Theor Appl Climatol

DOI 10.1007/s00704-014-1182-3

ORIGINAL PAPER

Modeling temporal and spatial variability of leaf wetness

duration in Brazil
Clayton Alcarde Alvares & Eduardo Mor de Mattos &
Paulo Cesar Sentelhas & Aline Cristina Miranda &
Jos Luiz Stape

Received: 25 March 2014 / Accepted: 22 May 2014

# Springer-Verlag Wien 2014

Abstract Leaf wetness duration (LWD) is recognized as a

very important conditioner of crops and forests diseases, but
clearly, there is a considerable gap in literature on temporal
models for prediction of LWD in broad regions from standard
meteorological data. The objective of this study was to develop
monthly LWD models based on the relationship between hours
of relative humidity (RH)90 % and average RH for Brazil and
based on these models to characterize the temporal and spatial
LWD variability across the country. Two different relative
humidity databases, being one in an hourly basis (RHh) and
C. A. Alvares : A. C. Miranda
Forestry Science and Research Institute (IPEF), 11 Pdua Dias
Avenue, PO box 530, Piracicaba, SP 13400-970, Brazil
A. C. Miranda
e-mail: aline.miranda@ipef.br
C. A. Alvares (*) : J. L. Stape
Forest Productivity Cooperative (FPC), 3108 Jordan Hall, Raleigh,
NC 27695-8008, USA
e-mail: clayton@ipef.br
C. A. Alvares
e-mail: caalvares@yahoo.com.br
J. L. Stape
e-mail: jlstape@ncsu.edu
E. M. de Mattos : P. C. Sentelhas : J. L. Stape
University of So PauloAgricultural College Luiz de Queiroz
(USP/ESALQ), Piracicaba, SP, Brazil11 Pdua Dias Avenue,
13418-900
E. M. de Mattos
e-mail: more.mattos@yahoo.com.br
P. C. Sentelhas
e-mail: pcsentel.esalq@usp.br
J. L. Stape
North Carolina State University (NCSU), 3108 Jordan Hall, Raleigh,
NC 27695-8008, USA

another in a monthly basis (RHm), were used. To elaborate the

LWD models, 58 automatic weather stations distributed across
the country were selected. Monthly LWD maps for the entire
country were prepared, and for that, the RHm from the 358
conventional weather stations were interpolated using
geostatistical techniques. RHm and LWD showed sigmoidal
relationship with determination coefficient above 0.84 and were
highly significant (p<0.0001). In relation to the validation of
the LWD monthly models, a very good performance for all
months was obtained, with very high precision with r between
0.92 and 0.96. Regarding the errors, mean error showed a slight
tendency of overestimation during February (0.29 h day1),
May (0.31 h day1), July (0.14 h day 1), and August
(0.34 h day1), whereas for the other months, the tendency
was of underestimation like January (0.27 h day1) and
March (0.25 h day1). Even as a first approach, the results
presented here represent a great advance in the climatology of
LWD for Brazil and will allow the development of studies
related to crop and forest diseases control plans.

1 Introduction
Leaf wetness is basically defined as the presence of free water
over plant tissues, and its duration is commonly named leaf
wetness duration (LWD). LWD is recognized as a very important conditioner of crop (De Weille 1965; Spotts 1977;
Huber and Gillespie 1992; Gleason et al. 1994; Kim et al.
2002; Sentelhas et al. 2006, 2008; Mersha et al. 2014) and
forest (Gadgil 1977; Park 1988; Hamelin et al. 1992; Rizzo
and Garbelotto 2003; Pinon et al. 2006; Silva et al. 2013)
diseases since it is fundamental for the fungal and bacterial
infection process, as well as for the sporulation of some of
them, so the risk of outbreaks of many plant diseases is
directly proportional to LWD (Huber and Gillespie 1992).

C.A. Alvares et al.

Furthermore, high humidity in the air, per se, may be an

abiotic agent of plant disease (Alfenas et al. 2009).
For this reason, LWD has been used together with air
temperature in many disease warning systems (DWSs), which
are decision support tools that help growers or other crop
managers to assess the risk of diseases outbreaks and advise
them when chemical or biological control is required, avoiding
economic losses (Gleason et al. 2008; Carisse et al. 2012; Rossi
et al. 2014). Therefore, DWSs are important components of the
Integrated Pest Management (IPM) (Strand 2000).
Leaf wetness is mainly caused by rain, fog, irrigation,
dewfall from the atmosphere, or distillation from the soil
(Monteith 1990). Dew is often the main source for leaf wetness and is formed as a result of radiative loss of heat from
plant surfaces during clear sky on humid and calm nights.
Even considering that dew formation can be easily understood
from the physical point of view, LWD is not an easy variable
to measure since it depends on the interactions between canopy characteristics and the atmospheric conditions (Huber and
Gillespie 1992; Dalla Marta et al. 2005; Sentelhas et al. 2005,
2006). These complexities are not taken into account neither
when LWD is measured by electronic sensors in standard
weather stations nor when sensors are inside the crop field,
which lead to errors depending on sensor position in the
canopy, angle of deployment, and interaction with surrounding leaves (Hirst 1957; Lomas and Shashoua 1970; Lau et al.
2000; Dalla Marta et al. 2004; Sentelhas et al. 2004a).
Since the 1980s, several studies have been carried out to
overcome the problem of LWD measurement through the
development of simulation models using different approaches.
According to these studies, LWD models can be divided into
two broad categories: empiricalbased on the relationship
between LWD and one or more weather variables (Gleason
et al. 1994, 2008; Rao et al. 1998; Kim et al. 2002, 2010;
Hamada et al. 2008; Sentelhas et al. 2008);and physical
based on the physical principles of dew formation and evaporation (Pedro Jr and Gillespie 1982a; Bass et al. 1991; Luo
and Goudriaan 2000; Magarey et al. 2006; Sentelhas et al.
2006; Leca et al. 2011). Further attempts have been made
using mixed approaches such as neural networks, a distributed
information processing system (Francl and Panigrahi 1997;
Chtioui et al. 1999; Dalla Marta et al. 2005) and fuzzy logic
system in which physical principles can be incorporated into
empirical models (Kim et al. 2004, 2005).
LWD models based on physical principles can be very
accurate and have been applied in different geographical and
climatic areas (Pedro Jr and Gillespie 1982b; Magarey et al.
2006; Sentelhas et al. 2006). However, for some conditions,
this kind of model is too complex, requiring inputs that are not
always available (Gleason et al. 2008). In contrast, empirical
models can simulate LWD by using simple relationships of
this variable with parameters measured at standard weather
stations. Their success depends on the accuracy of the weather

data, and they are usually site-specific (Huber and Gillespie

1992; Gleason et al. 1994; Kim et al. 2005; Sentelhas et al.
2008).
Among the LWD empirical models, the number of hours
with relative humidity (RH) above a specific threshold is the
most common and easy to apply (Bondalapati et al. 2012;
Durigon and van Lier 2013). Rao et al. (1998), estimating
LWD for corn ears in Canada, and Sentelhas et al. (2004b),
estimating LWD for a cotton crop in Brazil, showed that
wetness duration estimates from a simple threshold model
based on RH were as accurate as estimates from more complex physical models. Kruit et al. (2004), comparing two
LWD models based on physical models, proposed by Garratt
and Segal (1988) and Pedro Jr and Gillespie (1982a), and two
based on adjusted RH threshold (87 %) found that the empirical models had better performance, increasing the proportion
of correct estimates to almost 90 % of hours assessed in the
Netherlands. Sentelhas et al. (2008), evaluating three empirical models based on RH, found that number of hours with
RH90 % as surrogate for LWD resulted in estimates that
were as accurate as those derived from physical models.
According to the results presented above, it is clear that the
performance of empirical models can vary from place to place,
but it is also evident that this is the most suitable, simple, and
useful way to estimate LWD for evaluating regional potential
for disease outbreaks. Furthermore, there is still a lack of
spatial models to estimate LWD (Alves et al. 2011) and
estimating such weather variable for broad regions from standard meteorological data remains a challenge (Booth and
Jovanovic 2012). Moreover, the interface between LWD
models and geographic information systems (GIS), including
geostatistical and multivariate procedures, is a powerful way
to identify the patterns of spatial and temporal distribution of
this variable (Hartkamp et al. 1999). As for large scales, e.g.,
at a country level, the availability of weather data for estimating LWD is very restricted; the use of hours of RH 90 %
appears as the only alternative to assess the temporal and
spatial LWD variability. Considering that, the objective of this
study was to develop monthly LWD models based on the
relationship between hours of RH 90 % and average RH for
Brazil and based on these models to characterize the temporal
and spatial LWD variability across the country in order to
support crop and forest diseases control strategic plans.

2 Material and methods

The present study used two different relative humidity databases, one in an hourly basis (RHh) and another in a monthly
basis (RHm), both of them provided by the Brazilian National
Institute of Meteorology (INMET). This governmental agency
maintains, collects, and distributes at the national level meteorological data from a wide network of mechanical and

Modeling temporal and spatial variability of leaf wetness

automatic surface weather stations. Most of the mechanical

weather stations (MWS) have operated since 1961, and a total
of 291 of them were assessed for historical RHm data (Brazil
1992). In the MWS, the relative humidity is read at the
following local time: 9 a.m., 3 p.m., and 9 p.m., and the daily
average was calculated by applying Eq. 1 (Brazil 1992):


RH9am RH3pm 2RH9pm
RHD
1
4

where RHD is the daily relative humidity, RH9am is the relative

humidity measurement at 9 a.m., RH3pm is the relative humidity measurement at 3 p.m., and RH9pm is the relative humidity
measurement at 9 p.m.
The automatic weather station (AWS) network, with more
than 500 units, is more recent and has operated since 2000,
with the majority of them starting in 2008. The spatial distribution of AWS is available at www.inmet.gov.br/sonabra/
maps/automaticas.php. Each of the AWS is programmed to
compile all measurements at an hourly frequency.
The data series evaluation was conducted in several steps
that included data compilation, data consistency, exploratory
analysis, non-linear regression analysis, geographical information system applications, geostatistical methods, interpolation by kriging, and geoprocessing techniques, which were
done in a sequence of routines presented in Fig. 1.
After a first evaluation of the data series, 58 AWS distributed throughout Brazil were selected (Fig. 2), and their main
characteristics are presented in Table 1. The oldest AWS is in
Porto Alegre, Rio Grande do Sul state (Southern Brazil),
operating since September 2000, and the newest is in So
Gabriel da Cachoeira, Amazonas state (Northern Brazil), operating since September 2011. The AWSs selected cover the
entire Brazilian territory and represent an altitude range from
sea level to over 1,800 m.
The total number of months from the AWSs available until
30 April 2013 (N), the last date considered in this study, are
presented in Table 1. Within this period (N), all RHh data from
each AWS were compiled. At this step, the coordinated universal time (UTC) to Brazils time was corrected. The RH has
its maximum values during the night, and the purpose of this
study is to estimate the continuous LWD, so the data were
advanced for 12 h in order to have the entire RHh database
with the day starting at 12:01 of day d and ending at 12:00
of the day d+1. Based on this procedure, the maximum
LWD will be relevant to a 24-h period.
The following step was the screening of the RHh data by
verifying the presence of missing hourly data. All days that
included at least an hour of missing data were disregarded.
This procedure was performed in order to match our assumptions to calculate LWD based on the number of hours with
RHh 90 % (Sentelhas et al. 2008) since each missing hour in

Fig. 1 Flowchart of the procedures for leaf wetness duration (LWD)

modeling based on relative humidity (RH) data for Brazil, considering
automatic and mechanical weather stations (WS)

Fig. 2 Location of the automatic weather stations used for leaf wetness
duration (LWD) modeling with hourly relative humidity data (RHh) in Brazil

C.A. Alvares et al.

Table 1 Spatial distribution (latitude, longitude, and altitude) of the automatic weather stations in Brazil, their data period range, and number of months
(samples) used for modeling
COD
No.

Weather station
location

FU

Lat

Lon

degree

Alt

Period

from

to

months

Apui

AM

7.21

59.89

160

31 July 2008

30 April 2013

58

36

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

Ariquemes
Avar
Barreiras
Bauru
Belmonte
Boa Vista
Caador
Carolina
Chapado do sul
Chapadinha
Chu
Conceio do Araguaia
Cuiab
Curitibanos
Dores do Indai
Feira de Santana
Foz do Iguau
Governador Valadares

RO
SP
BA
SP
BA
RR
SC
MA
MS
MA
RS
PA
MT
SC
MG
BA
PR
MG

9.95
23.07
12.15
22.30
16.09
2.82
26.82
7.34
18.80
3.74
33.74
8.28
15.62
27.29
19.48
12.20
25.60
18.78

62.96
48.88
45.02
49.70
39.22
60.69
50.83
47.46
52.60
43.35
53.37
49.27
56.10
50.60
45.59
38.99
54.48

140
654
470
550
88
94
952
192
818
91
26
180
151
982
722
231
231

21 July 2008
28 September 2006
01 January 2002
30 August 2001
13 July 2009
27 August 2010
15 March 2008
05 December 2007
19 December 2006
18 September 2008
16 March 2008
05 June 2008
08 December 2002
29 February 2008
01 June 2007
26 May 2007
15 February 2008

30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013

58
80
138
142
46
33
62
66
77
56
62
60
127
63
72
72
63

43
45
72
96
38
31
54
37
50
49
47
41
88
56
55
64
49

41.98

263

29 May 2007

30 April 2013

72

64

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37

Guanhes
Gurupi
Ibaiti
Itapeva
Itapira
Ivai
Lbrea
Linhares
Maca
Major Vieira
Manaus
Montes Claros
Natal
Niquelndia
Oiapoque
Ouricuri
Paranaba
Peixe

MG
TO
PR
SP
SP
PR
AM
ES
RJ
SC
AM
MG
RN
GO
AP
PE
MS
TO

18.79
11.75
23.77
23.97
22.42
25.00
7.26
19.36
22.38
26.40
3.13
16.72
5.90
14.47
3.81
7.88
19.41
12.02

42.94
49.05
50.17
48.85
46.81
50.85
64.79
40.07
41.81
50.36
59.95
43.87
35.20
48.45
51.86
40.10
51.11
48.54

860
287
930
707
633
808
64
40
32
808
67
646
49
583
21
464
424
242

02 June 2007
20 January 2007
29 February 2008
25 July 2006
01 January 2008
11 July 2006
26 July 2008
27 October 2006
26 September 2006
10 February 2009
01 January 2002
19 December 2002
24 February 2003
01 June 2001
12 September 2008
16 August 2010
14 November 2006
04 December 2006

30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013

72
76
63
82
65
83
58
79
80
51
138
126
124
145
56
33
79
78

65
67
50
59
53
70
34
65
71
42
76
88
72
77
45
22
24
64

38
39
40

Piracicaba
Porto Alegre
Porto Grande

SP
RS
AP

22.70
30.05
0.70

47.62
51.17

573
47

26 September 2006
22 September 2000

30 April 2013
30 April 2013

80
153

65
110

41
42
43
44
45
46

Porto Murtinho
Porto Seguro
Porto Walter
Pradpolis
Rio Negrinho
Rio Verde

MS
BA
AC
SP
SC
GO

21.71
16.38
8.27
21.33
26.23
17.79

51.43
57.55
39.17
72.75
48.10
49.57
50.96

77
85
85
204
544
869
782

18 September 2008
24 October 2006
11 July 2008
30 October 2009
22 April 2008
24 March 2008
18 July 2007

30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013

56
79
58
43
61
62
72

22
75
43
34
50
52
57

Modeling temporal and spatial variability of leaf wetness

Table 1 (continued)
COD
No.

Weather station
location

FU

Lat

Lon

degree
47
48
49
50
51
52
53
54
55
56
57
58

So Gabriel
So Gabriel da Cachoeira
So Mateus do Sul
So Miguel Arcanjo
Sinop
Taubat
Teresina
Trs Lagoas
Trs Marias
Uberlndia
Urubici
Ventania

RS
AM
PR
SP
MT
SP
PI
MS
MG
MG
SC
PR

30.34
0.13
25.84
23.85
11.97
23.03
5.07
20.78
18.20
18.92
28.00
24.28

54.31
67.06
50.37
48.16
55.55
45.52
42.80
51.70
45.46
48.25
49.50
50.21

Alt

Period

from

to

months

126
90
788
644
371
571
74
313
921
869
1822
1106

11 July 2007
02 September 2011
18 April 2011
15 August 2006
29 October 2006
20 December 2006
13 May 2005
03 September 2001
26 August 2006
21 March 2003
18 June 2007
18 April 2011

30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013
30 April 2013

68
20
25
82
79
77
97
142
81
123
71
25

60
20
19
58
50
40
35
78
70
78
64
17

FU federation unit, Lat latitude, Lon longitude, Alt altitude, N months within the period, n months used for modeling

the database might underestimate LWD in 1 h. Having the

database hourly checked, we programmed a simple counter of
hours of RHh 90 %, and this number was admitted as the
LWD for that day.
After the consistency of the RHh data and estimation of LWD,
the RH and LWD databases were compiled creating a new
database in a monthly level, including the mean and standard
deviation for both variables. Then, a new screening process was
carried out aiming to exclude the months that included 10 or
more days of missing data. This procedure ensures that the
monthly means are representative, avoiding unreliable monthly
data, based on just few days. At the end of this process, the total
months with data for each AWS was obtained (n=3,156 samples,
Table 1), which was used in the modeling phase. The consisted
database was then randomly divided in two data sets. The first
referred to as the fitting set, with 80 % of database (n=2,525),
was used to adjust non-linear regression equations in order to
build the monthly LWD models. Another group, called the
testing set, with 20 % of the remaining data (n=631) was used
to perform the validation of the monthly and annual LWD
models. Fitting and testing sets presented very similar average
and standard deviation, both for monthly RH and LWD (Fig. 3).
The fitting data set was used in the LWD model development
through the relationship between this variable and RHm.
Exploratory analyses were conducted and indicated that such
relationship exhibits a sigmoidal variation, which is represented
by Eq. 2:
LWDi

a
1e

RHi c
b

annual (i=13) relative humidity; and a, b, and c are the

coefficients of the regression model.
The test data set (independent data) was used to validate the
LWD models. For evaluating the performance of the LWD
models, the following errors were determined: mean error
(ME), mean absolute error (MAE), and root-mean-square
error (RMSE). The performance of the LWD models was also

where LWDi is the monthly (i=1, 2,, 12) or annual (i=13)

leaf wetness duration; RHi is the monthly (i=1, 2,, 12) or

Fig. 3 Monthly relative humidity (RH) and leaf wetness duration (LWD)
from 58 automatic weather stations in Brazil. Bars represent data variability (standard deviation) among the weather stations

C.A. Alvares et al.

tested using the Performance index Pi (Alvares et al.

2013a), which is the product of the coefficient of correlation
"r", which defines models precision, and refined agreement
index "dr" (Willmott et al. 2012), which defines the models
accuracy. The criteria for interpreting the Pi are the following: Pi0.75, optimum performance; 0.6Pi<0.75, very good
performance; 0.45Pi<0.6, good performance; 0.3Pi<0.45,
tolerable performance; 0.15Pi<0.3, poor performance; 0Pi
<0.15, bad performance; and Pi<0, very bad performance
(Alvares et al. 2013a).
With the monthly models validated, the next step was to
map the monthly RH and LWD for the country. For that, the
RHm from the 358 MWS (Fig. 4) was interpolated using
geostatistical techniques. Normality hypothesis of RHm database was tested according to the W test at 5 % (Shapiro and
Wilk 1965). The experimental semivariograms were tested by
adjusting them to the theoretical spherical, exponential,
Gaussian, and linear models since they usually covered the
general dispersion of environmental spatial events (Burrough
and Mcdonnell 1998). The best fits were based on the smallest
reduced sums of squares (RSS) and on the highest determination coefficient (R2) (Robertson 2008). Furthermore, the quality of theoretical models was certified by cross-validation and
by the spatial dependence index (SDI). In the cross-validation
analysis, each measured point in the spatial domain is individually removed from the domain and its value estimated,
then the point is replaced and the next point is removed and
estimated, and so on until it covers all the data set (Robertson
2008). SDI was used as recommended by Alvares et al.
(2011), as a measure of the structural variance effect on total

variance (sill) of the sample. SDI comprises the following

classification: weak for SDI<25 %, moderate for SDI between
25 and 75 %, and strong for SDI>75 %. Through structural
parameters obtained from experimental semivariograms, RHm
maps were created with a geographic information system
(GIS), so a punctual ordinary kriging estimator was used for
geostatistic interpolation. Finally, using geoprocessing techniques as map algebra (Theobald 2007; Allen 2011), the
monthly LWD was estimated by the non-linear regression
models into maps using GIS, processing the independent
(RHm) and dependent (LWD) variables as raster layers, i.e.,
monthly maps LWD were calculated pixel by pixel.

Fig. 4 Location of the mechanical weather stations used for leaf wetness
duration (LWD) modeling with monthly relative humidity (RHm) in
Brazil

Fig. 5 General model to estimate mean monthly leaf wetness duration

(LWD) as a function of mean monthly relative humidity (RH) data in
Brazil

3 Results
3.1 LWD modeling
Monthly and general RH and LWD showed sigmoidal relationships, which are presented in Figs 5 and 6. Monthly and
general sigmoidal models produced determination coefficients
above 0.84 and were highly significant (p<0.0001) (Table 2).
The coefficients of the sigmoidal model were significant,
except for the a coefficients in July and August, which were
slightly not significant at the 5 % level of probability (Table 2).
Table 2 shows a clear seasonal pattern for the models
coefficients a, b, and c. Such coefficients have their lowest
values in the warmer and humid months, October to April,
whereas from May to September, they have the highest values,
mainly the coefficient a, which is the factor of curve flattening. For this reason, the curve of these 3 months seems to miss
the sigmoidal inflection point, becoming like an exponential
curve, as shown in Fig. 6.

Modeling temporal and spatial variability of leaf wetness

Fig. 6 Models to estimate mean
monthly leaf wetness duration
(LWD) as a function of mean
monthly relative humidity (RH),
in Brazil

The validation of the monthly LWD models, based on

independent data, showed a very good performance for all
months, with very high precision, r between 0.92 and 0.96,
good accuracy, dr ranging from 0.83 to 0.88, which resulted in
optimum performance, and Pi0.77 in all months. Regarding
the errors, ME showed a slight tendency of over estimation
during February (0.29 h day1), April (0.03 h day1), May
(0.31 h day1), July (0.14 h day1), August (0.34 h day1), and
November (0.01 h day1), whereas for the other months, such
tendency was of underestimation, with ME ranging from
0.27 to 0.03 (Table 2). The MAE ranged between 0.9 and
1.3 h day 1 , whereas RMSE varied between 1.2 and
1.7 h day1.

3.2 Geostatistical analysis

The geostatistical model with the best fit to monthly and
annual RH and LWD spatial variability in Brazil was the
spherical theoretical semivariogram, with coefficient of determination higher than 0.89 and cross-validation above 0.67
(Table 3). The nugget effect was low in the majority of the
months, with Co below 15 %, except for October (Co=
29.2 %) and November (Co = 16.2 %). In August and
September, on the other hand, the Nugget effect was very
low, with Co below 5 %. These data resulted in strong SDI
for all months, except March and April, which had moderate
SDI. The range of spatial dependence was markedly seasonal

C.A. Alvares et al.

Table 2 Coefficients of the monthly leaf wetness duration (LWD) models for Brazil, statistical performance indices, and validation with independent
data
Month

Coefficients of sigmoidal model

Goodness-of-fit statistics

Validation

R2adj

p value

ME
h day1

MAE

RMSE

dr

Pi

January
February
March
April
May
June

24.24
26.47
25.40
27.06
38.23
38.73

7.39
7.58
7.74
8.02
9.50
10.67

84.30
85.65
84.89
85.70
91.66
92.80

0.91
0.88
0.88
0.89
0.90
0.88

<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001

0.94
0.96
0.94
0.92
0.93
0.93

0.27
0.29
0.25
0.03
0.31
0.03

1.20
1.00
1.30
1.10
1.10
1.20

1.60
1.20
1.60
1.50
1.50
1.70

0.84
0.86
0.83
0.84
0.83
0.84

0.79
0.83
0.78
0.78
0.77
0.78

July
August
September
October
November
December
Ann

66.61 n.s.
102.16 n.s.
43.78
31.35
26.83
24.57
31.31

11.81
12.51
10.30
9.30
8.32
7.65
9.13

102.03
108.93
93.56
88.44
85.72
84.45
88.47

0.87
0.89
0.84
0.88
0.89
0.89
0.87

<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001

0.92
0.96
0.94
0.94
0.94
0.95
0.94

0.14
0.34
0.03
0.10
0.01
0.17
0.04

1.20
1.00
0.90
1.10
1.00
1.00
1.11

1.60
1.30
1.20
1.50
1.30
1.20
1.51

0.83
0.87
0.85
0.85
0.84
0.86
0.84

0.77
0.84
0.80
0.80
0.79
0.81
0.80

a, b, c Coefficients of the regression model; R2 adj adjusted coefficient of determination; p value significance probability; r coefficient of correlation; ME
mean error; MAE mean absolute error; RMSE root-mean-square error; dr refined agreement index; Pi performance index; n.s. not significant at the 0.05
level

since between mid-spring and late summer, which are warmer

and wetter months in most parts of Brazil, the ranges of the
semivariograms were around double of those obtained in the
cooler and drier months from April to September. Crossvalidation showed significant correlations (p<0.001) for all
months, with Pearson correlation coefficients ranging from
0.67 (April) to 0.82 (September, October, and November) and

kriging standard deviations varying from 2.8 % (March) to

4.6 % (August).
3.3 Spatialization of the RH and LWD
Monthly RH and LWD were spatially described in maps with
1-km resolution (100 ha per pixel) as presented in Figs. 7 and 8,

Table 3 Models, parameters, and quality of experimental semivariograms adjusted for monthly relative humidity (RHm) fitting dataset in Brazil
Month

Model

Co
%

January
February
March
April
May

Sph.
Sph.
Sph.
Sph.
Sph.

13.30
10.00
13.19
13.10
12.40

June
July
August
September
October
November
December

Sph.
Sph.
Sph.
Sph.
Sph.
Sph.
Sph.

19.40
13.60
0.10
4.50
29.20
16.20
10.20

Ao
degree

C/(Co+C)
%

SDI

R2

86.26
65.92
43.10
44.06
59.50

20.82
20.17
19.16
11.07
9.11

84.6
84.8
69.4
70.3
79.2

Strong
Strong
Moderate
Moderate
Strong

0.98
0.99
0.96
0.94
0.97

91.60
119.50
143.10
147.60
143.00
134.40
146.20

8.66
8.83
8.51
8.62
18.68
19.99
21.25

78.8
88.6
99.9
97.0
79.6
87.9
93.0

Strong
Strong
Strong
Strong
Strong
Strong
Strong

0.94
0.95
0.97
0.97
0.94
0.91
0.89

Co+C

RSS
%
122.0
35.90
39.0
47.5
52.0
228.0
378.0
461.0
371.0
836.0
139.0
2160.0

r
0.76a
0.79a
0.74a
0.67a
0.71a
0.72a
0.77a
0.81a
0.82a
0.82a
0.82a
0.81a

Sph spherical, Co nugget, Co+C Sill (C structural variance), Ao range (degrees), SDI spatial dependence index, R2 model adjustment determination
coefficient, RSS residue sum of squares, r crossed validation correlation coefficient
a

Significant at the 0.001 level

Modeling temporal and spatial variability of leaf wetness

Fig. 7 Spatial and temporal mean monthly relative humidity (RH) variability in Brazil

respectively. Considering every month, the average RHm

ranges from 38 to 91 %. March and April are the wettest
months since they have RHm below 70 % in only few small

areas of the Northeastern Brazil more specifically in the state of

Bahia, where the semi-arid climate predominates (Alvares et al.
2013b). On the other hand, August and September are the driest

C.A. Alvares et al.

Fig. 8 Spatial and temporal mean monthly leaf wetness duration (LWD) variability in Brazil

months, with RHm lower than 60 % in the great part of the

country from the northeast to the northern part of the south
region, passing by the central-west region, where the states of

Mato Grosso, Mato Grosso do Sul, and Gois are. The

other months represent the transitional conditions. The
spatial variability of RHm during the winter months is

Modeling temporal and spatial variability of leaf wetness

far greater than observed for the autumn-winter months

(Fig. 7). This pattern is highlighted by the range (Ao)
of the semivariograms previously described (Table 3).
As observed in the map of Fig. 8, average monthly LWD in
Brazil ranges from less than 2 h day1, in the semi-arid region of
the country, to more than 16 h day1, mainly in the Amazon
region. In this latter region, where the states of Amazonas and
Par are located, the average monthly LWD is higher than
10 h day1 in all months, as well as observed in the coastal
region, from the Santa Catarina state to the south of Bahia state
(Fig. 8).

4 Discussion
The results presented by this study showed that sigmoidal models
(Fig. 6), based on RHm data, were able to estimate monthly LWD
for the Brazilian territory with good accuracy and high precision
(Table 2). The comparison between LWD estimates using the
models and those determined by the number of hours of RHm
90 % showed mean absolute errors of 0.9 to 1.3 h, which are
similar to those obtained with other models. Pedro Jr and
Gillespie (1982a, b) and Francl and Panigrahi (1997), using
physical models, found errors of the same magnitude and
smaller than those obtained by Rao et al. (1998), of 1.8 h, Luo
and Goudriaan (2000), of 2.1 h, Madeira et al. (2002), of 1.5 h,
Sentelhas et al. (2004a), of 2 h, and Dalla Marta et al. (2005), of
2.3 h. Sentelhas et al. (2008) using the same kind of approach, as
used in the present study, for estimating LWD for four different
locations around the world (Ames, USA; Elora, Canada;
Florence, Italy; and Piracicaba, Brazil), found mean absolute
errors ranging from 1.7 to 2.6 h when using a fixed RH threshold
for all locations. When the threshold was locally adjusted, these
errors were reduced to 1.2 to 1.9 h, which was confirmed when
the method was applied for independent data of the same locations. Our purpose here, as also mentioned by Sentelhas et al.
(2008), was not to assert that RH-based models are better than
more complex models to estimate LWD since the latter ones are
highly accurate and can be readily portable among climates and
regions (Gleason et al. 2008), but to show how useful these
simple models can be to estimate LWD when only RH data are
available. Besides the monthly sigmoidal models generated and
considering that the models coefficients a, b, and c have a
smooth seasonal variation (Table 2), these coefficients can be
easily interpolated for the transitions between months in order to
estimate LWD for shorter periods of time, e.g., 5-day or 10-day
periods, as suggested by Valeriano and Picini (2000) for air
temperature models.
As there is no historical data of LWD for different regions
of the country, which could be used for local calibration of the
RH threshold for modeling LWD, it was assumed in this study
that a LWD error of 1.7 to 2.6 h, as obtained by Sentelhas et al.
(2008) for different locations with distinct climatic conditions

in the world, would be acceptable for the purpose of understanding the temporal and spatial variability of LWD across
Brazil. Based on that, we interpolated the relative humidity of
358 weather stations in Brazil by kriging (Fig. 7), and then by
applying the sigmoidal models (Fig. 6), LWD was mapped
month by month in order to have its spatial and temporal
variability. The results, presented in Fig. 8, are the first attempt
to demonstrate the temporal and spatial variabilities of LWD
in a country as large and so broad climatic complexity as
Brazil (Alvares et al. 2013b). Despite the work done by
Hamada et al. (2008) for only one state of the country, So
Paulo, this study was the first assessment of LWD for the
whole country and is extremely important not only to show
the ability to implementing it but also to encourage new
research in this direction. Recently, researchers have reported
the difficulty to spatialize the relative humidity (Phillips and
Marks 1996) and leaf wetness duration for wide areas, such as
the study of Booth and Jovanovic (2012), which ended up
adapting the amount of rainfall for this purpose, even though
De Weille (1965) had warned that rainfall is an erroneous
substitute for leaf wetness duration.
Despite being a first approach, the results presented in
Fig. 8 represent a great advance in the climatology of LWD
for Brazil and will allow the development of other
agroclimatic studies related to crop and forest diseases control
plans, such as those conducted for determining the best sowing dates for reducing potato late blight risk (Lozada-Garcia
et al. 2008), the climatic favorability zones for sugarcane
orange rust (Santos 2012), and the climatic risk zoning for
eucalyptus rust (Silva et al. 2013). In addition, since the LWD
models proposed in this study provide high accuracy, the
results obtained allow us to move towards developing a plant
disease warning system at national level.

5 Conclusions
Monthly RH was properly spatialized from the climatological
normals of Brazil, showing a significant correlation under
cross-validation, strong spatial dependence, and low kriging
standard deviations. The sigmoidal models that estimate
monthly LWD as a function of RHm were highly significant
for the whole Brazil, representing a feasible way to understand
the LWD climatology in a national level. The spatial and
temporal LWD variabilities presented in this study highlight the differences that occur in Brazil for this weather
variable, being the first time it is presented for the
country. This approach for estimating LWD can be used
to support crop and forest disease control strategic plans
around the country, as well as providing a basis for
developing disease warning systems for different crops
and planted forests.

C.A. Alvares et al.

Acknowledgments We thank the data provided by the Brazilian National Institute of Meteorology (INMET).

References
Alfenas AC, Zauza EAV, Mafia RG, Assis TF (2009) Clonagem e
doenas do eucalipto, 2nd edn. UFV, Viosa
Allen DW (2011) Getting to know ArcGIS ModelBuilder. ESRI,
Redlands
Alvares CA, Stape JL, Sentelhas PC, Gonalves JLM (2013a) Modeling
monthly mean air temperature for Brazil. Theor Appl Climatol 113:
407427
Alvares CA, Stape JL, Sentelhas PC, Gonalves JLM, Sparovek G
(2013b) Kppens climate classification map for Brazil. Meteorol
Z 22:711728
Alvares CA, Gonalves JLM, Vieira SR, Silva CR, Franciscatte W (2011)
Spatial variability of physical and chemical attributes of some forest
soils in southeastern of Brazil. Sci Agric 68:697705
Alves MDC, de Carvalho LG, Pozza EA, Sanches L, Maia JDS (2011)
Ecological zoning of soybean rust, coffee rust and banana black
sigatoka based on Brazilian climate changes. Proc Environ Sci 6:35
49
Bass B, Savdie I, Gillespie TJ (1991) Simulation of leaf wetness duration
for field corn. Agric For Meteorol 57:6984
Booth TH, Jovanovic T (2012) Assessing vulnerable areas for Puccinia
psidii Eucalyptus rust in Australia. Austr P Path 41:425429
Bondalapati KD, Stein JM, Neate SM, Halley SH, Osborne LE,
Hollingsworth CR (2012) Development of weather-based predictive
models for fusarium head blight and deoxynivalenol accumulation
for spring malting barley. Plant Dis 96:673680
Brazil (1992) Climatological Normal (1961-1990). SPI/EMBRAPA,
Braslia
Burrough PA, McDonnell RA (1998) Principles of geographical information systems. Oxford University, New York
Carisse O, Levasseur A, Van der Heyden H (2012) A new risk indicator
for botrytis leaf blight of onion caused by Botrytis squamosa based
on infection efficiency of airborne inoculum. Plant Path 61:1154
1164
Chtioui Y, Panigrahi S, Francl L (1999) A generalized regression neural
network and its application for leaf wetness prediction to forecast
plant disease. Chemometr Intell Lab 48:4758
Dalla Marta A, Magarey RD, Orlandini S (2005) Modelling leaf wetness
duration and downy mildew simulation on grapevine in Italy. Agric
For Meteorol 132:8495
Dalla Marta A, Orlandini S, Ghironi M, Sabatini F (2004) Influence of
different sensor positions on leaf wetness duration measurements
and their effect on the simulation of grapevine downy mildew
(Plasmopara viticola). Idojaras 108:253263
De Weille GA (1965) The epidemiology of plant disease as considered
within the scope of agrometeorology. Agr Meteorol 2:115
Durigon A, van Lier QDJ (2013) Durao do perodo de molhamento
foliar: Medio e estimativa em feijo sob diferentes tratamentos
hdricos. Rev Bras Eng Agric Ambient 17:200207
Francl LJ, Panigrahi S (1997) Artificial neural network models of wheat
leaf wetness. Agric For Meteorol 88:5765
Gadgil PD (1977) Duration of leaf wetness periods and infection of Pinus
radiata by Dothistroma pini. New Zeal J For Sci 7:8390
Garratt JR, Segal M (1988) On the contribution of atmospheric moisture
to dew formation. Boundary-Lay Meteorol 45:209236
Gleason ML, Taylor SE, Loughin TM, Koehler KJ (1994) Development
and validation of an empirical model to estimate the duration of dew
periods. Plant Dis 78:10111016

Gleason ML, Duttweiler KB, Batzer JC, Taylor SE, Sentelhas PC,
Monteiro JEBA, Gillespie TJ (2008) Obtaining weather data for
input to crop disease-warning systems: leaf wetness duration as a
case study. Sci Agric 65:7687
Hamada E, Ghini R, Fernandes JL, Pedro MJ Jr, Rossi P (2008) Spatial
and temporal variability of leaf wetness duration in the State of So
Paulo, Brazil. Sci Agri 65:2631
Hamelin RC, Shain L, Thielges BA (1992) Influence of leaf wetness,
temperature, and rain on poplar leaf rust epidemics. Can J For Res
22:12491254
Hartkamp AD, White JW, Hoogenboom G (1999) Interfacing geographic
information systems with agronomic modeling: a review. Agron J
91:761772
Hirst JM (1957) A simplified surfacewetness recorder. Plant Pathol 6:
5761
Huber L, Gillespie TJ (1992) Modeling leaf wetness in relation to plant
disease epidemiology. Annu Rev Phytopathol 30:553577
Kim KS, Taylor SE, Gleason ML, Koehler KJ (2002) Model to enhance
site-specific estimation of leaf wetness duration. Plant Dis 86:179
185
Kim KS, Taylor SE, Gleason ML (2004) Development of a leaf wetness
model using a fuzzy logic system. Agric For Meteorol 127:5364
Kim KS, Taylor SE, Gleason ML, Villalobos R, Arauz LF (2005)
Estimation of leaf wetness duration using empirical models in
northwestern Costa Rica. Agric For Meteorol 129:5367
Kim KS, Taylor SE, Gleason ML, Nutter FW Jr, Coop LB, Pfender WF,
Seem RC, Sentelhas PC, Gillespie TJ, Dalla Marta A (2010) Spatial
portability of numerical models of leaf wetness duration based on
empirical approaches. Agric For Meteorol 150:871880
Kruit RJW, van Pul AJ, Jacobs AFG, Heusinkveld BG (2004)
Comparison between four methods to estimate leaf wetness duration
caused by dew on grassland. In: Conference on Agricultural and
Forest Meteorology, 26. American Meteorological Society,
Vancouver. p 10.1
Lau YF, Gleason ML, Zriba N, Taylor SE, Hinz PN (2000) Effects of coating,
deployment angle, and compass orientation on performance of electronic wetness sensors during dew periods. Plant Dis 84:192197
Leca A, Parisi L, Lacointe A, Saudreau M (2011) Comparison of
Penman-Monteith and non-linear energy balance approaches for
estimating leaf wetness duration and apple scab infection. Agric
For Meteorol 151:11581162
Lomas J, Shashoua Y (1970) The performance of three types of leafwetness recorders. Agr Meteorol 7:159166
Lou WH, Goudriaan J (2000) Dew formation on rice under
varying durations of nocturnal radiative loss. Agric For Meteorol
104:303313
Lozada-Garcia BIL, Sentelhas PC, Tapia LR, Sparovek G (2008)
Climatic risk for potato late blight in the Andes region of
Venezuela. Sci Agric 65:3239
Madeira AC, Kim KS, Taylor SE, Gleason ML (2002) A simple cloudbased energy balance model to estimate dew. Agric For Meteorol
111:5563
Magarey RD, Russo JM, Seem RC (2006) Simulation of surface wetness
with a water-budget and energy balance approach. Agric For
Meteorol 139:373381
Mersha Z, Zhang S, Hau B (2014) Effects of temperature, wetness
duration and leaf age on incubation and latent periods of black leaf
mold (Pseudocercospora fuligena) on fresh market tomatoes. Eur J
Plant Pathol 138:3949
Monteith JL, Unsworth MH (1990) Principles of environmental physics,
2nd edn. Edward Arnold, New York
Park RF (1988) Effect of certain host, inoculum, and environmental
factors on infection of Eucalyptus species by two Mycosphaerella
species. T Brit Mycol Soc 90:221228
Pedro MJ Jr, Gillespie TJ (1982a) Estimating dew duration. I. Utilizing
micrometeorological data. Agric Meteorol 25:283296

Modeling temporal and spatial variability of leaf wetness

Pedro MJ Jr, Gillespie TJ (1982b) Estimating dew duration. II. Utilizing
standard weather station data. Agric Meteorol 25:297310
Phillips DL, Marks DG (1996) Spatial uncertainty analysis: propagation
of interpolation errors in spatially distributed models. Ecol Model
91:213229
Pinon J, Frey P, Husson C (2006) Wettability of poplar leaves influences
dew formation and infection by Melampsora larici-populina. Plant
Dis 90:177184
Rao PS, Gillespie TJ, Schaafsma AW (1998) Estimating wetness duration
on maize ears from meteorological observations. Can J Soil Sci 78:
149154
Rizzo DM, Garbelotto M (2003) Sudden oak death: endangering
California and Oregon forest ecosystems. Front Ecol Environ 1:
197204
Robertson GP (2008) GS+: geostatistics for the environmental sciences.
Gamma Design Software, Michigan
Rossi V, Salinari F, Poni S, Caffi T, Bettati T (2014) Addressing
the implementation problem in agricultural decision support
systems: the example of vite. net. Comput Electron Ag 100:
8899
Santos DL (2012) Climatic favorability zoning for the occurrence
of sugarcane orange rust in the main producing regions of
Brazil and Australia. MSc Thesis. ESALQ/University of So
Paulo, Brazil
Sentelhas PC, Gillespie TJ, Monteiro JEBA, Rowlandson T (2004a)
Estimating leaf wetness duration on a cotton crop from meteorological data. Rev Bras Agrometeorol 12:235245
Sentelhas PC, Gillespie TJ, Gleason ML, Monteiro JEBA, Helland ST
(2004b) Operational exposure of leaf wetness sensors. Agric For
Meteorol 126:5972

Sentelhas PC, Gillespie TJ, Batzer JC, Gleason ML, Monteiro JEBA,
Pezzopane JRM, Pedro MJ Jr (2005) Spatial variability of leaf
wetness duration in different crop canopies. Int J Biometeorol 49:
363370
Sentelhas PC, Gillespie TJ, Gleason ML, Monteiro JEBA, Pezzopane
JRM, Pedro MJ Jr (2006) Evaluation of a Penman-Monteith approach to provide reference and crop canopy leaf wetness duration
estimates. Agric For Meteorol 141:105117
Sentelhas PC, Dalla Marta A, Orlandini S, Santos EA, Gillespie
TJ, Gleason ML (2008) Suitability of relative humidity as an
estimator of leaf wetness duration. Agric For Meteorol 148:
392400
Shapiro SS, Wilk MB (1965) An analysis of variance test for normality:
complete samples. Biometrika 52:591611
Silva PHM, Miranda AC, Moraes MLT, Furtado EL, Stape JL, Alvares
CA, Sentelhas PC, Mori ES, Sebbenn AM (2013) Selecting for rust
(Puccinia psidii) resistance in Eucalyptus grandis in So Paulo
State, Brazil. For Ecol Manag 303:9197
Spotts RA (1977) Effect of leaf wetness duration and temperature on the
infectivity of Guignardia bidwellii on grape leaves. Phytopathol 67:
13781381
Strand JF (2000) Some agrometeorological aspects of pest and disease
management for the 21st century. Agric For Meteorol 103:7382
Theobald DM (2007) GIS Concepts and ArcGIS Methods, third ed.
Conservation Planning Technologies, Fort Collins
Valeriano MM, Picini AG (2000) Use of geographic information system
to generate maps of mean monthly temperature of the state of Sao
Paulo. Rev Bras Agrometeorol 8:255262
Willmott CJ, Robeson SM, Matsuura K (2012) A refined index of model
performance. Int J Climatol 32:20882094