Empirical vs Mechanistic models for primary infections of

Empirical vs Mechanistic models for primary
infections of Plasmopara viticola
Rossi V., Caffi T.
[email protected]
Cossu A.
Fronteddu F.
Wageningen,
Wageningen, 17 October 2006
Grape Downy mildew is a disease of major importance
• several yield and quality losses
• fungicidal control is necessary
Introduction
Fungicide applications begin in late April
14-18 applications/year
Many epidemiological models have been
elaborated to better manage fungicides, with
emphasis for primary infections
Spring-summer
penetration
inoculation
Spring – early summer
dispersal
(splash)
invasion
dispersal
(air)
zoospores
zoosporangia
macrosporangia
sporulation
oogonium
Introduction
anteridium
oosphere
Winter
oospores
Late summer - autumn
3 10 rule
Voghera, Italy (1947)
Ètat Potentiel d’Infection model
Bordeaux, France (1983)
Empirical approach
Downy Mildew foreCast
New York, USA (1997)
Introduction
Università Cattolica Sacro Cuore model
Piacenza, Italy (2004)
Mechanistic approach
The aim of this work was to compare:
ü the different approaches in elaborating epidemiological models for
primary infections of P. viticola
ü their accuracy in simulating infections occurred in Sardinia (Italy),
1996 to 2004
Oospore
Formation
Zoospore
Maturation Germination Release
3 10 Rule
Model comparison
EPI model
Dispersal
Infection
Incubation
f(T, R, GS)
Ep f(T, R)
Ec f(T, RH)
f(T, R)
DMCast model
f(R)
UCSC model
f(T, VPD) f(T, VPD) f(T, LW)
f(T, R, GS)
f(R)
f(T, LW,
f(T, RH)
GS)
10 mm of rain in 24-48 hours
Average daily temperature of 10°C
Alarm provided by the model
Vine shoots length of 10 cm
End of incubation period
20
30
25
15
T (°C)
10
15
10
5
5
0
0
3 10 rule
15/4
22/4
29/4
6/5
13/5
Date
20/5
27/5
3/6
R (mm)
20
The model is based on the assumption of an ecological adaptation of
P. viticola to local climate
Differences between actual meteorological data and a climatic series
Ep = Potential energy calculated in the elaboration interval (10 days)
Ep
ct = 1,2 in October-November; 1 in December; 0,8 in January, February, March
HM = climatic monthly rainfall [mm]
θc = climatic monthly temperature [°C]
NGP = climatic number of rainy days per month
EPI model
t = average monthly temperature [°C]
H = monthly rainfall [mm]
h = rainfall in the decade [mm]
ngp = number of rainy days in the decade
Ec = Cinetic energy calculated in the elaboration interval (daily)
Ucn = climatic monthly nocturnal average of RH [%]
θc = climatic monthly temperature [°C]
Ud = average diurnal RH between 10.00 h and 18.00 h [%]
T = average daily temperature [°C]
EPI =
March
September
October
April
∑ Ep + ∑ Ec
Area of risk from April 1st, -10 < EPI < 0
EPI model
When the index increases of 3 in 3 consecutive days the model
provides an alarm of infection
Oospore maturation (OSP) is based on the hypothesis that both an
excess and a lack of rainfall have a negative effect on oospore
maturation (Tran Manh Sung et al., 1990)
These limits are defined by the climatic monthly average of rainfall
Hm = minimum treshold =
Climatic monthly rainfall
Climatic number of rainy days
HM = maximum treshold =
Climatic monthly rainfall + st. dev of climatic rainfall
Number of rainy days in the month
DMCast model
Then for a day d:
if
Hm < Rd ≤ HM
if
Rd < Hm
if
Rd > HM
then
then
then
POS(d)=Rd
POS(d)=HM
LAC(d)=Hm-Rd
EXC(d)=Rd-HM
Monthly maturation Index = ΣPOS(d) – |Σ
ΣEXC(d) – ΣLAC(d)|
LAC(d)
σ ⋅ √ 2π
25
0.30
20
0.25
0.20
15
0.15
10
0.10
5
0.05
0
0.00
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
Date
Primary infection occurs when:
DMCast model
cumulative oospore germination > 3%
T>11°C and R>2mm for oospore germination
T>11°C and R>0mm for infection
Probability density function
of oospore germination
P(d) =
1
- (d - µ)2
⋅ e 2 σ2
°C, mm
An oospore maturation index (calculated at the end of January) is used to
determine mean (µ) and standard deviation (σ) of the normal distribution
density function for mature oospores
OLL
t
MMR
MMO
T
DOR
PMO
LLM
VPD
R
rainfall
A new approach:
approach:
- pathosystem analysis
- information/data collection
- construction of mathematical relationships
inoculum
- dynamic simulation (timePrimary
step
of 1season
hr from
1st January)
RH
T
S
U
R
GER
RH
T
GEO
LW
LW
T
January February March
R
Zoospore
dispersal
ZLL
April
May
Zoospore
release
70
60
50
40
End of
incubation
LW
30
T
20
ZCI
10
RH
INC
T
29/7
22/7
15/7
8/7
1/7
24/6
17/6
10/6
3/6
27/5
20/5
13/5
6/5
29/4
22/4
15/4
8/4
1/4
0
ISS
R (mm)
Oospore
germination
ZGL
UCSC model
July
Infection
LW
INF
June
To compare model outputs:
- meteorological data
Data were supplied by the regional
network for the nearest automatic
station to the vineyard in Siniscola
(NU, Italy)
- field observations on disease onset
Models comparison
Untreated plots against downy mildew until first disease onset
At weekly intervals
Onset of
primary
symptoms
Yield losses
(%)
1996
06-12 May
99
1997
02-08 May
10
Year
Phenological
susceptibility
1998
30 Apr - 04 May
19-25 May
12
1999
02-04 May
11-18 May
8
2000
24-27 Apr
11-18 May
36
2001
15-21 Apr
15-22 May
0
2002
21-24 Apr
02-09 May
8
2003
01-02 May
-
0
2004
03-08 May
17-24 May
99
15/04
22/04
29/04
06/05
13/05
20/05
27/05
Results
High variability of disease severity and consequent yield losses
Less variability in time of primary disease onset
1996
UCSC
DMCast
EPI
3 10
Actual onset
1997
?
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
1999
UCSC
DMCast
EPI
3 10
Actual onset
?
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
?
2001
?
2002
?
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
2004
2003
?
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6 15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
Results
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
2000
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6 15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
UCSC
DMCast
EPI
3 10
Actual onset
1998
?
15/4 22/4 29/4 6/5 13/5 20/5 27/5 3/6 10/6 17/6 24/6
Output comparison of the four models
3 10
rule
EPI
model
DMCast
model
UCSC
model
Correct simulations
3
1
2
8
False
alarms
3
3
--
1
Missed
alarms
3
5
7
--
Uncorrect
simulations
Simulated infection actually
produced symptom appearance
in the vineyard
Simulated infection but the
disease did not appear at the end
of incubation period
The model did not simulate an
infection that actually occurred
Model simulations were stopped at the first disease outbreak
In 2001 DMCast model provided 1 unjustified alarm
Results
In 2004 UCSC model provided 1 unjustified alarms
3-10 rule did not predict the primary infection 3 times: primary infection involves
several processes which are regulated by different weather variables (too
simple!)
EPI model underestimated primary infections 5 times; in 2004 the epidemic was
very destructive: the simple comparison between climatic and current data is not
sufficient for explaining a complex phenomenon such a primary P. viticola
infection; moreover splash effect of rainfall is a crucial event in the infection
chain and it is not considered by the model
Conclusion
DMCast model failed downy mildew primary infection 7 times. It was due to the fact
that the probability density function of oospore germination starts too late: the
model calculates this function based on the environmental conditions of January.
Probably, in the area considered in this work, oospore maturation depends also on
the weather conditions occurring in other winter periods; therefore, this models
should need calibrations to be used under climatic conditions different from those
where it has been developed
UCSC model produced 1 unjustified alarm only during 2003 when disease was
not observed in field: the model was elaborated on data obtained under
controlled conditions for each step of the infection process and does not require
any calibration or correction to work under different environmental conditions
Empiric approach
• too simple
• difficult to use in different areas
• low accuracy outside the observed interval of independent
variables
Mechanistic approach
• more complex
• accurate dynamic simulations of the pathosystem
Conclusion
• high robustness without any calibration
Dank voor de
aandacht