurica) severity on tomato (Solanum lycopersicum L.) leaflets. The symptoms of chlorosis and yellowing are rep-
resented in light gray, while the symptoms of necrosis are shown in dark gray. The numbers represent actual percentage of the leaflet area showing the sum of both chlorotic and
necrotic symptoms of the disease.
28 D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34

diseased leaf or leaflet tissue was estimated using Quant. The estimated severity (Y) and actual severity (X), which measures
diseased area determined electronically was considered the actual precision (variation) or the scattering of points around the best-
disease severity and used as the reference for evaluating the ac- fitting line. Cb is estimated as: Cb ¼ 2/[(y þ 1/y þ u2)], where
curacy and precision of assessments performed with and without y ¼ sy/sx, and sy and sx are the standard deviations of Y and X,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the SADs. Representative images were used to develop two SAD respectively; and y ¼ ðmy  mx Þ= ðsy $sx Þ, where my and mx are the
sets, one for the leaves and the other for leaflets. Each image on the mean values of Y and X, respectively. The term y measures the
diagram represents an actual image vectorized in graphic software difference in slope between the fitted line and the line of
(CorelDRAW version X4), seeking to maintain the actual form and concordance (scale bias). Equal slopes would have a y of 1. The
size of the lesions. Symptoms of chlorosis were represented in light term u is a measure of the location shift reflecting height dif-
gray, while symptoms of necrosis were shown in dark gray. ferences between the fitted line and the line of concordance
(location bias). Equal heights would have a u of 0. When there is
2.2. Validation of the SAD sets to assess the severity of powdery perfect agreement between estimated and actual severity, the
mildew on tomato leaflets and leaves points fall on the concordance line and my ¼ mx, sy ¼sx, r ¼ 1, and
by definition, y ¼ 1, u ¼ 0, Cb ¼ 1, and rc ¼ 1 (Bock et al., 2010;
In order to validate the diagram sets to tomato, 16 subjects were Nita et al., 2003).
recruited, divided into two equal groups of inexperienced and Inter-rater reliability of the estimates was determined in two
experienced raters, according to their knowledge of disease quan- ways. First, using the coefficient of determination (R2) from linear
tification. The group of inexperienced raters was composed of eight regression analyses of relationships between severity estimates for
undergraduates who had never assessed severity of plant diseases, all pairs of raters (Nutter and Schultz, 1995). Second, using the
while the group of experienced raters was made up of eight post- intracluster (or intra-class) correlation (r) for all raters combined,
graduate students familiar with disease severity assessments. as described by Shoukri and Pause (1999) and Nita et al. (2003).
Each of the SADs was validated at separate times, but followed the Precision was also determined with analysis of the absolute error
same procedures. The first stage involved each rater assessing the (estimated severity minus actual severity).
severity of powdery mildew of 42 randomly-presented images of For all parameters analyzed (r, Cb, y, u, and rc) and for the inter-
tomato leaves or leaflets with various levels of the disease. Later, rater reliability (R2), the differences between means (i.e. with SADs
the same images were presented for disease estimation, in a minus without SADs) were calculated and submitted to an equiv-
different random order, and raters were again asked to assess dis- alence test for significance (Bardsley and Ngugi, 2013; Yadav et al.,
ease severity with the help of the proposed diagrams. 2013; Yi et al., 2008). The equivalence test was used to calculate the
Accuracy and precision (agreement) of the estimates by each 95% confidence intervals (CIs) for each statistic (the difference be-
rater, with and without the use of the SADs, was determined tween the means) by bootstrapping using the percentile method
based on Lin's concordance correlation coefficient (LCCC, rc) (Lin, (with an equivalence test, the null hypothesis is the converse of H0,
1989), as described by Nita et al. (2003). The LCCC combines i.e. the null hypothesis is non-equivalence). All analyses were based
measures of accuracy and precision to assess the relational fit of on 2000 balanced bootstrap samples using PROC SURVEY SELECT
the data to the line of concordance (45 ), and is defined as: and calculating 95% CI using PROC UNIVARIATE (SAS Institute Inc.,
rc ¼ r  Cb, where Cb is a bias correction factor that measures Cary, NC, USA). If the CIs embraced zero, the difference was
how far the best-fitting line deviates from 45 and is thus a considered non-significant (a ¼ 0.05).
measure of accuracy, and r is the correlation coefficient between

Fig. 2. Standard area diagrams for powdery mildew (Leveillula taurica) severity on tomato (Solanum lycopersicum L.) leaves. The symptoms of chlorosis and yellowing are repre-
sented in light gray, while the symptoms of necrosis are shown in dark gray. The numbers represent actual percentage of the leaf area showing the sum of both chlorotic and
necrotic symptoms of the disease.
D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34 29

No SAD SAD used


100 100
A B
B
80 80

60 60

40 40

20 20

0 0

100 100
C D
80 80

60 60

40 40
Estimatedseverity(%)

20 20

0 0

100 100
E F
80 80

60 60

40 40

20 20

0 0

100 100
G H
80 80

60 60

40 40

20 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100

Actual severity (%)


Fig. 3. Relationship between actual and estimated severity of powdery mildew (Leveillula taurica) of tomato (Solanum lycopersicum L.) without and with the use of a set of standard
area diagrams (SADs) for 42 diseased leaflets (A, B, C and D) and 42 diseased leaves (E, F, G and H). A, B, E and F represent the 8 raters with experience and C, D, G and H represent
the 8 raters without experience in disease severity assessment. The solid line represents the best-fitting line, whereas the dotted line is the concordance line, which represents
perfect agreement between actual and estimated severity (slope of 1, intercept of 0).

3. Results Fig. 3 B, D). Accuracy, precision and agreement were generally


improved. Although the improvement was slight in the location
3.1. Standard area diagram set to assess the severity of powdery bias (Table 1, u ¼ 0.14 and 0.03 without and with SADs, respec-
mildew on tomato leaflets tively), the statistical parameters r, Cb, y, and rc were all signifi-
cantly improved when raters used the diagrams to estimate
3.1.1. Effect of leaflet SADs on accuracy and precision disease severity. The scale bias (y) was 1.26 and 1.05 without and
The SAD set designed in this study at the leaflet level has six with SADs, respectively and the correction factor (Cb) was 0.88 and
leaflet images, with powdery mildew severities from 1 to 60% 0.98 without and with SADs, respectively (Table 1). When SADs
(Fig. 1), based on the range of severities observed occurring natu- were not employed, the rc was 0.73 and when the SADs were used,
rally in tomato fields (0.5e65%). the rc significantly improved to 0.88. Without use of the SADs, the
Visual assessments of severity of powdery mildew on leaflets precision (r) was 0.82, but when the SADs were employed, pre-
were closer to the actual values when the diagrams were cision was 0.90 (Table 1). The more precise evaluation of powdery
employed, compared to unaided visual assessments (Fig. 3 A, C vs. mildew severity using the SADs was also confirmed by reduction
30 D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34

in absolute error, with most falling from c. ±40% (Fig. 4 A, C) to c. (Fig. 2), based on the observed range of severities observed in to-
±20% (Fig. 4 B, D). mato fields, which ranged from 0.5 to 62%.
Assessments were closer to the actual values when SADs
3.1.2. Rater experience on accuracy and precision with leaflet SADs were employed (Fig. 3 E, F, G, H). The accuracy, the precision
As expected, without the aid of the diagrams, experienced raters and the agreement of perceived and actual powdery mildew
estimated powdery mildew severity significantly better than assessments on leaves were improved. There was a slight
inexperienced raters for almost all statistical parameters (Cb, y, u improvement in the scale bias (y ¼ 1.06 and 0.99 without
and rc, Table 2), except for precision, which did not differ between and with SADs, respectively), but all other parameters (r, Cb, u,
groups (r ¼ 0.81 and 0.83). On the other hand, with aid of the SADs, and rc) were significantly improved when the diagrams were
almost all statistical parameters (except u) were not significantly employed. The location bias (u) was 0.12 and 0.05 without and
different between groups of inexperienced and experienced raters with SADs, respectively. The correction factor (Cb) was 0.92 and
(Table 2). Thus, the use of the SADs improved accuracy and preci- 0.98 without and with SADs, respectively. When SADs were not
sion of the estimates of powdery mildew severity by inexperienced employed, the rc was 0.81 and when the SADs were used, the rc
raters to almost the same level as experienced raters. increased to 0.91. Without use of the SADs, the precision (r)
Use of the diagrams was associated to a tendency to over- was 0.88, but when the SADs were employed, precision
estimate disease severity by the inexperienced raters. This was was 0.93 (Table 1). The more precise evaluation of powdery
made evident by the fact that 85.7% of the inexperienced raters mildew severity using the SADs was also confirmed by
showed positive location parameter (u) values (data not shown). reduction in absolute error, with most falling within ±20% (Fig. 4
Moreover, this trend was further confirmed by the fact that positive E, F, G, H).
absolute errors were detected for most estimates (Fig. 4B).
Contrarily, experienced raters using SADs were more disposed to
underestimate disease severity (75% of the experienced raters 3.2.2. Rater experience on accuracy and precision with leaf SADs
showed negatives u values, data not shown). This trend was also As anticipated, without the visual aid of the SADs, the group of
confirmed by the fact that negative absolute errors were detected experienced raters had significantly better agreement between
for most estimates (Fig. 4D). estimated and actual severity values than the group of inexperi-
enced raters, as determined by the Lin's concordance correlation
3.1.3. Inter-rater reliability coefficient (rc, Table 2). Conversely, with the aid of the diagrams, no
Based on the equivalence test, the statistical parameters of the differences were detected in the LCCC (rc) between the two groups,
intra-class correlation coefficient (r) and the inter-rater coefficient and only the scale and location parameters (y and u) remained
of determination (R2) were significantly enhanced when raters significantly different between the group of experienced and
used the SADs to estimate powdery mildew severity on leaflets. inexperienced raters (Table 2). Therefore, adoption of the diagrams
With aid of the diagrams, r advanced from r ¼ 0.689 to r ¼ 0.868 as visual aid to estimate tomato powdery mildew severity on leaves
(Table 3). Similarly, the R2 mean of the pairwise comparisons improved the performance of inexperienced raters to almost the
increased from 0.83 to 0.89 (Table 3), indicating that use of the same level as experienced raters.
diagrams improved inter-rater reliability. In a similar manner to what was observed with the leaflet SADs,
there was a tendency to overestimate disease severity on leaves by
3.2. Standard area diagram set to assess the severity of powdery inexperienced raters and 62.5% of the inexperienced raters showed
mildew on tomato leaves positive location parameter (u) values (data not shown). This is
confirmed by the fact that positive absolute errors were detected
3.2.1. Effect of leaf SADs on accuracy and precision for most estimates (Fig. 4F). On the other hand, experienced raters
The SAD set designed at the whole leaf level has six leaf images, using SADs tended to underestimate disease severity, and 62.5%
each with a distinct severity of powdery mildew from 1 to 60% had negative u values (data not shown). This trend was further

Table 1
Effect of using a standard area diagram set (SADs) as an assessment aid on the bias, accuracy, precision and agreement of assessments of severity of powdery mildew on 42
tomato leaflets and leaves as estimated by16 raters.

Diagram Parameters Meansa Differenceb between means 95% CIsc of the difference

W/o SAD With SAD

Leaflet diagram set Scale (y)d 1.26 (0.25) 1.05 (0.10) 0.21 (0.06) ¡0.320 to ¡0.109
Location (u)e 0.14 (0.47) 0.03 (0.19) 0.12 (0.10) 0.312 to 0.062
Coefficient of bias (Cb)f 0.88 (0.12) 0.98 (0.03) 0.10 (0.03) 0.049e0.155
Correlation coefficient (r)g 0.82 (0.05) 0.90 (0.04) 0.08 (0.01) 0.055e0.098
LCCC (rc)h 0.73 (0.11) 0.88 (0.05) 0.16 (0.02) 0.117e0.199
Leaf diagram set Scale (y) 1.06 (0.27) 0.99 (0.08) 0.07 (0.06) 0.195 to 0.022
Location (u) 0.12 (0.39) 0.05 (0.19) 0.18 (0.07) ¡0.318 to ¡0.042
Coefficient of bias (Cb) 0.92 (0.10) 0.98 (0.02) 0.06 (0.02) 0.019e0.113
Correlation coefficient (r) 0.88 (0.04) 0.93 (0.02) 0.05 (0.01) 0.033e0.064
LCCC (rc) 0.81 (0.09) 0.91 (0.03) 0.10 (0.02) 0.070e0.144
a
The values for standard deviation are in parentheses.
b
Mean of the difference between each rating. The values for standard errors are in parentheses (bootstrap calculated values).
c
2000 bootstrap samples were used to obtain the confidence intervals (CIs). If the CIs embrace zero, the difference was not significant (a ¼ 0.05). Bold numbers represent
significance of the difference.
d
Scale bias or slope shift (y, 1 ¼ no bias relative to the concordance line).
e
Location bias or height shift (u, 0 ¼ no bias relative to the concordance line).

f
The correction factor (Cb) measures how far the best-fit line deviates from 45 and is a way to measure accuracy.
g
The precision is measured by the correlation coefficient (r).
h
Lin's concordance correlation coefficient (LCCC) combines both measures of precision (r) and accuracy (Cb) to measure agreement with the true value.
D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34 31

confirmed by the fact that negative absolute errors were detected 4. Discussion
for most estimates by experienced raters (Fig. 4H).
Standard area diagrams are important for quantification of
3.2.3. Inter-rater reliability disease severity to be used in epidemiological studies, screening
Both the intra-class correlation coefficient (r) and the coefficient for partial resistance and assessment of disease management
of determination (R2) were significantly improved when the raters schemes. Results demonstrated that SADs improve accuracy,
used the SADs to estimate powdery mildew severity on the leaves. precision and reliability of the tomato powdery mildew data
The r value was greater when the SADs were used (r ¼ 0.717 un- collection and will help minimize the risk of erroneous decisions
aided vs. r ¼ 0.875 with SADs). Without the aid of the SADs, the R2 or Type II errors in statistical analysis (Bock et al., 2010; Nutter
mean of the pairwise comparisons was 0.85, but when the dia- et al., 1993). SADs have been demonstrated to improve the accu-
grams were employed, the R2 mean increased to 0.90. All param- racy and reliability of assessment of diseases in different crops
eters thus indicate that use of the leaf SADs provided greater inter- including corn white spot (Capucho et al., 2010), potato early
rater reliability (Table 3). blight (Duarte et al., 2013), wheat blast (Rios et al., 2013), pecan

No SAD SAD used


60 60
A
A B
B
40 40

20 20

0 0

-20 -20

-40 -40

-60 -60

60 60
CC DD
40 40

20 20

0 0

-20 -20
Absolute error

-40 -40

-60 -60

60 60
AE B
F
40 40

20 20

0 0

-20 -20

-40 -40

-60 -60

60 60
C
G DH
40 40

20 20

0 0

-20 -20

-40 -40

-60 -60
0 10 20 30 40 50 60 0 10 20 30 40 50 60

Actual severity (%)


Fig. 4. Absolute error (estimated severity minus actual severity) of estimates of severity of powdery mildew (Leveillula taurica) on tomato (Solanum lycopersicum L.) without and
with the use of a set of standard area diagrams (SADs) for 42 diseased leaflets (A, B, C and D) and 42 diseased leaves (E, F, G and H). A, B, E and F represent 8 raters with experience
and C, D, G and H represent 8 raters without experience in disease severity assessment. Low absolute errors indicate that the estimated severity was similar to the actual severity.
32 D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34

Table 2
Effect of rater experience without and with the use of standard area diagram sets (SADs) as an assessment aid on the bias, accuracy, precision and agreement of assessments of
severity of powdery mildew on 42 tomato leaflets and leaves as estimated by16 raters.

Diagram SADs Parameters Means of experiencea Differenceb between means 95% CIsc of the difference

Inexperienced (n ¼ 8) Experienced (n ¼ 8)

Leaflet diagram set W/o SAD Scale (y)


d
1.41 (0.24) 1.10 (0.16) 0.31 (0.10) ¡0.484 to ¡0.095
Location (u)e 0.45 (0.46) 0.17 (0.18) 0.57 (0.22) ¡0.943 to ¡0.090
Coefficient of bias (Cb)f 0.81 (0.13) 0.96 (0.02) 0.15 (0.04) 0.071¡0.238
Correlation coefficient (r)g 0.81 (0.06) 0.83 (0.11) 0.01 (0.03) 0.035 to 0.062
LCCC (rc)h 0.66 (0.11) 0.80 (0.06) 0.14 (0.04) 0.065¡0.205
With SAD Scale (y) 1.08 (0.11) 1.02 (0.08) 0.05 (0.04) 0.131 to 0.019
Location (u) 0.13 (0.18) 0.08 (0.13) 0.21 (0.07) ¡0.341 to 0.093
Coefficient of bias (Cb) 0.97 (0.04) 0.99 (0.01) 0.01 (0.01) 0.007 to 0.442
Correlation coefficient (r) 0.89 (0.04) 0.92 (0.03) 0.03 (0.02) 0.001 to 0.062
LCCC (rc) 0.86 (0.05) 0.90 (0.03) 0.03 (0.02) 0.011 to 0.068
Leaf diagram set W/o SAD Scale (y) 1.13 (0.37) 0.99 (0.08) 0.14 (0.12) 0.396 to 0.070
Location (u) 0.30 (0.43) 0.06 (0.27) 0.37 (0.17) 0.661 to 0.008
Coefficient of bias (Cb) 0.87 (0.12) 0.97 (0.03) 0.09 (0.05) 0.004¡0.192
Correlation coefficient (r) 0.89 (0.04) 0.88 (0.04) 0.01 (0.02) 0.039 to 0.029
LCCC (rc) 0.77 (0.11) 0.85 (0.03) 0.08 (0.04) 0.014¡0.156
With SAD Scale (y) 1.03 (0.10) 0.95 (0.04) 0.07 (0.03) ¡0.129 to ¡0.008
Location (u) 0.05 (0.17) 0.16 (0.16) 0.20 (0.08) ¡0.361 to 0.065
Coefficient of bias (Cb) 0.98 (0.02) 0.98 (0.02) 0.01 (0.01) 0.022 to 0.012
Correlation coefficient (r) 0.92 (0.02) 0.94 (0.02) 0.02 (0.01) 0.001 to 0.034
LCCC (rc) 0.91 (0.04) 0.92 (0.02) 0.011 (0.01) 0.015 to 0.037
a
The values for standard deviation are in parentheses.
b
Mean of the difference between each rating. The values for standard errors are in parentheses (bootstrap calculated values).
c
2000 bootstrap samples were used to obtain the confidence intervals (CIs). If the CIs embrace zero, the difference was not significant (a ¼ 0.05). Bold numbers represent
significance of the difference.
d
Scale bias or slope shift (y, 1 ¼ no bias relative to the concordance line).
e
Location bias or height shift (u, 0 ¼ no bias relative to the concordance line).

f
The correction factor (Cb) measures how far the best-fit line deviates from 45 and is a way to measure accuracy.
g
The precision is measured by the correlation coefficient (r).
h
Lin's concordance correlation coefficient (LCCC) combines both measures of precision (r) and accuracy (Cb) to measure agreement with the true value.

scab (Yadav et al., 2013) and pepper anthracnose (Pedroso et al., mildew observed in the field. Too few diagrams in one set can
2011). compromise accuracy and precision, while an excessive number of
Powdery mildew is a disease that leads to heavy damage in diagrams can be time consuming and affect the efficiency of as-
tomato production. Accurate assessments of actual disease se- sessments (Madden et al., 2007). Correa et al. (2009) have shown
verities are paramount for implementing powdery mildew yield that six diagrams are sufficient for tomato leaf blight and more than
loss estimates, supporting breeding programs, and testing the ten diagrams are excessive and result in loss of time.
efficacy of various control methods, among other studies. How-
ever, the variation in symptoms, along with the composite anat-
Table 3
omy of the tomato leaf, may lead to reduced accuracy and
Inter-rater reliability of assessments by 16 raters of powdery mildew on 42 leaflets
precision in the estimates. These were the main problems high- and leaves of tomato both without and with the use of a standard area diagram set
lighted by raters who helped to evaluate the diagrammatic scales (SADs) assessment aid. Inter-rater reliability is measured by the intra-class corre-
that we developed. Furthermore, the diagrammatic scales lation coefficient (r) and coefficient of determination (R2).
currently available for evaluating other leaf diseases in tomato do Diagram Statistics Without aid With aid of SADs
not realistically represent symptoms of powdery mildew on to- of SADs
mato leaves and/or leaflets, which justified the need to develop Leaflet Intra-class 0.689 F, P > Fa: 0.868 F, P > F:
specific scales for this disease. diagram correlation L ¼ 75 L ¼ 132
The two scales for assessing disease in leaflets and leaves are set coefficient (r) (P < 0.0001); (P < 0.0001);
designed for use in different approaches to study tomato powdery R ¼ 46 R ¼ 12
(P < 0.0001) (P < 0.0001)
mildew. Currently, it is common to find published reports in which
Mean inter-rater 0.83 (0.62e0.94) 0.89 (0.81e0.95)
several tomato diseases, including the L. taurica mildew, are studied coefficient of Mean differencec ¼ 0.065 (0.0002),
at leaflet level, as in host resistance and plantepathogen in- determination (R2)b 95% CIs 0.047 e 0.086
teractions in controlled conditions (e.g. Guzman-Plazola et al., Leaf Intra-class correlation 0.717 F, 0.875 F,
2003). In these types of experiments the leaflet diagrams would diagram coefficient (r) P > Fa: L ¼ 75 P > F: L ¼ 151
set (P < 0.0001); (P < 0.0001);
be useful, because often only leaflets are inoculated or the assay is R ¼ 36 R ¼ 16 (P < 0.0001)
conducted in detached leaflets. On the other hand, the use of whole (P < 0.0001)
leaves for severity assessment is also common, as in comparison of Mean inter-rater 0.85 (0.72e0.94) 0.90 (0.83e0.98)
fungicide effects, host partial resistance or other management coefficient Mean differencec ¼ 0.051 (0.0001),
n-Plazola et al., 2011). The proposed leaf di- of determination (R2)b 95% CIs 0.037 e 0.066
strategies (e.g. Guzma
a
agrams will be useful when raters wish to collect severity data on F-value for L ¼ leaf, R ¼ rater. F e value in parentheses.
b
whole leaves or at the plant level. Mean coefficients of determination estimated from pairwise comparisons of
assessments by all visual raters.
The two SADs proposed in this study have six images with a c
Mean of the difference between each rating, with standard errors in parentheses
range of severity distributed from 1 to 60%. The number of diagrams (bootstrap calculated value), confidence intervals (CIs) were based on 2000 boot-
is quite sufficient to represent the range of intensities of powdery strap samples. If the CIs embrace zero, the difference is not significant (a ¼ 0.05).
D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34 33

Without aid of SADs, there was a significant effect of rater Acknowledgments


experience on rc (which combines the measures of both precision
and accuracy), where experienced raters were more accurate and We gratefully acknowledge Dr. Emerson Del Ponte (UFV) for the
precise. However, there were no significant differences on rc when review of the statistical analysis and valuable discussions. The first
the experienced and inexperienced groups rated severity using the author was partially supported by a doctorate sholarship of the
SADs. Therefore, performance of inexperienced raters improved up Brazilian National Research Council (CNPq). AC Cafe -Filho and WA
to a point where no significant differences were detected between Marouelli are recipients of CNPq Research Fellowships, grant
the two groups. These effects have been observed in previous numbers 305792/2013-0 and 302954/2010-4.
studies (e.g., Gonzalez-Domínguez et al., 2014; Pedroso et al., 2011;
Yadav et al., 2013). Overall, results showed that both of the SAD sets
developed and validated in this study improved the accuracy, References
precision and reliability of disease severity estimates, and are good
Bardsley, S.J., Ngugi, H.K., 2013. Reliability and accuracy of visual methods used to
tools for quantifying powdery mildew severity on leaflets and quantify severity of foliar bacterial spot symptoms on peach and nectarine.
leaves. Plant Pathol. 62, 460e474.
Even as absolute errors were consistently reduced, a tendency Bock, C.H., Poole, G., Parker, P.E., Gottwald, T.R., 2010. Plant disease severity esti-
mated visually, by digital photography and image analysis, and by hyperspectral
to overestimate disease severity by inexperienced raters and a imaging. Crit. Rev. Plant Sci. 29, 59e107.
tendency to underestimate disease severity by inexperienced Capucho, A.S., Zambolim, L., Duarte, H.S.S., Parreira, D.F., Ferreira, P.A., Lanza, F.E
raters using SADs was observed in this study. According to Nutter Costa, R.V., Casela, C.R., Cota, L.V., 2010. Influence of leaf position that corre-
spond to whole plant severity and diagrammatic scale for white spot of corn.
and Schultz (1995), solutions to correct or minimize inaccurate Crop Prot. 29, 1015e1020.
disease severity assessments vary according to the magnitude of Cabral, R.N., Marouelli, W.A., Lage, D.A.C., Cafe -Filho, A.C., 2013. Septoria leaf spot in
the error and the amount of training received by the rater. organic tomatoes under diverse irrigation systems and water management
strategies. Hort. Bras. 31, 392e400.
Computer-based training of raters is one valuable approach to Correa, F.M., Bueno Filho, J.S.S., Carmo, M.G.F., 2009. Comparison of three dia-
reduce errors and increase the accuracy of tomato powdery grammatic keys for the quantification of late blight in tomato leaves. Plant
mildew severity estimates, as suggested by several authors Pathol. 58, 1128e1133.
Correll, J.C., Gordon, T.R., Elliott, V.J., 1988. Powdery mildew of tomato: the effect of
(Bardsley and Ngugi, 2013; Nutter and Schultz, 1995; Nutter and planting date and triadimefon on disease onset, progress, incidence and
Worawitlikit, 1989). severity. Phytopathology 78, 512e519.
One of the most commonly used approaches for validating Diver, S., Kuepper, G., Born, H., 1999. Organic Tomato Production. Horticultural
Production Guide. ATTRA, USDA, Fayetteville.
disease assessment methods and estimating the accuracy and
Drinkwater, L.E., Letourneau, D.K., Workneh, F., Van Bruggen, A.H.C., Shennan, C.,
precision of estimates is linear regression (Bock et al., 2010; 1995. Fundamental differences between conventional and organic tomato
Capucho et al., 2010; Nutter et al., 1993; Nutter and Schultz, agroecosystems in California. Ecol. Appl. 5, 1098e1112.
1995; Nutter and Esker, 2006). We have indeed resorted to Duarte, H.S.S., Zambolim, L., Capucho, A.S., Junior, A.F.N., Rosado, A.W.C.,
Cardoso, C.R., Paul, P.A., Mizubuti, E.S.G., 2013. Development and validation of a
linear regression to demonstrate the relationship between actual set of standard area diagrams to estimate severity of potato early blight. Eur. J.
and estimated severity in this paper (Fig. 3). However, there has Plant Pathol. 137, 249e257.
Gonza lez-Domínguez, E., Martins, R.B., Del Ponte, E.M., Michereff, S.J., García-
been an argument that erroneous conclusions may be drawn from
Jimenez, J., Armengol, J., 2014. Development and validation of a standard area
linear regression analysis under certain circumstances (Lin, 1989; diagram set to aid assessment of severity of loquat scab on fruit. Eur. J. Plant
Madden et al., 2007). Linear regression analysis may not detect Pathol. 139, 1e10.
departure from intercept 0 and slope 1 if the data are very scat- Guzman-Plazola, R.A., Davis, R.M., Marois, J.J., 2003. Effects of relative humidity and
high temperature on spore germination and development of tomato powdery
tered. The less precise the data, the less likely the null hypothesis mildew (Leveillula taurica). Crop Prot. 22, 1157e1168.
(b0 ¼ 0 and b1 ¼ 1) will be rejected, and conversely, if the system is Guzm an-Plazola, R.A., Fajardo-Franco, M.L., Coffey, M.D., 2011. Control of tomato pow-
highly reproducible, the hypothesis could be rejected due to very dery mildew (Leveillula taurica) in the Comarca Lagunera, Coahuila State, Mexico,
supported by the spray forecast model tomato PM. Crop Prot. 30, 1006e1014.
small error (Lin, 1989). Thus, we have also employed Lin's
Jones, J.B., Jones, J.P., Stall, R.E., Zitter, T.A., 1991. Compendium of Tomato Diseases.
concordance correlation because it combines measures of accu- APS Press, St Paul.
racy and precision and avoids these problems associated with Jones, W.B., Thomson, S.V., 1987. Source of inoculum, yield and quality of tomato as
affected by Leveillula taurica. Plant Dis. 71, 266e268.
linear regression. In the present study, accuracy, precision and
Kasselaki, A.M., Shaw, M.W., Malathrakis, N.E., Haralambous, J., 2006. Control of
reliability of SADs to estimate powdery mildew severity were Leveillula taurica in tomato by Acremonium alternatum is by induction of
determined using an equivalence test, recommended for agree- resistance not hyperparasitism. Eur. J. Plant Pathol. 115, 263e267.
ment studies (Yi et al., 2008). The equivalence test has been used Kiss, L., Cook, R.T.A., Saenz, G.S., Cunnington, J.H., Takamatsu, S., Pascoe, I.,
Bardin, M., Nicot, P.C., Sato, Y., Rossman, A.Y., 2001. Identification of two pow-
in the analysis of other disease diagrams to statistically estimate dery mildew, Oidium neolycopersici sp. nov. and Oidium lycopersici, infecting
agreement and reliability (Bardsley and Ngugi, 2013; Yadav et al., tomato in different parts of the world. Mycol. Res. 105, 684e697.
2013). Konstantinidou-Doltsinis, S., Markellou, E., Kasselaki, A.M., Fanouraki, M.N.,
Koumaki, C.M., Schmitt, A., Liopa-Tsakalidis, A., Malathrakis, N.E., 2006. Efficacy
While SADs are important for the quantitative appraisal of the of Milsana, a formulated plant extract from Reynoutria sachalinensis, against
efficacy of disease management practices, they are not designed for powdery mildew of tomato (Leveillula taurica). BioControl 51, 375e392.
real-time management decisions, such as time of fungicide appli- Lin, L.I., 1989. Aconcordance correlation coefficient to evaluate reproducibility.
Biometrics 45, 255e268.
cations. Because SADs are based on symptoms, and not on pathogen Madden, L.V., Hughes, G., Van den Bosch, F., 2007. The Study of Plant Disease Epi-
signs, they do not adequately measure latent periods. In fact, in demics. APS Press, St. Paul.
some specific environmental conditions, sporulation ahead of Nita, M., Ellis, M.A., Madden, L.V., 2003. Reliability and accuracy of visual estimation
of Phomopsis leaf blight of strawberry. Phytopathology 93, 995e1005.
chlorosis may be detected in some tomato genotypes and the use of Nutter Jr., F.W., Gleason, M.L., Jenco, J.H., Christians, N.C., 1993. Assessing the ac-
the diagrams alone will not account for this important epidemio- curacy, intra-rater repeatability, and inter-rater reliability of disease assessment
logical variable. systems. Phytopathology 83, 806e812.
Nutter Jr., F.W., Schultz, P.M., 1995. Improving the accuracy and precision of disease
In summary, the standard area diagrams proposed in this study
assessments: selection of methods and use of computer-aided training pro-
consistently and significantly improved the accuracy, precision and grams. Can. J. Plant Pathol. 17, 174e184.
reliability of the estimates of L. taurica powdery mildew severity on Nutter, F.W., Esker, P.D., 2006. The role of psychophysics in phytopathology. Eur. J.
tomato leaflets and leaves. It is thus hoped that these SADs will Plant Pathol. 114, 199e213.
Nutter, F.W., Worawitlikit, O., 1989. Disease.Pro: a computer program for evaluating
prove to be useful tools for quantitative studies in epidemiology and improving a person ability to assess disease proportion. Phytopathology 79,
and disease management in this important pathosystem. 111e135.
34 D.A. Costa Lage et al. / Crop Protection 67 (2015) 26e34

Nutter, F.W., Teng, P.S., Shokes, F.M., 1991. Disease assessment term and concepts. Rotem, J., Palti, J., 1969. Irrigation and plant diseases. Annu. Rev. Phytopathol. 7,
Plant Dis. 75, 1187e1188. 267e288.
Palti, J., 1988. The Leveillula mildews. Bot. Rev. 54, 423e535. Shoukri, M.M., Pause, C.A., 1999. Statistical Methods for Health Science, second ed.
Pedroso, C., Lage, D.A.C., Henz, G.P., Cafe-Filho, A.C., 2011. Development and vali- CRC Press, Boca Raton, FL.
dation of a diagrammatic scale for estimation of anthracnose on sweet pepper Sutton, T.B., Jones, A.L., 1979. Analysis of factors affecting dispersal of Podosphaera
fruits for epidemiological studies. J. Plant Pathol. 93, 219e225. leucotricha conidia. Phytopathology 69, 380e383.
Reis, A., Boiteux, L.S., Paz-Lima, M.L., Silva, P.P., Lopes, C.A., 2005. New hosts of Vale, F.X.R., Fernandes-Filho, E.I., Liberato, J.R., 2003. A software for plant disease
Oidiopsis haplophylli in the Solanaceae family in Brazil. Fitopatol. Bras. 30, severity assessment. In: Annals of XVIII International Congress of Plant Pa-
195e198. thology. New Zealand, Christchurch, p. 105 (Abstract).
Reis, A., Boiteux, L.S., Paz-Lima, M.L., 2007. Powdery mildew of ornamental species Yadav, N.V., Vos, S.M., Bock, C.H., Wood, B.W., 2013. Development and validation of
caused by Oidiopsis haplophylli in Brazil. Summa Phytopathol. 33, 405e408. standard area diagrams to aid assessment of pecan scab symptoms on fruit.
Rios, J.A., Debona, D., Duarte, H.S.S., Rodrigues, F.A., 2013. Development and vali- Plant Pathol. 62, 325e335.
dation of a standard area diagram set to assess blast severity on wheat leaves. Yi, Q., Wang, P.P., He, Y., 2008. Reliability analysis for continuous measurements:
Eur. J. Plant Pathol. 136, 603e611. equivalence test for agreement. Stat. Med. 27, 2816e2825.

You might also like