Received: 28 July 2023 | Accepted: 5 November 2023

DOI: 10.1111/ppa.13839

REVIEW ARTICLE

Weather-based models for forecasting Fusarium head blight

risks in wheat and barley: A review

Taurai T. Matengu1 | Paul R. Bullock1 | Manasah S. Mkhabela1,2 | Francis Zvomuya1 |

Maria A. Henriquez1,3 | E. RoTimi Ojo1,4 | W. G. Dilantha Fernando5

1
Department of Soil Science, University of
Manitoba, Winnipeg, Manitoba, Canada Abstract
2
Climate Resilience Unit, Manitoba Fusarium head blight (FHB) is one of the most devastating crop diseases worldwide,
Agriculture, Winnipeg, Manitoba, Canada
3
significantly reducing the yield and quality of small-cereal crops such as wheat and
Morden Research and Development
Centre, Agriculture and Agri-Food barley when favourable weather conditions exist during anthesis. Additionally, FHB-
Canada, Morden, Manitoba, Canada associated mycotoxins significantly impact global food and feed safety. Controlling
4
Soil and Agriculture Weather
FHB with fungicides applied near anthesis reduces visual FHB symptoms and associ-
Surveillance, Manitoba Agriculture,
Winnipeg, Manitoba, Canada ated mycotoxin production, thereby lowering disease-related costs. However, when
5
Department of Plant Science, University weather conditions are unfavourable for FHB occurrence, fungicide application can
of Manitoba, Winnipeg, Manitoba, Canada
be costly and environmentally undesirable. Thus, fungicides should be used sparingly
Correspondence only when the pathogen is present and weather conditions are favourable. Modelling
Paul R. Bullock, Department of Soil
Science, University of Manitoba, of FHB risk using weather data has grown rapidly in recent decades and plays an es-
Winnipeg, Manitoba, Canada. sential role in integrated crop disease management. In this review, several weather-
Email: paul.bullock@umanitoba.ca
based FHB models are selected and described in detail. The models were developed
Funding information globally for assessing the real-time risk of FHB epidemics in various regions/countries.
Agriculture and Agri-Food Canada;
Alberta Wheat Commission; Brewing Most of these models are site-specific and predict FHB visual observations such as
and Malting Barley Research Institute; the incidence and severity of FHB, Fusarium-damaged kernels (FDK), and also deox-
Canadian Agricultural Partnership;
Integrated Crop Agronomy Cluster; ynivalenol (DON) levels. The review also highlights the limitations of these existing
Manitoba Crop Alliance; Saskatchewan models, including their narrow applicability, low accuracy for high-risk contamination
Wheat Development Commission;
Western Grain Research Foundation; situations, and omissions of certain factors. Also discussed are potential avenues for
Prairie Oat Growers Association improvement and enhanced predictive capabilities including consideration of addi-
tional disease risk factors as well as a broader range of varieties. These predictive
models can assist producers, regulatory agencies, and industry to mitigate potential
food and feed security and safety concerns.

KEYWORDS
barley, deoxynivalenol, Fusarium head blight, weather-based models, wheat

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2023 The Authors. Plant Pathology published by John Wiley & Sons Ltd on behalf of British Society for Plant Pathology.

Plant Pathology. 2023;00:1–14.  wileyonlinelibrary.com/journal/ppa | 1

 |
2 MATENGU et al.
1 | I NTRO D U C TI O N
Epidemics of Fusarium head blight (FHB) in small-cereal crops such
as wheat and barley are frequently triggered when weather condi-
tions favourable for infection by Fusarium species, predominantly
Fusarium graminearum, coincide with flowering and early kernel fill-
ing (De Wolf et al., 2003; McMullen et al., 2012). Numerous studies
and surveys have established that temperature and moisture (rain-
fall and humidity) are the primary determinants of FHB develop-
ment (Del Ponte et al., 2009; Hooker et al., 2002; Shah, De Wolf,
et al., 2019). If FHB is left uncontrolled, wheat and barley heads are
blighted, and severe yield reductions occur due to flower abortion,
which reduces the number of kernels formed (Bai & Shaner, 2004).
Grain filling is also impacted, leading to Fusarium-damaged kernels
(FDK), often light in weight (Góral et al., 2018). These yield losses are
exacerbated by the production of mycotoxins such as deoxynivale-
nol (DON) by the causative pathogen (i.e., Fusarium species), which
are toxic to both humans and livestock (Tamburic-Ilincic et al., 2015).
DON-contaminated grains may be unfit for human consumption or
the production of products such as bread, beer and animal feed,
and are thus frequently downgraded during marketing (Dahl &
Wilson, 2018).
Due to the high cost of FHB damage to a crop, some produc-
ers use preventative fungicide applications to protect the crop
from disease without knowing if FHB will cause significant damage
(Nita, 2013). Under low predicted risk, fungicide overuse can be det-
rimental to the environment, and the marginal cost-to-revenue ratio
of wheat makes it critical to use inputs cost-effectively, including
fungicides (Wallhead & Zhu, 2017). Fungicide spray for FHB sup-
pression in wheat and barley production can be reduced if the risk
of a disease epidemic can be accurately predicted. This could result
in more sustainable and environmentally-friendly wheat and barley
production. Forecasting models are appropriate for FHB due to the
disease's sporadic manifestation, reliance on weather factors, and
the relatively short time window for pathogen sporulation, inoculum
dispersal and host infection (De Wolf et al., 2003; Góral et al., 2018).
Researchers worldwide have created, validated and adapted weath-
er-based models to predict the likelihood of the presence and sever-
ity of FHB and DON toxin levels so that informed in-season decisions
can be made and post-season marketing decisions can be anticipated
(Birr et al., 2019; Del Ponte et al., 2005; Rossi, Giosue, Pattori, & Del
Vecchio, 2003). These models are primarily based on meteorological
data such as temperature, relative humidity and precipitation, which
are sometimes combined with agronomic variables such as crop res-
idues, tillage, crop rotation and crop variety. Some of these models
are specific to where they are developed, so researchers have tested
these models in different climatic zones to determine how well they
work. After a few modifications, some of these models worked well
in other locations and crop types (Giroux et al., 2016; Schaafsma &
Hooker, 2007) and sometimes they did not (Landschoot et al., 2013).
This review discusses various weather-based FHB models developed
around the world for assessing the real-time risk of FHB epidem-
ics, as well as their application and adoption in different regions/
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
countries. The selection criteria included only published field stud-
ies that offer predictions on FHB occurrence in wheat and barley,
anchored primarily by weather-driven variables. Additionally, only
those models with published model equations were considered.
The goal is to review a range of FHB risk model examples to identify
some common characteristics as well as knowledge gaps. When the
model developers did not specify a name for a model, the author's
name(s) is used.
2 | FU SA R I U M H E A D B LI G HT CO NTRO L
Control of FHB outbreaks is crucial due to the substantial loss in
crop yield and quality and the adverse health effects of mycotoxins
on consumers. Besides cereals, Fusarium species infect other crops
grown in western Canada, causing diseases such as root rot in soy-
beans, peas and beans; ear rot in maize; and Fusarium wilt in canola
(Foroud et al., 2019; Gilbert & Fernando, 2004). As a result, FHB
disease management with a single strategy such as crop rotation is
challenging (Harris et al., 2016). Therefore, it is essential to combine
multiple control strategies to manage FHB (Bai & Shaner, 2004).
Tillage and crop rotation with non-host crops are suitable agri-
cultural activities that reduce the likelihood of FHB epidemics by
eliminating wintering plant residues and reducing inoculum sources
(Spolti et al., 2014). While reducing inoculum levels and disease epi-
demics with high-tillage levels is significant, it is not enough to con-
trol FHB (Gilbert & Fernando, 2004). The selection of varieties with
some resistance to FHB also plays a crucial role in managing FHB.
Several FHB resistance types have been identified, for example,
Type I (resistance to initial infection) and Type II (resistance to patho-
gen spread throughout the spike; Góral et al., 2020; Ha et al., 2016).
Additionally, the application of fungicides during flowering greatly
reduces FHB epidemics (Yoshida et al., 2012). The efficacy of chemi-
cal controls relies on predicting the best possible time for fungicide
application, as their effectiveness is much lower if applied either
prior to or after anthesis (Bolanos-Carriel et al., 2020).
3 | D I S E A S E FO R EC A S TI N G M O D E L S
FHB forecasting has gained popularity in recent decades to maximize
fungicide efficacy while also minimizing its cost (Musa et al., 2007).
FHB is ideal for forecasting due to its dependency on environmental
conditions and the limited time period near the anthesis stage for
host infection (De Wolf et al., 2003; Rossi, Giosue, Pattori, & Del
Vecchio, 2003). FHB disease forecasting, days or weeks before in-
fection or major epidemic, helps producers to respond quickly and
efficiently by modifying crop management practices to include fun-
gicide application (Shah et al., 2013). FHB forecasting has adopted
different strategies including (i) predicting the risk of disease oc-
currence based on FHB incidence and severity, (ii) forecasting the
FDK risk, and (iii) forecasting the concentration of DON in harvested
grain. Because FHB is a global threat to cereals, the origins for

MATENGU et al.
disease prediction models have come from different research organ-
izations in many different countries. These groups conduct their re-
search locally and the approaches that have been developed for FHB
disease risk prediction have tended to be regional in nature. Thus,
this review has taken the approach to present the models based on
these geographical distinctions in the types of models to provide a
comprehensive understanding of FHB forecasting strategies.
3.1 | Research on FHB prediction models from the
United States
The first models for FHB risk assessment in wheat in the United States
were logistic regression models developed by De Wolf et al. (2003).
These models used data gathered from four states (Ohio, North
Dakota, Missouri and Kansas) for a total of 50 site-years. Field FHB
severity ≥10% (congruent to an FHB index ≥10%) was considered
an epidemic in these forecasting models. Disease forecasting used
independent variables constructed from observed hourly weather
data for two time periods, either from 7 days before 50% anthesis or
10 days after 50% anthesis. A total of 49 variables were constructed
from (i) the calculation of daily minimums, maximums, averages and
sums; (ii) the duration that prespecified conditions were met; (iii) the
functional response to temperature and moisture; and (iv) the calcu-
lation of interaction terms between predictor variables to determine
which were most aligned with FHB severity ≥10%.
Eleven different prediction models were developed and three
were selected for closer scrutiny, namely model A, model B and
model I (De Wolf A, B and I, respectively). The De Wolf A model
used only the duration of temperature between 15°C and 30°C and
relative humidity (RH) ≥90% for 10 days post-anthesis (TRH9010;
Equation 1), while De Wolf B used the interaction of duration of tem-
perature between 15°C and 30°C for 7 days pre-anthesis (T15307)
and TRH9010 (Equation 2). In contrast, De Wolf I used only pre-an-
thesis weather variables including T15307 and the duration of pre-
cipitation in the 7 days pre-anthesis (DPPT7; Equation 3).
(1)
( )
P = 1 ∕ (1 + exp − − 3.3756 + 6.8128TRH9010
(2)
( )
P = 1 ∕ (1 + exp − − 3.7251 + 10.5097INT3
P = 1 ∕ (1 + exp − − 8.2175 + 8.4358T15307 + 4.7319DPPT7 (3)
( )
where P is the probability (from 0 to 1) of FHB severity ≥10%, TRH9010
is the duration of temperature between 15°C and 30°C and RH ≥90%
for 10 days post-anthesis, T15307 is the duration of temperature be-
tween 15°C and 30°C for 7 days pre-anthesis, DPPT7 is the duration
of precipitation in the 7 days pre-anthesis and INT3 is an interaction
term between T15307 and TRH9010 (De Wolf et al., 2003). To use
Equations (1)–(3), variables must first be placed on the same scale as
the data used to develop the models. This is done by dividing TRH9010,
T15307 or DPPT7 by the observed maximum during the study or 136,
168 and 39, respectively (TRH9010/136; T15307/168; DPPT7/39) (De
Wolf et al., 2003).
|
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3
De Wolf et al. (2003) noted that the models with post-anthesis
weather variables had higher accuracy than those with just pre-an-
thesis weather variables. Model accuracy (i.e., successful classifica-
tion of FHB epidemics and non-epidemics) was 84%, 84% and 70%
for Models A, B and I, respectively. Further validation of Models A
and B with an independent dataset showed an accuracy of 78%,
which was perceived to be good (De Wolf et al., 2003). Despite its
lower accuracy, Model I was of interest because it had the potential
to provide predictions early enough to make management decisions
before anthesis. De Wolf et al. (2003) suggested that the accuracy
of FHB prediction models could be improved using new represen-
tations of precipitation variables, such as incorporating the inten-
sity and frequency of rainfall events during anthesis, as opposed to
solely focusing on the duration of precipitation. Additionally, they
suggested that incorporating information about local inoculum
sources, including the presence and type of crop residues, could be
useful predictors of disease at a local level, when coupled with key
weather variables.
The De Wolf models were used in an online FHB risk assessment
tool (i.e., Fusarium Risk Tool; www.​wheat​scab.​psu.​edu), for wheat
production in 31 states across the United States, despite initially
being developed using data from just four states. While it is unclear
how well these four states represent the other states where the
models were used, the widespread and successful adoption of these
models is an indication of their effectiveness in assessing FHB risk
in wheat production. Nevertheless, it is essential to note that the
performance of the models may differ between states due to envi-
ronmental and crop factors that are unique to each state. The tool
provides risk maps showing the probability of a severe FHB outbreak
(FHB ≥10%) based on weather data from multiple weather stations in
the states covered by the prediction system (McMullen et al., 2012).
Users enter data such as flowering date, crop type (winter wheat or
spring wheat), and crop variety (very susceptible, susceptible, mod-
erately susceptible or moderately resistant) into the tool to get FHB
risk at the field scale (McMullen et al., 2012; Shah et al., 2021).
More recently, efforts were made by Shah et al. (2013, 2014)
to understand the role of pre-and post-anthesis weather indicators
on FHB epidemic risk forecasting (i.e., FHB epidemics again defined
as an FHB index ≥10%). They used an enormous dataset collected
across 15 states in the United States over 27 years and an extensive
number of RH, temperature and rainfall predictors for time periods
ranging from 15 days pre-anthesis to 15 days post-anthesis. These
models added wheat resistance category (RESIT) as an ordinal vari-
able, where 0, 1, 2 and 3 represented very susceptible, susceptible,
moderately susceptible and moderately resistant, respectively (Shah
et al., 2013). The odds of FHB index ≥10% on susceptible, moder-
ately susceptible and moderately resistant cultivars were 47%, 73%
and 78% lower, respectively, than the odds of a major FHB epidemic
on a very susceptible cultivar. Out of 380 total weather-based pre-
dictors through the various pre- and post-anthesis periods, their
selection process identified 21 variables as the most important
predictors, and these were integrated into 15 separate logistic re-
gression models. These models differed in the pre- or post-anthesis
 |
4 MATENGU et al.
duration of their predictors, with specific RH, temperature or rainfall
conditions (Shah et al., 2013). All 15 models offered a better sensi-
tivity-specificity balance than updated versions of the original mod-
els developed by De Wolf et al. (2003). The presence of a greater
number of weather-based predictors in a model was found to be
linked to a reduction in misclassification rates. Rainfall was generally
not crucial for forecasting the likelihood of epidemics in this study
(Shah et al., 2013). Their winter wheat (Equation 4) and spring wheat
(Equation 5) FHB risk models were as follows:
logit(𝜇) = − 1.7954 + 0.0245TH2 (4)
logit(𝜇) = − 11.008 − 0.9578RESISTC + 0.1516H1 (5)
where μ is the probability of a significant FHB epidemic (FHB index
≥10%), H1 is the mean hourly RH in the 7 days pre-anthesis, and TH2
is the number of hours during which the following two conditions are
met simultaneously within a given hour: temperature of 9°C–30°C and
RH ≥90% in the 7 days pre-anthesis. RESISTC is a categorical variable
with four different levels of disease resistance: very susceptible = 0;
moderately susceptible = 1; moderately resistant = 2; and resistant = 3
(Shah et al., 2013).
The US FHB prediction models described above were devel-
oped using temperature and moisture-based variables during short
periods of ≤15 days preceding and following anthesis (De Wolf
et al., 2003; Shah et al., 2013, 2014). This was based on the idea that
FHB prediction should be made in time to provide fungicide applica-
tion recommendations during anthesis or no later than 5 days after
anthesis to control the disease effectively (De Wolf et al., 2003;
Shah et al., 2013) and corresponding to the period of peak suscepti-
bility of wheat to the FHB fungal pathogen (Kazan et al., 2012).
Shah, De Wolf, et al. (2019) and Shah, Paul, et al. (2019) con-
ducted a functional data analysis of entire weather time series from
120 days before to 30 days after anthesis at 865 site-years and com-
pared FHB epidemics (FHB index ≥10%) to FHB non-epidemics (FHB
index <10%). The two classes were compared to determine whether
the mean functional weather curves differed statistically between
FHB epidemics and non-epidemics. Although the largest differences
between weather time series for epidemics and non-epidemics were
found close to anthesis, the results indicated that FHB epidemics
and non-epidemics have distinct differences in some weather pa-
rameters, particularly atmospheric moisture variables such as RH, as
early as 40 days before anthesis (Shah, De Wolf, et al., 2019; Shah,
Paul, et al., 2019). Differences in the mean functional curves of daily
average temperature between epidemics and non-epidemics were
relatively small and confined to narrow time windows around anthe-
sis. A better temperature variable was the daily difference between
the maximum and minimum observed hourly temperature, as had
been reported previously by Landschoot et al. (2013) in Belgium. The
findings suggest that earlier detection of FHB epidemics is possible
but must be captured in predictive models that use weather time
series for variables that are consistently well-separated between
epidemics and non-epidemics for extended time periods (Shah, De
Wolf, et al., 2019; Shah, Paul, et al., 2019).
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bondalapati et al. (2012) evaluated and reported significant pos-
itive correlations between DON concentration and both FHB inci-
dence and FHB severity in the northern Great Plains of the United
States. Using disease and weather data from six growing seasons
(51 location-years), they developed and validated predictive models
for DON epidemics (DON concentration ≥0.5 ppm) in barley using
logistic regression and the Weibull function to represent the non-
linear relationship of DON to temperature and wetness. The logis-
tic regression approach was the same as that utilised by De Wolf
et al. (2003) and Shah et al. (2013, 2014). Bondalapati et al. (2012)
developed and evaluated nine prediction models based on different
temperature and wetness duration variables near anthesis (±5 days),
which are indicative of ideal conditions for FHB infection. The tem-
perature variables included average temperature (AvgT), average
minimum temperature (AvgMinT) and average maximum tempera-
ture (AvgMaxT). The wetness duration variables encompassed hours
with RH ≥90% (RH90), duration of continuous hours with RH ≥90%
(drRH90), and weighted hours with RH ≥90% (wRH90). Their Model
3, which used AvgT and the weighted time (in hours) when the RH
≥90% (wRH90) during the critical period (anthesis ±5 days), was found
to have the best overall performance (a combination of accuracy,
sensitivity and specificity) for predicting risk of DON ≥0.5 ppm. The
model accuracy was 84% using the training dataset and 79% using
an independent validation dataset. Bondalapati et al. (2012) noted
that, even though cultivar was not a significant factor when included
in the model, their observations indicated that a model with differ-
ent thresholds varying with cultivar resistance might provide more
accurate predictions. They also commented that their study was lim-
ited to six growing seasons and did not necessarily represent all of
the possible weather conditions that could occur in the region and
influence FHB disease development or DON accumulation in barley.
Therefore, they concluded that further validation was required.
3.2 | Research on FHB prediction models
from Canada
In western Canada, some US models have been adopted for pre-
dicting FHB severity greater than 10%. In Manitoba and Alberta,
FHB risk assessment maps were developed based on De Wolf I
(Equation 3), whereas in Saskatchewan, FHB risk assessment maps
were originally based on the second-generation model developed
by Shah et al. (2013) that incorporated agronomic and climatic
variables (ACIS, 2021; MARD, 2021; SWDC, 2021). The lack of
validation research on FHB risk assessment models used in the
prairie provinces of western Canada can be partly attributed to
the time-consuming and costly nature of gathering and analys-
ing such data. A recent study used a logistic regression approach
similar to De Wolf et al. (2003) to identify weather variables
most closely related to FHB, FDK and DON epidemics in spring
wheat, winter wheat, barley and durum wheat on the prairies
(Matengu, 2022). A series of several different risk models were
developed using data collected at plot sites in 2019 and 2020 and

MATENGU et al.
then were independently validated using data collected in more
than 200 producer fields across the prairies. Seven of those risk
models, shown below (Equations 6–12), were selected to be used
in a prairie-wide FHB risk mapping tool (https://​w ww.​youtu​b e.​
com/​watch?​v=​z8Ewf​r w_​j6c&​t= ​31s).
(6)
{ [ ( )]}
FHBww = 1 ∕ 1 + exp − − 5.1095 + 0.0312RH8014
FHBsw =
(7)
{ [ ( )]}
1∕ 1 + exp − − 34.5786 + 0.3513RHmx14 + 0.0435T252814
{ [ ( )]}
FHBb = 1 ∕ 1 + exp − − 3.7724 + 0.2146R14 + 0.0495T252814
(8)
(9)
{ [ ( )]}
FHBd = 1 ∕ 1 + exp − − 2.0665 + 0.0326TRH8010
(10)
{ [ ( )]}
FDKsw = 1 ∕ 1 + exp − − 31.6372 + 0.4004RH10
(11)
{ [ ( )]}
FDKd = 1 ∕ 1 + exp − − 11.9932 + 0.0847RH8010
(12)
{ [ ( )]}
DONd = 1 ∕ 1 + exp − 24.1039 + 0.3114RH14
where FHBww, FHBsw, FHBb and FHBd are the probability values for
FHB index values ≥5% in winter wheat, spring wheat, barley and durum
wheat, respectively; FDKsw is the probability for FDK ≥0.3% in spring
wheat; FDKd is the probability for FDK ≥2.0% in durum; DONd is the
probability for DON ≥1 ppm in durum; RH8014 is the duration (hours)
[ ( )2
DON = exp − 2.15 + 2.21RAINA − 0.61 RAINA + 0.85RAINB + 0.52RAINC − 0.30TMIN − 1.10TMAX − 0.1
of RH ≥80% in the 14 days prior to mid-anthesis; RHmax14 is the mean
daily maximum RH (%) in the 14 days prior to mid-anthesis; T252814 is
the duration (hours) with 28 ≥ air temperature ≥ 25 in the 14 days prior
to mid-anthesis; R14 is the mean daily rainfall (mm) in the 14 days prior
to mid-anthesis; TRH8010 is the duration (hours) with both 30 ≥ air tem-
perature ≥ 15 and RH ≥80% in the 10 days prior to mid-anthesis; RH10
is the mean daily RH (%) in the 10 days prior to mid-anthesis; RH8010 is
the duration (hours) with RH ≥80% in the 10 days prior to mid-anthesis
and RH14 is the mean RH (%) in the 14 days prior to mid-anthesis. The
model accuracies ranged from 75% to 85% on the training data and
from 70% to 100% on the validation data. It should be noted that the
dry weather conditions during the study created a limited number of
epidemic samples in the validation dataset and the accuracy of the dif-
ferent models is mainly or completely a result of the models' specificity.
Although the Matengu (2022) study was conducted during a drier pe-
riod with limited FHB disease pressure, the models developed showed
higher accuracy than the De Wolf Model I and the Shah et al. (2013)
models being used in the prairie provinces, when they were tested
using the same weather and FHB data. Very recently, Manitoba has
adopted the spring wheat FHB risk model (Equation 7) as its new in-
dicator of FHB risk in the province commencing in the spring of 2023.
Current research is underway to build a larger dataset of weather and
FHB disease observations on the prairies to improve the accuracy of
the models and delineate differences in FHB response between culti-
vars with varying genetic resistance to the disease. The development
|
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
5
of common FHB models across all prairie provinces (Matengu, 2022)
will improve understanding of FHB epidemics and enhance regional
collaboration among researchers and stakeholders to promote effec-
tive FHB management.
Forecasting the FHB index is useful but some infected kernels
are, in some instances, asymptomatic even though they are in-
fected with mycotoxin (Hooker et al., 2002). One limitation of the
predictability of DON epidemics is that visual FHB symptoms ob-
served during grain filling are not necessarily associated with DON
incidence (Cowger et al., 2020; Kelly et al., 2015; Matengu, 2022;
Paul et al., 2005). In Ontario, Canada, a model predicting DON con-
centration in wheat grain was developed by Hooker et al. (2002)
and consisted of three regression equations. The model was further
improved through the years into a forecasting tool called DONcast
(Schaafsma & Hooker, 2007). In the model, the grain DON level is de-
pendent on three critical periods. The first critical period is 4–7 days
before heading and calculated by Equation 13:
[ ( )2 ]
DON = exp − 0.30 + 1.84RAINA − 0.43 RAINA − 0.56TMIN − 0.1
(13)
where DON concentration (mg/g) is a function of RAINA (the num-
ber of days with rain >5 mm/day), and TMIN is the number of days of
temperature < 10°C between 4 and 7 days before heading (Hooker
et al., 2002). The second critical period is 7 days before heading to
10 days after heading and calculated using Equation 14:
]
(14)
where RAINB is the number of days of rain >3 mm/day in the period
3–6 days after heading, TMIN is the number of days with tempera-
ture < 10°C between 4 and 7 days before heading, and TMAX is the
number of days with temperature > 32°C in the period 4 and 7 days
before heading. The third critical period included RAINC (number of
days of rain >3 mm/day in the period 7–10 days after heading), and is
given in Equation 15:
(15)
( )
DON = exp − 0.84 + 0.78RAINA + 0.40RAINc − 0.42TMIN − 0.1
Rainfall, high humidity and warm temperatures were favour-
able for disease development during all three periods, and DON
concentrations in the increased set of weather variables for the
three periods varied marginally. Daily rainfall >5 mm increased the
potential DON concentration in the first critical period. In contrast,
DON concentration decreased when the daily minimum tempera-
ture was <10°C. In the second critical period, DON levels were re-
duced with mean temperature >32°C. In both the second and third
critical periods, daily rainfall >3 mm and RH > 80% increased DON
concentration in the grain, while temperature < 10°C reduced DON
concentration (Schaafsma & Hooker, 2007).
The model developed by Hooker et al. (2002) explained 73% of
the variability in DON concentration across 5 years and 399 sam-
ples. Prediction of DON concentration lower than 1 mg/L showed
a high accuracy of 89%. The Hooker model was further enhanced
by incorporating FHB observations into the dataset and variables of
 |
6 MATENGU et al.
crop history, tillage and host susceptibility to render it field-specific.
The Hooker model was further developed by Weather Innovations
(WIN) Consultancy LP into the world's first commercialized DON
forecasting system, (i.e., DONcast), accessible in different parts of
the world (Schaafsma & Hooker, 2007). DONcast is reasonably sta-
ble across the years, continents and cropping systems due to the
robustness of the data used for its development. For example, one
study found DONcast accuracy of 80%–85% and explained 72% of
DON variation in over 1000 field samples from four countries using
a 1 mg/L threshold level (Schaafsma & Hooker, 2007).
The DONcast model considers a wide range of agronomic fac-
tors, such as varietal susceptibility to disease and the amount of
crop residue on the soil surface. However, it was difficult to get ac-
curate DON forecasts and fungicide treatment recommendations
due to over-generalized agronomic factors and regional variability,
which led to low system adoption (Pitblado et al., 2007). Thus, the
DONcast model was refined further by including other meteorologi-
cal variables such as leaf wetness and agronomic factors such as crop
history, tillage and crop FHB resistance to make it more field-spe-
cific and improve its forecasting accuracy (Pitblado et al., 2007;
Schaafsma & Hooker, 2007). As a result, the DONcast has been
renamed site-specific DONcast (ssDONcast) (Pitblado et al., 2007;
Schaafsma & Hooker, 2007).
A study in Quebec, Canada, evaluated nine different FHB models
for their ability to forecast FHB index, FDK and DON under local
conditions (Giroux et al., 2016). The models that were tested included
Hooker et al. (2002), DONcast (Schaafsma & Hooker, 2007), De
Wolf A, B and I, Molineros (2007), Moschini et al. (2001), and the two
indices developed by Rossi, Giosue, Pattori, and Del Vecchio (2003).
All of these FHB models, except Molineros (2007), are described in
separate sections of this paper. Molineros (2007) developed models
based on the De Wolf et al. (2003) models but using only average
RH (%) for the 7 days before flowering and integrating four levels of
cultivar susceptibility. The De Wolf A, B, I, Molineros, and Moschini
models are empirical and make predictions using polynomial equa-
tions. The Hooker et al. (2002), DONcast and Rossi models are more
complex with FHB risk determined using forecasts at different plant
development stages.
During the Giroux et al. (2016) study period, there was insuffi-
cient disease pressure to create enough cases with FHB index ≥10%,
FHB index ≥5% and DON ≥2 ppm for a statistically robust receiver
operating characteristic (ROC) analysis. However, the models were
evaluated using ROC analysis for their accuracy to predict FHB ep-
idemics defined by FDK ≥1.5%, FDK ≥1.0% and DON ≥1 ppm be-
cause there were a sufficient number of epidemic cases for robust
analysis. The DON ≥1 ppm disease indicator best differentiated the
differences in the performance of the nine models. The De Wolf A,
B and Moschini models all displayed a good combination of accu-
racy, sensitivity, and specificity for DON ≥1 ppm, even though these
models were not designed to predict DON levels directly but rather
field FHB index. The Molineros model underperformed with re-
gard to specificity and the Hooker et al. (2002), DONcast and Rossi
models underperformed in regards to sensitivity. Thus, some of the
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
simplest models had the best performance despite being developed
in a different geographical region. The De Wolf I model never pro-
vided reliable predictions and had low overall accuracy, sensitivity
and specificity. Giroux et al. (2016) commented that the De Wolf I
model and the Molineros model both use only pre-anthesis weather
conditions, which may explain their low performance.
3.3 | Research on FHB prediction models
from Argentina
A site and year-specific empirical model to predict the incidence
of FHB (predictive index or PI%) was developed by Moschini and
Fortugno (1996) for the cooler Pergamino wheat region in Argentina
using cultivars of different FHB susceptibility, temperature and pre-
cipitation as predictor variables. Equations 16 and 17 were used to
predict head blight incidence (PI).
PI( % ) = 20.37 + 8.63 × NP2 − 0.49 × DD926 (16)
PI( % ) = 18.34 + 4.12 × NP12 − 0.45 × DD1026 (17)
where NP2 is the number of 2-day periods of precipitation ≥0.2 mm,
and RH > 81% and ≥ 78% for the first and second day, respectively.
NP12 is the total NP2 periods plus the number of days with precipi-
tation ≥0.2 mm and average RH > 83%. Maximum and minimum tem-
peratures were used to calculate degree days between 9 and 26°C.
The observations start 8 days before heading (Zadoks stage 51) and
extend up to 530-degree days accumulation. In the equations, DD926
and DD1026 represent 926 and 1026 accumulated extreme tempera-
tures, respectively, which are calculated using Equations 18 and 19:
∑ [(
(18)
) ( )]
DD926 = Tmax − 26 + 9 − Tmin
∑ [(
(19)
) ( )]
DD1026 = Tmax − 26 + 10 − Tmin
If the maximum daily temperature is ≤26°C, the accumulation of
Tmax − 26 is set to zero, and if the minimum daily temperature is ≥9°C,
the 9 − Tmin is set to zero (Equation 18). Similarly, if the minimum
daily temperature is ≥10°C, 10 − Tmin is set to zero (Equation 19).
Therefore, DD values accumulate only on days with extreme high
and low temperatures.
Predicted disease severity is estimated using the calculated PI
values in Equations (20) and (21).
PSms = − 1.010225 + exp(0.137449 × PI) (20)
PSs = − 1.02573985 + exp(0.27209 × PI) (21)
where PSms is predicted head blight severity (0 to 1) for moder-
ately susceptible wheat cultivars, PSs is predicted head blight se-
verity (0 to 1) for susceptible wheat cultivars and PI is calculated
in Equations (16) and (17). Equations (20) and (21) were developed
from 64 and 31 data points, respectively with R 2 values of 0.947 and
0.907, respectively.

MATENGU et al.
When the equations above were used in warmer wheat regions
of Argentina, there was a tendency for underestimation at high dis-
ease levels. The model was adapted to these warmer regions in more
northern Argentina (Moschini et al., 2001). The maximum daily tem-
perature threshold in Equations (18) and (19) was changed from 26°C
to 30°C and the number of degree days in the critical period was
increased from 530 to 550. The revised model offered predictions
closer to the observed infection levels (Moschini et al., 2001). Unlike
logistic regression models, this Argentinian model does not have an
action threshold (i.e., a specific level that triggers fungicide appli-
cation). It directly predicts the level of FHB incidence and severity
based on air temperature and precipitation during the critical time
period for infection (Moschini et al., 2001).
3.4 | Research on FHB prediction models from Italy
Rossi et al. (2002); Rossi, Giosue, Pattori, and Del Vecchio (2003);
Rossi, Giosue, and Delogu (2003) developed an epidemiological
model in Italy, in which weather variables and information on wheat
growth stages were used to predict the risk of FHB and DON. Model
development was based on four problematic FHB pathogens in Italy
(Fusarium culmorum, Fusarium graminearum, Fusarium avenaceum
and Microdochium nivalis). The model pivots on equations related
to the phases in the epidemiological cycle of FHB in wheat. In the
first equation, the amount of inoculum produced, or the sporulation
rate (SPO), was determined using four equations (one for each fun-
gal species) under controlled incubation conditions (Rossi, Giosue,
Pattori, & Del Vecchio, 2003). In Gibberella zeae, SPO was estimated
using the following equation:
)]0.24 [
[
8.59
∕ 1 + exp(5.52 − 0.51t) (22)
( ]
SPO = 25.98Teq × 1 − Teq
where Teq = equivalent of temperature calculated as (T − Tmin)/
(Tmax − Tmin); T = temperature (Tmin = 5°C and Tmax = 35°C) and t = incuba-
tion time (days).
Spore dispersal rate (DIS) was determined using precipitation,
which is the main factor influencing the rate of spore dispersal. Two
regression equations were developed, one adjusted to rainy days
(rain >0.2 mm) and the other for non-rainy days (rain <0.2 mm). The
regression equations also included rainfall intensity, daily average
temperature, and duration of hours with RH > 80% as shown below
(Rossi et al., 2002):
DIS1 = − 839.7 + 410.3W + 4.08T 2 + 115.45RintMax_P − 455.9Y95 (23)
DIS2 = − 682.3 + 45.68W + 21.5T 2 + RH80 + 107.0Rtot_P (24)
3
where DIS1 and DIS2 are conidia numbers (m of air per day) estimated
in days with and without rainfall, respectively; W is an empirical weight
assigned each day in a sequence of n successive rainy days, as fol-
lows: first rainy day = 1.1, second rainy day = 2.5, third rainy day = 1.2,
fourth or later rainy day = 0.8; T is the average air temperature (°C);
RintMax_P = maximum intensity of rain (mm/h) in the preceding day;
|
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7
Y95 = dummy variable for 1995, equal to zero (for the years 1994, 1996
and 1997) and 1 (for 1995); RH80 = number of hours with RH > 80%;
and Rtot_P = total rainfall (mm) on a preceding day (Rossi et al., 2002).
Disease infection frequency (INF) was based on leaf wetness du-
ration and temperature following inoculation and considering wheat
growth stages (GS) (Rossi, Giosue, Pattori, & Del Vecchio, 2003). For
G. zeae, INF was estimated using the following equation:
INF = − 0.099 − 0.363t + 0.07808 (T × t) − 0.00591 T 2 × t
( )
( 3 ) ( 4 ) (25)
+ 0.000199 T × t − 0.0000024 T × t
where T = temperature (°C) and t = incubation time (hours).
Finally, the invasion rate of mycelium into head tissues (INV) was
related to temperature. For G. zeae, INV was estimated using the
following equation:
)]1.35
(26)
[ (
INV = 5.33 × Teq 1.55 × 1 − Teq
where Teq = T − Tmin ∕ Tmax − Tmin , Tmin = 0 and Tmax = 38.
( ) ( )
Two risk indices are given by the Rossi model and are calculated
as below:
FHB _ risk = ΣSPO × DIS × INF × GS (27)
TOX _ risk = ΣINF × GS × INV (28)
The risk of FHB infection (FHB_risk) is used for the four Fusarium
species and the risk of mycotoxin development (TOX_risk) for F.
culmorum and F. graminearum (Rossi, Giosue, & Delogu, 2003). The
indices are computed daily and accumulate throughout the grow-
ing season. Data from various winter wheat crops were used to
validate the model. A comparison of actual infection and mycotoxin
against predicted data produced satisfactory results (R 2 > 0.8) (Rossi,
Giosue, & Delogu, 2003).
3.5 | FHB prediction model research from Brazil
A process-based FHB simulation model was developed in Brazil, ac-
counting for host, environment and inoculum dynamics during in-
fection (Del Ponte et al., 2005). Firstly, the host factor calculates
the groups of heads that emerged (cohort) on the same day (HNG)
(Equation 29). The daily rate in the cohort of heads as a function of
temperature (ANText) was then calculated (Equation 30).
HNG = 1 − exp − 0.0127t 2.4352 (29)
( )
ANText = 1 − exp at b (30)
( )
where t = 1 day, a = 0.225–0.029T + 0.0009T2 and
b = −5.773 + 0.966T − 0.0278T2, and T is daily mean temperature
(°C). Equation (29) assumes that every part of the cohort has its
first part extruded 3 days later. An empirical rule was determined
when anthers were attached to wheat spikelets before falling on
the ground (anther longevity). A rule was established based on em-
pirical observations reporting prolongation of flowering during a
 |
8 MATENGU et al.
series of cloudy days, indicating that flowers remained attached for
longer. The rule suggested that flower longevity is a minimum of
2 days and, if the daily solar radiation on the second or next day is
<10 MJ m−2 day−1, the anther remains attached for an extra day up
to a maximum of 5 days. Therefore, the proportion of anthers pres-
ent in 1 day (ANT) results from the sum of extruded and attached
anthers in each cohort of heads subtracted from anthers removed
from the cohorts (Del Ponte et al., 2005).
Coefficients for determining the proportion of susceptible tis-
sue (ST) were developed using ANT and coefficients for post-peak
anthesis infections. ST is ANT until ANT reaches the peak and de-
creases to 0.25. ST = 0.25 for the next 7 days after anthesis (ANT
<0.01), while ST = 0.10 for the next 8–14 days after anthesis. Such
guidelines were developed to compensate for potential late infec-
tions from post-peak flowering to kernel filling levels (Del Ponte
et al., 2005).
Secondly, the inoculum factor was derived from the night and
daytime inoculum observations. Spore cloud relative density was
estimated by adjusting a linear equation to the relative density of
colony-forming units observed during the night using Equation (31).
GZ = ( − 0.6306 + 0.0152RH + 0.176CRD)2
where RH = daily mean relative humidity (%) and CRD is a variable for
the rainy-day position (>0.3 mm) in consecutive rainy days (for four
consecutive days: CRD = 1, 2, 2.5 or 0.3 for each subsequent day); and
GZ is a fraction (0 < GZ <1), which adjusts the daily index of infections
by accounting for lower or higher inoculum pressure in the course of
an infection event.
Thirdly, daily rainfall and mean RH combinations were used to de-
termine head wetness duration between 30 and 48 h. Within 2 days,
infection events were recorded, and infection happens when precip-
itation (PREC) > 0.3 mm occurs in both days with average RH > 80%
for the 2 days or when PREC >0.3 mm with mean RH ≥80% followed
by a non-rainy day with mean RH ≥85%. An exponential equation to
calculate F. graminearum infection frequency (INF) under tempera-
ture effect (10°C to 30°C) for 48 h of head wetness was developed
as follows:
INF = 0.001029 exp(0.1957T)
In this case, T = average mean daily temperature in the 2-day win-
dow of the infection event.
Finally, four models calculating the daily infection risk index (GIB)
were developed. The daily GIB is then added and multiplied by 100
to get the accumulated infection index (% GIB). The GIBs are calcu-
lated using Equations (33)–(36).
GIB1 = ANT × INF
GIB2 = ANT × INF × GZ
GIB3 = ST × INF
GIB4 = ST × INF × GZ
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
where ANT is the daily mean proportion of anthers during a 2-day
infection event (IE), ranging from 0 to 1; ST = mean daily proportion
of susceptible tissue during IE; INF = infection frequency at the sec-
ond day of IE and GZ = mean spore cloud density during IE. The accu-
mulated infection index (% GIB) is the summation of partial infection
indices by the four different models along the susceptible period
(Equation 37):
GIB % = Σ(GIB × 100) (37)
The models were evaluated based on FHB incidence, FHB sever-
ity and FDK. Because the models estimate an infection index rather
than a disease level, regression was used to validate each model
using the coefficient of determination to verify the adequacy of the
index to explain the observations of the disease level (Del Ponte
et al., 2005). The coefficient of determination for the regression
analysis between predicted and actual infection ranged from 0.73 to
0.93, 0.43 to 0.69, and 0.14 to 0.37 for FHB severity, incidence and
FDK, respectively.
3.6 | Research on FHB prediction models
(31)
from Germany
Birr et al. (2019) developed three multiple regression models to
predict concentrations of DON and zearalenone (ZEA) mycotoxin
in wheat grain at harvest in a maize-free crop rotation in northern
Germany. The model for the highly susceptible cultivar (Model 1)
was based on data collected from eight farmers' fields from 2008
to 2017. The moderately to highly susceptible cultivar (Model 2) and
the low to moderately susceptible cultivar (Model 3) were based on
data from 2012 to 2016. The models used cumulative precipitation
and average temperature covariates and the interaction term of
precipitation and temperature during wheat flowering. In collabo-
ration with the Schleswig-Holstein Plant Protection Service (SCA),
the three models were adapted to farms using representative local
weather stations for weather-based DON and ZEA predictions as
part of the regional monitoring of leaf pathogens and FHB in wheat
in northern Germany.
(32)
Model 1: highly susceptible (Ritmo).
DON = 646 − 4.50P − 31.94T + 2.70PT (38)
ZEA = − 40 + 3.86P + 2.41T + 0.18PT (39)
Model 2: moderately to highly susceptible (Inspiration).
DON = 481 − 18.42P − 26.37T + 2.87PT (40)
(33)
ZEA = − 75 + 2.17P + 5.34T + 0.14PT (41)
(34)
Model 3: low to moderately susceptible (Dekan).
(35) DON = 305 − 24.09P − 18.74T + 2.73PT (42)
(36) ZEA = − 39 + 1.18P + 2.87T + 0.11PT (43)

MATENGU et al.
where P is the cumulative precipitation (mm) during wheat flowering
(GS61 to GS69), T is the average temperature (°C) during wheat flow-
ering (GS61 to GS69), PT is the interaction term of precipitation and
temperature. Model evaluation with independent data not used in the
development of the models but from similar sampling locations and
years produced satisfactory results (Birr et al., 2019). The coefficients
of determination for Models 1, 2 and 3 were 0.89, 0.91 and 0.86, re-
spectively, indicating that the variation in the predicted DON values ac-
counted for a substantial proportion of the observed variation in DON
values. In 95.2% of the cases, Model 1 correctly predicted whether
DON and ZEA concentrations were lower or higher than the European
limit of 1250 μg DON/kg and 100 μg ZEA/kg for both mycotoxins. The
accuracy of Models 2 and 3 were 85.7% (14.3% false negative) and
100%, respectively, for DON predictions when the 2017 dataset was
used (Birr et al., 2019). Models 2 and 3 accurately estimated in 100%
and 85.7% (14.3% false positive) of cases, ZEA concentrations below
or above the maximum level of 100 μg ZEA/kg, respectively, when the
2017 validation dataset was used (Birr et al., 2019).
In contrast to other published models (Franz et al., 2009; Hooker
et al., 2002; van der Fels-Klerx et al., 2010), all three models of
Birr et al. (2019) produced reliable DON and ZEA concentration
estimates in wheat grain without including post-anthesis weather
conditions. Therefore, the models are used not only as a tool for
identifying years, regions, or fields under risk, but also to assist
producers in determining whether to make a fungicide application
at flowering based on weather data from local weather stations in
order to reduce the risk of food and feed contamination with DON
and ZEA (Birr et al., 2019). The models use only temperature and
precipitation as input variables, which may not capture all factors
that affect disease development such as tillage practices and crop
rotation. Also, the models are based on linear relationships between
weather variables and DON and ZEA concentrations, which may not
hold true in all cases, affecting model accuracy. These models are
still relatively new, and further refinement may be needed to im-
prove their accuracy and usability.
3.7 | Research on FHB prediction models from the
Netherlands
In the Netherlands, the application of the DONcast regression equa-
tions developed by Hooker et al. (2002) to Dutch data resulted in
poor quantitative predictive performance (Franz et al., 2009). The
poor performance of the DONcast model was attributed to the fact
that predictive models for DON levels in wheat may not be univer-
sally applicable. Thus, two multiple regression models specific to the
Dutch conditions were developed to predict DON concentrations in
wheat grain in Dutch-specific situations (Franz et al., 2009). These
models use pre- and post-heading weather data, cultivar resistance,
region (REG) classified into south-west (SW), central (C) and north-
east (NE), fungicide use (SPRAY) classified into Yes or No, and head-
ing dates (HD) which include eight blocks of 6 days around the HD
(numbered 1 to 8 [as a suffix] to the weather variables as well as
|
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
9
for the entire pre- and post-heading time blocks of 24 days each)
(Franz et al., 2009). The heading date was estimated by linear re-
gression between the observed heading date and the pre-heading
temperature sum (Tsum) when not recorded. The first model focused
on average climatic variables in a 24-day pre-heading and a 24-day
post-heading period. The second model focused on average climatic
variables in eight-time blocks of 6 days. Models 1 and 2 explained
59% and 56% of DON variation across all fields, respectively, indi-
cating good predictive performance. Climate variables accounted for
30% of the overall variance in Model 1.
3.7.1 | Model 1
DON (𝜇g∕kg) = 3.771 + 0.209T (post − heading) + 0.005Pmm (pre − heading)
+ 0.004Pmm (post − heading) − 0.002Th25 (pre − heading)
− 0.005Th25 (post − heading) + 0.001RHh90 (post − heading)
− 0.031HD + REG + RES + SPRAY
(44)
where T (post-heading) is the average hourly post-heading tempera-
ture, Pmm (pre-heading) and Pmm (post-heading) are the total hourly
pre- and post-heading precipitation (mm), respectively, Th25 (pre-head-
ing) and Th25 (post-heading) are the number of hours with a tempera-
ture greater than 25°C and RHh90 is the number of post-heading hours
with RH > 90%. Non-climatic variables included HD (as factorial),
which are the heading dates; REG is the wheat-producing region with
coefficients: C REG (reference level); −0.366SW REG and 0.3442NE
REG. RES is the wheat resistance category with coefficients: RES 5
(reference level); −0.415RES 5.5; 0.057RES 6; 0.375RES 6.5; 0.482RES
7; 0.336RES 7.5; 0.525RES 8.5). SPRAY (factorial) is a fungicide ap-
plication with coefficients: No (reference level); −0.192Yes (Franz
et al., 2009).
3.7.2 | Model 2
DON (𝜇g∕kg) = 4.427 + 0.1320T (8) + 0.025RH (4) − 0.018Th25 (8)
− 0.034HD + REG + RES + SPRAY
(45)
where HD is heading date and T (8) is the average hourly tempera-
ture in the eighth of eight 6-day time blocks bracketing HD (i.e. 18
–24 days after heading), RH (4) is the average hourly relative humidity
in the fourth of eight 6-day time blocks bracketing HD (i.e., 0–6 days
before heading), and Th25 (8) is the number of hours with a T greater
than 25°C in the eighth of eight 6-day time blocks bracketing HD
(i.e., 18–24 days after heading). REG is the region where the wheat
is produced (factorial) with coefficients as C REG, reference level;
−0.385SW REG; 0.259NE REG, RES is wheat resistance (factorial) with
coefficients: −0.441RES 5.5; −0.037RES 6; −0.319RES 6.5; −0.508RES
7; −0.270RES 7.5; −0.618RES 8.5 and SPRAY (factorial) with coeffi-
cients No, reference level; −0.211Yes (Franz et al., 2009).
Both models predict concentrations of 10 log(x + 1) in DON
(μg/kg). Models 1 and 2 correctly predicted whether the DON

TA B L E 1 Risk prediction models developed worldwide for wheat and barley Fusarium head blight (FHB) and deoxynivalenol (DON).
Country Authors Modelling technique
USA De Wolf et al. (2003) Logistic regression
Shah et al. (2013, Artificial neural networks
2014) and decision trees within
logistic regression
Bondalapati Logistic regression
et al. (2012)
Canada Matengu (2022) Logistic regression
Hooker et al. (2002) Multiple regression
Argentina Moschini and Degree-day model for the
Fortugno (1996), critical infection period,
Moschini regression models for
et al. (2001) fusarium incidence and
severity
Italy Rossi et al. (2002, Regression models for
2003, 2003) epidemiological stages of
sporulation (rate, spore
dispersal), infection
frequency, disease
invasion rate
Brazil Del Ponte Second order polynomial Daily mean temp., daily mean RH,
et al. (2005) model for daily heading
and inoculum production,
non-linear regression for
infection frequency
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10
|
Input variables for days before [−]/
after [+] anthesis Response variables Model performance Limitations
Hours of temp. 15°C–30°C and FHB index (spring/winter Accuracy 70%–84% Low accuracy, does not consider
RH ≥90% [+10], hours of temp. wheat) (training data) agronomic practices
15°C–30°C [−7], hours of rainfall
[−7]
FHB genetic resistance mean RH [−7], FHB index (spring/winter Misclassification rate Low accuracy, does not consider
hours of temp. 9°C–30°C and RH wheat) 0.26 (training data), agronomic practices
≥90% [−7] 0.34 (test data)
Hours of RH ≥90% [±5] FHB index, DON (barley) 84% (training data) 79% Those above plus site-specific
(test data)
Hours of RH ≥80% [−14], mean daily FHB index (winter/spring 75%–85% accuracy Low accuracy, limited dataset
max. RH% [−14], hours of temp. wheat, durum, barley), (training data) with low number of epidemic
25°C–28°C [−14], mean daily FDK (spring wheat, 70%–100% accuracy samples, does not consider
rainfall (mm) [−14], hours of temp. durum), DON (durum) (test data) agronomic practices
15°C–30°C and RH ≥80% [−10],
mean daily RH% [−10], hours of RH
≥80% [−10], mean daily RH% [−14]
Days with rainfall >5 mm [−4 to −7 DON (wheat) R2 = 0.73 Poor performance reported in
heading], days with min. temp. some region-specific studies
<10°C [−4 to −7 heading], days
with rainfall >3 mm [+3 to +6
heading], days with max. temp.
>32°C [−4 to −7 heading], days
with rainfall >3 mm [+7 to +10
heading]
Number of 2-day periods of rainfall FHB incidence and severity R2 = 0.947 (moderately Site and year specific, limited to
≥0.2 mm and RH > 81% (1st day) (wheat) susceptible moderately susceptible and
and RH ≥78% (2nd day), max. and varieties), R2 = 0.907 susceptible varieties
min. temp. (susceptible varieties)
Spore incubation time, temp., total FHB risk of infection, and R2 = 0.51 (FHB) R2 = 0.90 Low accuracy for high FHB and
rainfall, rainy days (≥0.2 mm), non- DON (spring and winter (DON) DON cases
rainy days, hours of RH >80%, leaf wheat)
wetness
FHB infection index, related For the GIB4% model: Applicable only to the specific
consecutive rainy days to FHB severity, incidence R2 = 0.93 (FHB region studied
and FDK severity) R2 = 0.69
(FHB incidence),
R2 = 0.37 (FDK)
MATENGU et al.
 |

13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MATENGU et al. 11

concentration was lower or higher than the maximum level of

tillage practice on DON and

ZEA not directly included in

regions, requires validation

1250 μg/kg in 92% and 88% of the cases, respectively. The estimated

Developed for Dutch-specific

Effects of previous crop and

for use in other regions

DON amount increased in both models with increasing average tem-
perature, precipitation and RH but decreased with an increased
number of hours at temperatures above 25°C (Franz et al., 2009).

the models
There was an apparent regional effect on DON concentrations,
Limitations

with considerably greater levels when progressing from the south-

west to the north-east regions of the Netherlands that climatic and
non-climatic variables could not explain (Franz et al., 2009). Region
0.84 (model 2), 0.89
effects were thought to be caused by differences in infection pres-
0.86 (model 3) ZEA
R2 = 0.91 (model 1),

sure in different parts of the country because of how much wheat is

model 2: R2 = 0.56,
DON R = 0.89 (model
1), 0.91 (model 2),

sensitivity = 0.63;

sensitivity = 0.59
Model performance

grown. Wheat-after-wheat rotations are more typical in the north-

Model 1: R2 = 0.59,

east, with more intensive wheat cultivation. Another possible ex-

(model 3)

planation for the region effect were differences in temperature

2

sensitivity between populations of Fusarium species in different loca-

tions (Franz et al., 2009). Northern populations may have adapted to
lower average temperatures, resulting in increased growth, and divid-
zearalenone (ZEA) (winter

ing regional impact into underlying variables could greatly increase

model accuracy, making prediction models less site-specific and gen-
Deoxynivalenol (DON),

DON (winter wheat)

erally applicable to producers, industry and control authorities (Franz

Response variables

et al., 2009).
wheat)

4 | CO N C LU S I O N
total hourly pre- and post-heading

resistance, sprayed or not sprayed

Cumulative rainfall and average temp.

This paper reviews and summarizes several weather-based mod-

Average hourly post-heading temp.,

post-heading RH >90%, cultivar

Input variables for days before [−]/

rainfall, hours of pre- and post-

during wheat flowering period

heading temp >25°C, hours of

els for FHB disease forecasting in cereal crops that have been de-
veloped and used by a number of researchers in the past couple
of decades (Table 1). From the models reviewed in this study, it is
evident that there is a limited number that directly predict FDK
and DON levels. Moreover, the majority of these models focus on
after [+] anthesis

spring wheat, with only a few addressing barley and durum wheat.
This highlights a significant research gap and emphasizes the need
for the development of more comprehensive and diverse models
capable of accurately predicting FDK and DON levels in a variety
of cereal crops, including durum wheat and barley. While many of
Multiple regression models

Multiple regression models

the weather-based FHB models examined in this study, like the De

Wolf models in the United States, employ regression modelling,
Modelling technique

there are much more complex models, such as those developed

by Rossi for Italy and the Del Ponte model in Brazil, that simulate
the entire dynamic cycle of the FHB pathogen. The quest for iden-
tifying a crucial weather period conducive to FHB epidemics is
common in all models reviewed in this paper. The timeframe for
prediction of the FHB and the probability of outbreak varies from
Franz et al. (2009)

120 days before to 14 days after the commencement of anthesis.

Birr et al. (2019)

Most models produce more accurate predictions using weather pa-

rameters for periods within 14 days before and after anthesis. The
TA B L E 1 (Continued)

Authors

evidence for validity of these models is how well they link theoreti-
cal, scientific and small-scale knowledge to large-scale knowledge
of the disease, and this is assessed by different measures of ac-
Netherlands

curacy (R 2 and accuracy sensu stricto) (Hollingsworth et al., 2006).

Germany
Country

Parsimony, the principle of modelling that favours simplicity rather

than complexity, is preferable (Prandini et al., 2009; Rossi, Giosue,
 |
12
Pattori, & Del Vecchio, 2003). Of the models reviewed in this
study, Birr et al. (2019) and De Wolf et al. (2003) are examples of
the most parsimonious models, while Del Ponte et al. (2005) and
Rossi et al. (2002), Rossi, Giosue, Pattori, and Del Vecchio (2003),
Rossi, Giosue, and Delogu (2003) are considered less parsimonious
models, as they include more complex epidemiological factors that
account for disease development. Models must be as accurate as
possible while using the fewest possible easily measured variables.
While complex models can be more accurate and portable than
empirical models based on statistical relationships, complexity can
reduce model usability, adaptability and applicability. The long-term
accuracy of models may be impacted by effects of climate change
on FHB species evolution, which could lead to epidemic variation
and weather patterns outside those captured in the data used to
develop the reviewed models. In the future, it is likely that continu-
ous data collection for model development, adaptation and ongo-
ing validation will be essential (Vaughan et al., 2016). The recently
developed logistic regression risk models for FHB index (FHBi), FDK
and DON in spring wheat, winter wheat, durum wheat and barley
in western Canada (Matengu, 2022) are expected to be continu-
ously updated and validated in the future. The correlation between
these FHB damage indicators is variable, and the risk of infection
varies by crop type. These models power an online viewer that pro-
cesses near real-time weather information to generate FHBi, FDK
and DON estimates. Successful management of FHB will require
not just the development of FHB risk models but also collaboration
with local agriculture offices and agronomists as well as extension
staff to educate producers about the disease and the best use of
integrated management practices, including the risk assessments
that are provided by the models. Disease risk models should be in-
tegrated into disease management decisions using a multifaceted
approach. Web-based and smartphone app-based decision aid tools
enable widespread access to risk information. However, only a small
number of the models examined here, including the De Wolf mod-
els, the Shah models and DONcast, have been used to create a suc-
cessful commercial or public online tool for assessing Fusarium risk
(Table S1). Prediction of FHB days before anthesis is an essential
element of any FHB model if it is to be used as a tool for produc-
ers to decide on fungicide spraying. Models that use post-anthesis
weather conditions to evaluate the risk of FHB will not provide the
assessment in time for producers to intervene, although they can
still be helpful for anticipating FHB impacts on grain marketing and
food systems (De Wolf et al., 2003; Rossi, Giosue, Pattori, & Del
Vecchio, 2003).
AC K N O​W L E D
​ G E ​M E N T S
We are grateful for the financial assistance provided by the
Integrated Crop Agronomy Cluster, Western Grain Research
Foundation, Agriculture and Agri-Food Canada, Manitoba Crop
Alliance, Saskatchewan Wheat Development Commission, Alberta
Wheat Commission, Brewing and Malting Barley Research Institute,
Canadian Agricultural Partnership, and Prairie Oat Growers
Association.
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MATENGU et al.
C O N FL I C T O F I N T E R E S T S TAT E M E N T
The authors declare that there is no conflict of interest.
DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were cre-
ated or analysed in this study.
ORCID
Taurai T. Matengu https://orcid.org/0009-0003-3883-190X
W. G. Dilantha Fernando https://orcid.
org/0000-0002-2839-1539
REFERENCES
ACIS. (2021) Alberta Climate Service -Fusarium Head Blight Map.
Available from: https://​agric​ulture.​alber​t a.​c a/​acis/m#​!fusarium.
[Accessed 16th November 2021].
Bai, G. & Shaner, G. (2004) Management and resistance in wheat and
barley to Fusarium head blight. Annual Review of Phytopathology, 42,
135–161.
Birr, T., Verreet, J.-A. & Klink, H. (2019) Prediction of deoxynivalenol and
zearalenone in winter wheat grain in a maize-free crop rotation
based on cultivar susceptibility and meteorological factors. Journal
of Plant Diseases and Protection, 126, 13–27.
Bolanos-Carriel, C., Wegulo, S.N., Baenziger, P.S., Funnell-Harris, D.
& Hallen-Adams, H.E. (2020) Effects of fungicide chemical class,
fungicide application timing, and environment on Fusarium head
blight in winter wheat. European Journal of Plant Pathology, 158,
667–679.
Bondalapati, K.D., Stein, J.M., Neate, S.M., Halley, S.H., Osborne, L.E. &
Hollingsworth, C.R. (2012) Development of weather-based predic-
tive models for Fusarium head blight and deoxynivalenol accumula-
tion for spring malting barley. Plant Disease, 96, 673–680.
Cowger, C., Ward, T.J., Nilsson, K., Arellano, C., McCormick, S.P. &
Busman, M. (2020) Regional and field-specific differences in
Fusarium species and mycotoxins associated with blighted North
Carolina wheat. International Journal of Food Microbiology, 323,
108594.
Dahl, B. & Wilson, W.W. (2018) Risk premiums due to Fusarium head
blight (FHB) in wheat and barley. Agricultural Systems, 162, 145–153.
De Wolf, E.D., Madden, L.V. & Lipps, P.E. (2003) Risk assessment models
for wheat Fusarium head blight epidemics based on within-season
weather data. Phytopathology, 93, 428–435.
Del Ponte, E.M., Fernandes, J.M.C. & Pavan, W. (2005) A risk infection
simulation model for Fusarium head blight of wheat. Fitopatologia
Brasileira, 30, 634–642.
Del Ponte, E.M., Fernandes, J.M.C., Pavan, W. & Baethgen, W.E. (2009)
A model-based assessment of the impacts of climate variability on
Fusarium head blight seasonal risk in southern Brazil: modeling im-
pacts of climate variability on Fusarium head blight. Phytopathology,
157, 675–681.
Foroud, N.A., Baines, D., Gagkaeva, T.Y., Thakor, N. & Badea, A. (2019)
Trichothecenes in cereal grains—an update. Toxins, 11, 634.
Franz, E., Booij, K. & Van der Fels-Klerx, I. (2009) Prediction of deoxyni-
valenol content in Dutch winter wheat. Journal of Food Protection,
72, 2170–2177.
Gilbert, J. & Fernando, W.G.D. (2004) Epidemiology and biological con-
trol of Gibberella zeae/Fusarium graminearum. Canadian Journal of
Plant Pathology, 26, 464–472.
Giroux, M.-E., Bourgeois, G., Dion, Y., Rioux, S., Pageau, D., Zoghlami, S.
et al. (2016) Evaluation of forecasting models for Fusarium head
blight of wheat under growing conditions of Quebec, Canada. Plant
Disease, 100, 1192–1201.

MATENGU et al.
Góral, T., Wiśniewska, H., Ochodzki, P., Nielsen, L., Walentyn-Góral, D.
& Stępień, Ł. (2018) Relationship between Fusarium head blight,
kernel damage, concentration of Fusarium biomass, and Fusarium
toxins in grain of winter wheat inoculated with Fusarium culmorum.
Toxins, 11, 2.
Góral, T., Wiśniewska, H., Ochodzki, P., Twardawska, A. & Walentyn-
Góral, D. (2020) Resistance to fusarium head blight, kernel damage
and concentration of Fusarium mycotoxins in grain of winter triti-
cale (x Triticosecale Wittmack) lines. Agronomy, 11, 16.
Ha, X., Koopmann, B. & von Tiedemann, A. (2016) Wheat blast and
Fusarium head blight display contrasting interaction patterns on
ears of wheat genotypes differing in resistance. Phytopathology,
106, 270–281.
Harris, L.J., Balcerzak, M., Johnston, A., Schneiderman, D. & Ouellet, T.
(2016) Host-preferential Fusarium graminearum gene expression
during infection of wheat, barley, and maize. Fungal Biology, 120,
111–123.
Hollingsworth, C.R., Mewes, J.J., Motteberg, C.D. & Thompson, W.G.
(2006) Predictive accuracy of a Fusarium head blight epidemic risk
forecasting system deployed in Minnesota. Plant Health Progress,
7, 4.
Hooker, D.C., Schaafsma, A.W. & Tamburic-Ilincic, L. (2002) Using
weather variables pre- and post-heading to predict deoxynivalenol
content in winter wheat. Plant Disease, 86, 611–619.
Kazan, K., Gardiner, D.M. & Manners, J.M. (2012) On the trail of a cereal
killer: recent advances in Fusarium graminearum pathogenomics and
host resistance. Molecular Plant Pathology, 13, 399–413.
Kelly, A.C., Clear, R.M., O'Donnell, K., McCormick, S., Turkington, T.K.,
Tekauz, A. et al. (2015) Diversity of fusarium head blight popula-
tions and trichothecene toxin types reveals regional differences in
pathogen composition and temporal dynamics. Fungal Genetics and
Biology, 82, 22–31.
Landschoot, S., Waegeman, W., Audenaert, K., Van Damme, P.,
Vandepitte, J., De Baets, B. et al. (2013) A field-specific web tool for
the prediction of fusarium head blight and deoxynivalenol content
in Belgium. Computers and Electronics in Agriculture, 93, 140–148.
MARD. (2021) Fusarium head blight report-Manitoba agriculture and re-
source development. Province of Manitoba–agriculture. Available
from: https://​w ww.​gov.​mb.​c a/​agric​ulture/​. [accessed 16th
November 2021].
Matengu, T.T. (2022) Developing risk models to mitigate Fusarium head
blight in western Canadian cereal production. MSc thesis, University
of Manitoba. Available from: http://​hdl.​handle.​net/​1993/​36855​
[accessed 7th November 2023].
McMullen, M., Bergstrom, G., De Wolf, E., Dill-Macky, R., Hershman, D.,
Shaner, G. et al. (2012) A unified effort to fight an enemy of wheat
and barley: Fusarium head blight. Plant Disease, 96, 1712–1728.
Molineros, J. (2007) Understanding the challenge of Fusarium head blight
forecasting. PhD thesis, University Park, PA: Pennsylvania State
University.
Moschini, R.C. & Fortugno, C. (1996) Predicting wheat head blight inci-
dence using models based on meteorological factors in Pergamino,
Argentina. European Journal of Plant Pathology, 102, 211–218.
Moschini, R.C., Pioli, R., Carmona, M. & Sacchi, O. (2001) Empirical pre-
dictions of wheat head blight in the northern Argentinean pampas
region. Crop Science, 41, 1541–1545.
Musa, T., Hecker, A., Vogelgsang, S. & Forrer, H.R. (2007) Forecasting of
fusarium head blight and deoxynivalenol content in winter wheat
with FusaProg. EPPO Bulletin, 7, 283–289.
Nita, M. (2013) Fungicides–Showcases of integrated plant disease man-
agement from around the world. https://​doi.​org/​10.​5772/​3251
London: InTech.
Paul, P.A., Lipps, P.E. & Madden, L.V. (2005) Relationship be-
tween visual estimates of Fusarium head blight intensity and
|
13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
13
deoxynivalenol accumulation in harvested wheat grain: a me-
ta-analysis. Phytopathology, 95, 1225–1236.
Pitblado, R., Hooker, D.C., Nichols, I., Danford, R. & Schaafsma, A.W.
(2007) DONCAST: seven years of predicting DON in wheat on
a commercial scale. In: Proceedings of the National Fusarium Head
Blight Forum. The Westin crown center. Kansas City, Missouri.
Available from: https://​scabu​sa.​org/​pdfs/​forum​07_​proc_​compl​ete.​
pdf. [Accessed 6th May 2022]: ASAP Printing.
Prandini, A., Sigolo, S., Filippi, L., Battilani, P. & Piva, G. (2009) Review of
predictive models for Fusarium head blight and related mycotoxin
contamination in wheat. Food and Chemical Toxicology, 47, 927–931.
Rossi, V., Giosue, S. & Delogu, G. (2002) Dynamics of airborne of
Fusarium macroconidia in wheat fields naturally affected by head
blight. Journal of Plant Pathology, 84, 53–64.
Rossi, V., Giosue, S. & Delogu, G. (2003) A model estimating risk for
Fusarium mycotoxins in wheat kernels. Aspects of Applied Biology,
68, 229–234.
Rossi, V., Giosue, S., Pattori, E. & Del Vecchio, A. (2003) A model esti-
mating the risk of Fusarium head blight on wheat. EPPO Bulletin,
33, 421–425.
Schaafsma, A.W. & Hooker, D.C. (2007) Climatic models to predict occur-
rence of Fusarium toxins in wheat and maize. International Journal of
Food Microbiology, 119, 116–125.
Shah, D.A., De Wolf, E.D., Paul, P.A. & Madden, L.V. (2014) Predicting
Fusarium head blight epidemics with boosted regression trees.
Phytopathology, 104, 702–714.
Shah, D.A., De Wolf, E.D., Paul, P.A. & Madden, L.V. (2019) Functional
data analysis of weather variables linked to Fusarium head blight
epidemics in the United States. Phytopathology, 109, 96–110.
Shah, D.A., De Wolf, E.D., Paul, P.A. & Madden, L.V. (2021) Accuracy
in the prediction of disease epidemics when ensembling sim-
ple but highly correlated models. PLoS Computational Biology, 17,
e1008831.
Shah, D.A., Molineros, J.E., Paul, P.A., Willyerd, K.T., Madden, L.V. & De
Wolf, E.D. (2013) Predicting Fusarium head blight epidemics with
weather-driven pre- and post-anthesis logistic regression models.
Phytopathology, 103, 906–919.
Shah, D.A., Paul, P.A., De Wolf, E.D. & Madden, L.V. (2019) Predicting
plant disease epidemics from functionally represented weather se-
ries. Philosophical Transactions of the Royal Society B, 374, 20180273.
Spolti, P., Del Ponte, E.M., Dong, Y., Cummings, J.A. & Bergstrom, G.C.
(2014) Triazole sensitivity in a contemporary population of Fusarium
graminearum from New York wheat and competitiveness of a tebu-
conazole-resistant isolate. Plant Disease, 98, 607–613.
SWDC. (2021) Fusarium resources—Saskatchewan wheat development
commission. Sask wheat development commission. Available
from: https://​saskw​heat.​c a/​fusar​ium-​resou​rces [accessed 16th
November 2021]
Tamburic-Ilincic, L., Wragg, A. & Schaafsma, A. (2015) Mycotoxin accu-
mulation and Fusarium graminearum chemotype diversity in winter
wheat grown in southwestern Ontario. Canadian Journal of Plant
Science, 95, 931–938.
van Der Fels-Klerx, H.J., Burgers, S.L.G.E. & Booij, C.J.H. (2010)
Descriptive modelling to predict deoxynivalenol in winter wheat in
the Netherlands. Food Additives & Contaminants. Part A, Chemistry,
Analysis, Control, Exposure & Risk Assessment, 27, 636–643.
Vaughan, M., Backhouse, D. & Ponte, E.M.D. (2016) Climate change
impacts on the ecology of Fusarium graminearum species complex
and susceptibility of wheat to Fusarium head blight: a review. World
Mycotoxin Journal, 9, 685–700.
Wallhead, M. & Zhu, H. (2017) Decision support systems for plant disease
and insect management in commercial nurseries in the Midwest:
a perspective review. Journal of Environmental Horticulture, 35,
84–92.
 |

13653059, 0, Downloaded from https://bsppjournals.onlinelibrary.wiley.com/doi/10.1111/ppa.13839 by Cochrane Poland, Wiley Online Library on [26/01/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 MATENGU et al.

Yoshida, M., Nakajima, T., Tomimura, K., Suzuki, F., Arai, M. & Miyasaka,
A. (2012) Effect of the timing of fungicide application on fusarium How to cite this article: Matengu, T.T., Bullock, P.R.,
head blight and mycotoxin contamination in wheat. Plant Disease,
Mkhabela, M.S., Zvomuya, F., Henriquez, M.A., Ojo, E.R.
96, 845–851.
et al. (2023) Weather-based models for forecasting Fusarium
head blight risks in wheat and barley: A review. Plant
S U P P O R T I N G I N FO R M AT I O N
Pathology, 00, 1–14. Available from: https://doi.org/10.1111/
Additional supporting information can be found online in the
ppa.13839
Supporting Information section at the end of this article.