.17 1..
5-,]... V .1

.1. , , . . V , H V . . I

. . . .. 2.5V .

.. :1 s. .
my

._......

 

 

,7 A . x AXLE.
.., ,5 .51 .33 51,.“ :t:

 

 

 

THESIS

”119%

This is to certify that the

dissertation entitled

SPATIAL AND TEMPORAL MODELING FOR RISK ASSESSMENT
AND MANAGEMENT OF POTATO LATE BLIGHT
(PH YTOPHTHORA INFESTANS)
IN MICHIGAN

presented by

Kathleen Marie Baker

has been accepted towards fulﬁllment
of the requirements for

Ph . D . degree in Geography

 

 

 

Date 21 MWL

MS U i: an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

 

 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/C|RC/DateDuo.pBS-p. 15

 

SPATIAL AND TEMPORAL MODELING FOR RISK ASSESSMENT AND
MANAGEMENT OF POTATO LATE BLIGHT (PHYTOPHTHORA INFESTANS)
1N MICHIGAN
By

_ Kathleen Marie Baker

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCI‘ OR OF PHILOSOPHY
Department of Geography

2002

ABSTRACT
SPATIAL AND TEMPORAL MODELING FOR RISK ASSESSMENT AND
MANAGEMENT OF POTATO LATE BLIGHT (PH YT OPHTHORA INFESTANS)
IN MICHIGAN
By
Kathleen Marie Baker

In the Midwestern United States, potato late blight (Phytophthora infestans) is a
temporally sporadic disease, occurring only when microclimate conditions within the
canopy are favorable and inoculum is present. Favorable environmental conditions
include air temperatures between 7 and 27 degrees C and relatively long periods (10
hours or more) of leaf wetness. Increasing concern in the agricultural community over
observed and projected climate change has prompted numerous studies on the possible
implications for crop yields. However, relatively little work has focused on disease
management.

Historical trends in hourly weather variables and potato late blight risk as
expressed by a modiﬁed Wallin Disease Severity Value index were analyzed at 7 stations
in Wisconsin, Michigan, and Ohio from 1948-1999. All Sites Showed Significant trends in
at least one of the risk estimates. While late blight risk was greatest at all stations in
August, periods of increasing risk occurred across the region particularly during July. The
increases in disease risk appeared to be associated with upward trends in dry bulb and
dew point temperature at nearly all of the stations, particularly during July and August.

These results correspond with historical climatological trends in the Upper Great Lakes

that point to warmer and wetter growing season conditions. Signiﬁcant trends in

precipitation totals and in the frequency of days with precipitation have also been
documented.

Because potato late blight Spreads rapidly, limited options are available to
growers once the crop canopy has become infected. Spatial and temporal dynamics of
the leading edge of a late blight field infection with respect to weather—based explanatory
vaiables were characterized using regression models. To improve management decisiOnS,
desiccation zone parameters were identiﬁed by exploring Spatial consistency of localized
late blight infection Spread in experimental plots both between replicates and between
years. Time in days Since inoculation was the variable most strongly related to distance
of lesion farthest from the point of initial inoculation. The best linear model that did not
include time in days used both the hours above 90 percent relative humidity and hours

between 70 and 80 percent relative humidity.

Copyright by
KATHLEEN MARIE BAKER

2002

Ever to my parents, who continue to impress upon me the importance

of dedication, Simplicity, love and vegetables.

ACKNOWLEDGEMENTS

Love and gratitude to everyone who smiled tolerantly at vegetable geography and
supported my constant department hopping, especially my committee: Drs. Jeff
Andresen, Willie Kirk, Jay Harman and Ray Hammerschmidt, and to the folks at MSU
GREEEN for funding the project.

To all of my potato boys (and the occasional itinerant spud babe), may there
always be someone to pull you out of the muck, may your plates never know
contamination, and may you never be without a pie when you really need one.

As always, extra Special thanks to: scum-of-the-earth-cat, who learned to speak
English; Laura, the strong-wristed, who was more than a Sister; Tom, who replenished
my stock of honey-roasted macadamia nuts; Ilya, who used the unusual theme of
landmines to inspire and admonish me; Gregory, who shared his cows and snuck in as
quietly as possible; and Grandma, who made some mean banana bread. To my circle of
friends and neighbors (who are often both at once!), I appreciate your smiles and support
so much. And, of course, to the girls who tried to burn down my garage — I’ll never forget
you.

In our lives and our research, let us follow the advice of Sergei Chesnokov and try
to reconcile ‘the gap between the truth of science and the truth embodied in the soul of

the artist’.

vi

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. ix

LIST OF FIGURES ............................................................................................................ xi
CHAPTER 1

INTRODUCTION ............................................................................................................... 1

Global and Local Potato Production ............................................................................ 2

Canopy Microclimate ................................................................................................... 6

Leaf Wetness .................................................................................................... 6

Leaf Wetness Sensors ...................................................................................... 7

Leaf Wetness Algorithms ................................................................................. 8

Potato Canopy Characteristics ....................................................................... 10

Potato Late Blight ...................................................................................................... ll

Potato Late Blight Models ......................................... 13

Research Questions .................................................................................................... 15

Objectives ................................................................................................................... 16
CHAPTER 2

HISTORICAL TREND ANALYSIS OF REGIONAL POTATO LATE

BLIGHT RISK .................................................................................................................. 18

Introduction ................................................................................................................ 1 8

Methods and Materials ............................................................................................... 19

Data Characteristics ....................................................................................... 19

Data Handling Methods ................................................................................. 21

Relative Humidity Thresholds .................................................................. 21

Hourly and Three-Hour Data .................................................................... 25

Trend Analysis ............................................................................................... 25

Seasonal DSV Accumulation .................................................................... 27

DSV Day Type Analysis ........................................................................... 27

Threshold Analysis .................................................................................... 29

Analysis of Highly Conducive Periods ..................................................... 29

Timing of Change in Risk ......................................................................... 30

Environmental Variable Trends ................................................................ 30

Results ........................................................................................................................ 31

Comparison of Hourly and Three-Hour Data ................................................ 31

Seasonal DSV Accumulation ......................................................................... 31

DSV Day Type Analysis ................................................................................ 37

Threshold Analysis ......................................................................................... 4]

Analysis of Highly Conducive Periods .......................................................... 43

Timing of Change in Risk .............................................................................. 47

Environmental Variable Trends ..................................................................... 49

Discussion .................................................................................................................. 52

vii

Conclusions ................................................................................................................ 56

CHAPTER 3
CHARACTERIZING THE LEADING EDGE OF A POTATO LATE BLIGHT
FIELD INFECTION .......................................................................................................... 58
Introduction ................................................................................................................ 58
Spatial Methods in Plant Pathology ............................................................... 59
Objectives ....................................................................................................... 61
Methods ...................................................................................................................... 62
Field Plots ...................................................................................................... 62
Regression Analysis ....................................................................................... 64
Application of Leading Edge Regression ...................................................... 70
Results ........................................................................................................................ 71
Time Dependent Variables ............................................................................ 71
Weather Variables .......................................................................................... 73
Disease Severity Value Variables .................................................................. 75
Combined Variables ....................................................................................... 77
Application of Regression Model .................................................................. 79
Discussion .................................................................................................................. 81
REFERENCES .................................................................................................................. 89

viii

LIST OF TABLES

Table 1. Potato canopy air temperature and relative humidity thresholds in the
computation of Wallin disease severity values for late blight risk estimation (W allin,
1962) .................................................................................................................................. 14

Table 2. Indicators of potato late blight risk estimated for each growing season and
location included in the 1948-1999 analyses and derived as secondary variables from
ambient air temperature, dew point temperature, and time in days during the growing
season (from May 1 through September 30) ..................................................................... 28

Table 3. Analysis of accumulated DSV per growing season, May 1 through September
30, at seven locations in the greater Michigan region from 1948-1999 and from 1948-
1990 including a) the median DSV accumulated per growing season and b) the rate of
change in DSV accumulated per growing season ............................................................. 36

Table 4. Analysis of speciﬁc DSV value day types per growing season, May 1 through
September 30, at seven locations in the greater Michigan region from 1948-1999: a)
median number of days with each value per growing season and b) rate of change in
number of days of each value per growing season ............................................................ 38

Table 5. Analysis of the accumulation of DSV to thresholds of 18 and 30 per growing
season, May 1 through September 30, at seven locations in the greater Michigan region
from 1948-1999 including a) the median DSV accumulated per growing season and b)
the rate of change in DSV accumulated per growing season. ........................................... 42

Table 6. Analysis of the number of ﬁve day periods with an accumulation of ten or more
DSV per growing season, May 1 through September 30, at seven locations in the greater
Michigan region from 1948-1999 and from 1948-1990 including a) the median DSV
accumulated per growing season and b) the rate of change in DSV accumulated per
growing season. ................................................................................................................. 46

Table 7. Analysis of DSV accumulation during 30-day periods per growing season, May
1 through September 30, at seven locations in the greater Michigan region from 1948-
1999 including: a)the median DSV accumulation (2v) during 30-day periods per growing
season and b) rate of change in DSV accumulation during 30-day periods per growing
season. ............................................................................................................................... 48

Table 8. Analysis of rate of change per growing season, May 1 through September 30, at
seven locations in the greater Michigan region from 1948-1999 for: a) dew point
temperature and b) air temperature. .................................................................................. 51

Table 9. Environmental variables and indicators of potato late blight risk derived as

secondary variables from hourly ambient air temperature and dew point temperature from
day of initial inoculation to day of assessment .................................................................. 69

.ix

Table 10. Linear regression results for time and weather independent variable models.
Dummy slope and intercept variables were included to test significant differences
between the models created from two randomly chosen data sets (A and B). Dummy
variables were identiﬁed as signiﬁcant (<0.0001) or not signiﬁcantly different (A=B)... 74

Table 11. Linear regression results for DSV independent variable models. Dummy slope
and intercept variables were included to test signiﬁcant differences between the models
created from two randomly chosen data sets (A and B). Dummy variables were identiﬁed
S signiﬁcant (<0.0001) or not Signiﬁcantly different (A=B). ............................................ 76

Table 12. Linear regression results for time, weather and DSV combined independent
variable models. Dummy Slope and intercept variables were included to test signiﬁcance
between the two halves of the data set (A and B). Dummy variables were identified as
Signiﬁcant (<0.0001) or not Signiﬁcantly different (A=B) ............................................... 78

LIST OF FIGURES

Figure 1. Value of state potato production in the coterrninous United States (National
Potato Council, 2000), a. Production value in thousands of dollars. b. Production value
Standardized by area in dollars per thousand hectares. ....................................................... 4

Figure 2. Potato acreage by Michigan county (Michigan Agricultural Statistics, 1999-
2000; MPIC 2001; unpublished MSU Extension data) ....................................................... 5

Figure 3. National Weather Service (NWS) stations used in historical trend analysis 20

Figure 4. Comparison of in-canopy and ambient air relative humidity from a potato
canopy in Entrican, MI during the growing seasons of 1997, 1998, and 1999. Mean and
standard deviation of ambient relative humidity are Shown for each one percent interval
of canOpy air relative humidity (approximately 125 ambient values per canopy
percentage). Thel :1 relationship line is Shown for clarity, and the 90% canopy relative
humidity is shown as the common criteria for estimating leaf wetness. ........................... 23

Figure 5. Disease severity value accumulations for 1997, 1998, 1999 at Entrican,
Michigan. DSV were based on relative humidity thresholds of 90 percent in the canopy
and both 80 and 90 percent in the ambient air. DSV were accumulated in days after
potato emergence and billing, when sensors were placed in the canopy (approximately
May 15-30). ....................................................................................................................... 24

Figure 6 (a-g). DSV accumulation (2v) by year for: a) Sault Ste Marie (Y62), b) Alpena
(APN), c) Traverse City (TVC), (1) Green Bay (GRB), e) Muskegon (MKG), f) Grand
Rapids (GR), and g) Toledo (TOL). Values are overlaid with location-speciﬁc median
value line and smoothed trend line. ................................................................................... 32

Figure 7. Accumulated potato late blight DSV (2v) from May 1 to September 30 at
seven locations in the Upper Great Lakes region, locations described in Figure 6.
Summary statistics, from top to bottom of each boxplot, include maximum (above
maximum point), 75% quartile (right of upper whisker), median (box midline), 25%
quartile (right of lower whisker) and minimum values (below minimum). ...................... 34

Figure 8 (a-g). Number of days per growing season of each DSV value (l,2,3,4) stacked
to show proportion of overall annual DSV accumulation for: a) Sault Ste Marie (Y62),

b) Alpena (APN), c) Traverse City (TVC), (1) Green Bay (GRB), e) Muskegon (MKG), f)
Grand Rapids (GR), and g) Toledo (TOL). Total growing season days were 153 in this
analysis; DSV=O each year was 153-Evy”. ...................................................................... 39

Figure 9 (a—g). Number of five-day intervals with disease severity value accumulation of
ten or more per growing season for: a) Sault Ste Marie (Y62), b) Alpena (APN), c)
Traverse City (TVC), (1) Green Bay (GRB), e) Muskegon (MKG), 0 Grand Rapids
(GR), and g) Toledo (TOL). Note change of scale in 9g.Years when the total number of

xi

high accumulation periods was zero are shown with a black box at the origin, while years
without a bar or origin point were not included in the analysis. ....................................... 44

Figure 10. Mean 30-day DSV accumulation with starting dates at 15-day intervals
throughout the growing season. Locations where the increase in 30-day DSV
accumulation during 1948-1999 was Signiﬁcantly greater than zero at the p=0.05 level

are circled. Associated 30-day accumulation rates of change and signiﬁcantly different
were extracted from Table 6. Estimates of potato canopy leaf area index (LAI) for
northern growing regions (e. g. latitude 42N+) are included for each 30-day period (Allen
and Scott, 1992). LAI = 3 is considered full canopy ........................................................ 50

Figure 11. Example of along and across-row distance components using an example
quadrant from the analysis. Instead of estimating a circular disease expansion radius in
Euclidean distance from the inoculation (1') point (6.4 for this quadrant), the leading edge
expansion model uses the maximum distance from the inoculation point of both an along-
row and across-row component (5, in both cases). ........................................................... 67

Figure 12. Regression results for lesion distal to inoculation (LD) and: a) days after
inoculation, b) hours since inoculation with relative humidity greater than 90 percent, and
c) days Since inoculation with DSV > 0. ........................................................................... 72

Figure 13. Observed values and predicted linear regression line. Distance from initial
inoculation point in meters is predicted using days after inoculation ............................... 80

Figure 14. All data from sets A and B and the recommended desiccation zone in meters
both along and across-row from the inoculation point in days after inoculation. ............. 82

Figure 15. The increase in the area of predicted observable disease, the predicted
observable disease, and the total predicted disease. Late blight expansion becomes
visible in the field approximately 8 days after inoculation. .............................................. 83

Figure 16. Observable disease from an example 2001 research plot after 13 days after
inoculation with the model predicted observable disease, the predicted observable disease
99% conﬁdence bound, and the total predicted diseased area with 99% conﬁdence
(recommended desiccation zone). ..................................................................................... 84

xii

CHAPTER 1

INTRODUCTION

Late blight of potato, caused by Phytophthora infestans (Mont. De Bary), has
recently re-emerged as a major factor in potato (Solonum tuberosum) production in North
America (Fry & Goodwin, 1997). Potato late blight is the most important potato disease
throughout the potato growing regions of the world (Inglis et al., 1996) and is considered
one of the most important plant diseases relative to global food production in the future
(Agrios, 1997). Prior to 1992, North American populations of P. infestans (A1 mating
type) were easily controlled by metalaxyl-based fungicide applications. A new aggressive
genotype of P. infestans (U88, A2 mating type, metalaxyl-insensitive) has replaced the
previous clonal lineage and the foliage of all currently available commercial varieties is
highly susceptible to current populations (Inglis et al., 1996; Douches et al., 1997).

No fungicides currently in use completely arrest epidemic development, and even
plants protected with five-day fungicide applications at a maximum rate of active
ingredient became infected when environmental conditions were conducive to late blight
development and inoculum was plentiful (Kirk et al., 2000; Mayton et al., 2001). Even
plants protected with ﬁve-day fungicide applications at a maximum rate of active
ingredient become infected when environmental conditions are conducive to late blight
develoPment and inoculum is plentiful (Kirk et al., 2000). In Michigan, the Michigan
Potato Industry Commission Pest Survey estimated that 7 - 18 protective applications
were made during a typical growing season depending on length of growing season and

end use of crop. These fungicide applications reduce the profitability of the industry and

could have attracted criticism from environmental groups seeking to reduce reliance on
pesticides (MPIC, 2000).

The increasing economic impact of potato late blight, in terms of both yield loss
and increased fungicide use, makes understanding disease dynamics in space and time
important for estimation of risk. Since epidemics occur only when environmental
conditions are conducive and inoculum is present in the potato canopy, potato late blight
is a suitable disease for geographically—based epidemiological research at a wide range of
Spatial and temporal scales. Two geographically-based research topics, varying in scale,
follow, including: a) the quantiﬁcation of historical trends in late blight conducive
climate conditions; and b) the characterization of the spatial and temporal dynamics of an
epidemic within a single ﬁeld to enhance the effectiveness of grower late blight
management techniques after a canopy has become infected.

The following sections focus on the background of potato production, the canopy
microclimate conditions necessary for foliar infection by late blight, and the estimation

procedure for late blight risk. Speciﬁc research questions and objectives follow.

GLOBAL AND LOCAL POTATO PRODUCTION

The potato is generally considered to be the fourth most important global food
crop behind rice, wheat and maize, and is grown throughout the world in a wider range of
latitudes than any other agricultural crop (Hijmans et al., 2000). About seven percent of
the total global potato production (NPC, 2001) occurs in the United States; close to

550,000 hectares were grown annually in the US. between 1959 and 1998. During this

same time, however, average yield nearly doubled from 3 to 6 metric tons per hectare
(NPC, 1999). Major potato growing regions of the US. include the Northwest,lUpper
Midwest, Northeast and Southeast. The States of Washington and Idaho led in revenue
from potato production during 2000 (Figure la). When production value is standardized
by state area (dollars per 1000 hectares), North Dakota, Wisconsin, Michigan, Maine,
Delaware and Florida are also important producers (Figure lb).

In Michigan, production in the last several years has remained relatively steady at
680,000 metric tons which is just over three percent of the nation’s production (NPC,
I999; MASS, 1999-2000). Approximately 80 percent of that production, as of 2000, was
dedicated to varieties particularly used for chipping and the snack-food industry. In
terms of farmgate value, the state’s crop is worth nearly 90 million dollars annually
(MPIC, 2000). Michigan’s production by county is Shown in Figure 2. Montcalm County
has consistently grown twice the area of any other Michigan county, reporting 5,300
harvested hectares in 1999. Production zones in Michigan are linked to soil texture, with
most potatoes in the state grown in well-drained loamy-sand and sandy-loam soils. A
high percentage of the state’s production is located in the central lower Peninsula, but the
southwest and northern Lower Peninsula and the central Upper Peninsula also have

substantial potato production.

 

    

- 300000 - 600000

 

   

Production, by area
0

 

 

 

1:1
I: 0 - 500
500-1500
-1500-3000
- 3000-7000
" 500 0 500 1000 1500 Kilometers
+ I I I I

 

 

Figure 1. Value of State potato production in the coterminous United
States (National Potato Council, 2000), a. Production value in
thousands of dollars. b. Production value standardized by area in
dollars per thousand hectares.

 

 

 

 

 

 

 

Hectares

     

 

 

 

E 0
[j 1 - 200
m 200 — 500
500 - 2500
- 2500 - 5500
+ 100 0 100 200 Kilometers
| I l A

 

 

Figure 2. Potato acreage by Michigan county (Michigan
Agricultural Statistics, 1999-2000; MPIC 2001; unpublished
MSU Extension data).

 

CANOPY MICROCLIMATE

Leaf Wetness

Canopy wetness and air temperature affect the initiation and rate of late blight
spread once inoculum is present in the canopy. The wetness period or wetness duration of
a leaf generally refers to the length of time that liquid water is present on the leaf surface.
As an agricultural variable, leaf wetness has many beneﬁcial aspects, such as temperature
regulation, soil water conservation and leaf turgor enhancement, which aids in
photosynthesis (Armstrong et al., 1993). In contrast to these positive effects, leaf wetness
duration is the most important factor for a large number of crop diseases including potato
late blight (Fry et al., 1997; Wilks et al., 1991); apple scab (Penrose and Nicol, 1996);
bacterial spot of peach; (Zehr and Shepard, 1996); Septoria of wheat; Rhynchosporium
of barley; Athracnose of watermelon (Monroe et al., 1997); and white mold disease of
dry beans (Deshpande et al., 1995).

Water on the leaf surface may be the result of a precipitation event, or may be
deposited by dew or fog. Most potato crops in Michigan also receive regular irrigation.
Depending on the type of wetting event and the characteristics of the leaf (e. g. wax on
cuticle, hair density), the water on the surface may be in the form of droplets, or may
cover the leaf in a thin ﬁlm (Thompson, 1981). In some studies, a microscopic ﬁlm of
water was still detected on grass blades an hour after moisture was no longer visible

(Giesler et al., 1996).

Meteorologically, the likelihood and magnitude of infection is correlated strongly
with the duration of wetting period and temperature. Fungi and bacteria generally
produce secondary inoculum in increasing quantities as duration of wetting period
increases within pathogen speciﬁc temperature ranges and when the host is in a
vulnerable condition, e. g. immature leaf tissue (Kirk, personal communication). The
relationship between wetting period and disease has been assessed with various models

and sensors.

Leaf Wetness Sensors

The most frequently used leaf wetness sensors are made with electrical resistance
grids. Equally spaced ﬁngers of copper or other conductive metal form an alternating
circuit over the surface of a ﬁberglass board and surface moisture is detected by changes
in the electrical resistance of the grid (Gillespie and Kidd, 1978). Color, orientation
modiﬁcations and coverings of absorbent cloth have been tested to more accurately
Simulate leaf surface conditions, but these changes can cause the measured wetness
period to vary by about three hours (Armstrong et al., 1993; Gillespie et al., 1993). In
low biomass canopies, such as turfgrass, surface grids do not closely approximate the leaf
surface. Instead of resistance grids, small stainless steel pins can be inserted directly into
blades of grass. This type of sensor is thought to be sensitive to microscopic water ﬁlms
(Giesler et al., 1996). A beta ray gauge is perhaps the most accurate measure of leaf

wetness, but is also costly and sensitive to inclement weather (Armstrong et al., 1993).

The duration of leaf wetness has been Shown to be exceedingly variable, even at
multiple locations in close proximity to one another within the same field (Andresen and
Edson, 2001). Comparisons of different devices for leaf wetness measurement showed
that while most sensors accurately modeled shape of the leaf’s drying curve , the estimate
of leaf wetness duration was inaccurate (Armstrong et al., 1993). However, it is duration
of leaf wetness that is critical to pathogen models (Thompson, 1981). Penrose and Nicol
(Penrose and Nicol, 1996) advocate the use of at least three wetness sensors per Site to

improve reliability for disease warning estimates.

Leaf Wetness Algorithms

As sensors vary in method and accuracy, so do models. Mathematical models for
estimating leaf wetness differ in both the number of variables used and the complexity of
the relationships among those variables. Perhaps the most complete, as well as one of the
earliest algorithms was based on the energy balance equation for a surface, and
subdivides the canopy into layers approximately 10 cm in depth (Thompson, 1981). For
each layer, resistances and energy balance parameters were calculated. Other models
have approximated leaf wetness using more easily measurable meteorological variables.
Little standardization exists among these models due to the wide range of disciplines in
which they originated.

Both climatological and agronomic variables inﬂuence duration of leaf wetness.
Net radiation, wind speed, vapor pressure and soil moisture are critical climatological

components. Although net radiation is difﬁcult to estimate accurately given the number

of variables involved in the surface energy budget and the influence of cloud cover, it is
essential for estimating the quantity of energy available for evaporation. Ambient vapor
pressure and vapor pressure gradients affect the rate and direction of movement of
moisture from the leaf surface. Soil moisture and wind Speeds also influence the vapor
pressure gradient and movement of moisture to and from the canopy, although some
studies have speciﬁed that leaf movement does not Signiﬁcantly inﬂuence drying time,
(Armstrong et al., 1993). Canopy structure and density, and leaf orientation, temperature,
and surface properties influence interception of precipitation and radiation, as well as
aerodynamic resistance.

Due to the difficulty in monitoring or estimating leaf wetness, it has often been
approximated using the so-called “90 percent criterion”. If relative humidity within the
canopy remains above 90 percent, leaves in that canopy are considered wet. Relative
humidity is a temperature-dependent proxy for vapor pressure, as a proportion of the
saturation vapor pressure of air at the same temperature. Because saturation vapor
pressure increases with temperature, relative humidity decreases when vapor pressure
remains constant while the air warms. Thompson (1981) found that the 90 percent
criterion often underestimated leaf wetness duration, especially on days when dew, as
opposed to precipitation, was the primary source of canopy moisture. This was
especially true for ground crops, while the criterion may have been a good approximation
for tree crOpS as there is air movement through the canopy, and the canopy is often less
dense. The 90 percent criterion was not considered the best leaf wetness estimate for
disease alert systems and other applications of leaf wetness, especially in ground crop

situations where dew is an integral aspect of wetness duration (Gleason et al., 1994).

Both leaf wetness sensor values and relative humidity approximations are used to
quantify moisture in the growing environment for a number of agricultural purposes.
Harvest moisture forecasts have been implemented in these types of approximations in
wheat and barley (Atzema, 1998), rice (Lu and Siebenmorgen, 1994), cut grass (Atzema,
1998), hay (Barr and Brown, 1995), and alfalfa (Wilks et al., 1993). Predictive
epidemiological models have been extensively developed for potato early and late blight
(Wallin, 1962; Nutter and MacHardy, 1980; Shtienberg and Fry, 1990a; Shtienberg and
Fry, 1990b; Doster and Fry, 1991; Raposo et al., 1993; Johnson, 1998) to optimize the
initiation and effectiveness of fungicide applications. Similar strategies have been
developed for tomatoes (Gillespie et al., 1993) and Botrytis in both greenhouse
vegetables (Shtienberg and Elad, 1997) and onions (Vincelli and Lorbeer, 1989), as well

as for Sclerotinia stem rot in Oilseed rape (Twengstrom et al., 1998).

Potato Canopy Characteristics

Under non-limiting conditions of water, nutrients and solar radiation, potato
plants increase leaf area very quickly and create a microclimate favorable to late blight
(Dorozhkin, 1971; Harris, 1992). Leaf wetness in potato canopies must include auxiliary
branches and stems as well as laminar tissue since P. infestans can affect all tissue that
makes up the potato canopy. The canopy is generally considered closed when the ratio of
total leaf area to ground surface area, or leaf area index (LAI), is equal to 3 (Jeffries and
Heilbronn, 1995). However, the canopy continues to increase in density, reaching

maximum LAI’s upwards of 8, depending on growing conditions, approximately two

10

months after planting (Allen and Scott, 1992). The increasing density of the potato
canopy increases the duration of leaf wetness by capturing a large percentage of incident
radiation but limiting the amount of radiation reaching lower leaves that could evaporate
water. Water vapor pressure also remains higher than that of the ambient air as the
canopy limits airflow, and shading by upper branches protects lower canopy levels from
high temperatures (Thompson, 1981). The importance of leaf wetness to potato late blight
and the suitability of the physical environment within the potato canopy to extending leaf
wetness duration together create a situation so conducive to disease that potatoes have

been considered ‘suicidal’ (Dorozhkin, 1971).

POTATO LATE BLIGHT

As potato production is global, the re-emergence of late blight has become a
worldwide problem (Fry and Goodwin, 1997). The infection of the potato plant occurs
only after extended periods of leaf wetness at relatively cool temperatures, although
infection can occur even at air temperatures in excess of 30°C (Kirk, personal
communication). Late blight is readily transmitted by seed-bome inoculum and,
consequently stems and leaves may be exposed to late blight from infected seed pieces
and in-ﬁeld volunteer plants (Lambert et al., 1998).

In the canopy, Sporangia can be transported by wind or splash from plant to plant
during precipitation or irrigation events. Symptoms appear in the canopy within 5-8 days
of inoculation and include necrotic lesions on leaves and stems that appear gray and

water soaked (Lacy and Hammerschmidt, 1995).

ll

The influence of canopy air temperature on the development and growth of P.
infestans has been well documented (Harrison, 1992). Although the optimal temperature
for hyphal growth in foliage varies among potato cultivars, it is typically between 15 and
20°C (Harrison and Lowe, 1989; Harrison, 1992). P. infestans is an obligate pathogen
and the base temperature for late blight development may be related to that for potato leaf
and sprout development, i.e., 4°C (Kirk et al., 1985). A decreased rate of Sporangia]

development occurs at temperatures below the optimal temperature of 10°C (Kadish and
Cohen, 1992).

Many potato late blight protection strategies depend upon high level inputs of
agrochemicals. Estimation of ‘current’ ﬁeld conditions that influence late blight
development enables growers to apply appropriate fungicides in response. There are
currently several types of fungicide applied routinely for prevention of late blight in
Michigan and several others with more speciﬁc activity that could be used when
conditions are more favorable for late blight development or when infection has occurred.
In order to effectively advise growers on the appropriate late blight control strategy it is
necessary to have information based on knowledge of fungicide efﬁcacy, foliar
development, varietal susceptibility to late blight, intended market, and past and current
weather records.

The control of potato late blight and estimation of associated ﬁeld conditions
throughout the growing season have continued to be top research priorities for the
Michigan potato industry since 1995 (MPIC, 2000). Because conditions are only
sporadically favorable and inoculum is not always present, timely and effective

management recommendations are crucial to the economics of potato production. The

12

Upper Great Lakes region is important to the study of late blight field epidemic
characteristics as the disease occurs more sporadically than in other US. production areas
because of the variability of growing season weather conditions. For this reason, 90
percent of Michigan growers incorporate some form of weather monitoring into their

disease management practices (MPIC, 2000).

Potato Late Blight Models

 

The 90 percent relative humidity criterion has been used for potato late blight
monitoring Since early analyses of canopy based records of the physical environment in
relation to disease development (W allin, 1962). Average air temperatures and duration
of wetting period are combined in the daily assignment of Wallin disease severity values,
which have become standard in estimating potato late blight risk (MacKenzie, 1981).
Microclimate-based late blight models were established originally in the 1940’s and
fungicide spray programs and other management decisions were based upon the
accumulation of ‘blight favorable dayS’, or thresholds of the disease severity values
(DSV) beginning in the 1960’s (Beaumont, 1947;Wa11in and Schuster, 1960; Wallin,
1962; MacKenzie, 1981). DSV range from 0 to 4 and are calculated using the average air
temperature during extended periods of high relative humidity (Table l). The severity of
late blight risk is estimated by both the number of hours above the relative humidity
threshold and the air temperature during that time. Longer leaf wetness durations,
approximated by 90 percent relative humidity, and higher air temperatures between the

thresholds of 7.2 to 270°C are indicative of greater late blight risk (Wallin, 1962). The

13

traditional Blightcast model developed during predominance of the US] genotype (Al
mating type, metalaxyl-sensitive) recommended initiation of fungicide applications after
the accumulation of either ten consecutive favorable days or 18 DSV following

emergence (M acKenzie, 1981).

Table 1. Potato canopy air temperature and relative humidity thresholds in the
computation of Wallin disease severity values for late blight risk estimation (Wallin,
1962).

Severity values for the number of hours relative humidity > 90%

 

Ave Temp 0 (none) 1 (minimal) 2 (slight) 3 (moderate) 4 (severe)
7.2-11.7 <16 hours 16-18 hours 19-21 hours 22-24 hours n/a

11.7-15.0 <13 hours 13-15 hours 16—18 hours 19-21 hours 22-24 hours
15.0-27.0 <10 hours 10-12 hours 13-15 hours 16-18 hours 19-21 hours

Michigan State University (MSU) has been using a modified Wallin model for
making management recommendations to growers for several years (Baker et al., 2000).
The model was adapted for use with hourly temperature and relative humidity data. Each
of the hourly values above the relative humidity threshold are assigned to a particular
Wallin-based temperature category and totaled for the day. Relative humidity thresholds
are set at 80 percent, instead of the traditional 90 percent criterion, to account for the
underestimation of the 90 percent threshold especially with respect to the canopy
environment. Justification for using an 80 percent relative humidity threshold is
described in Chapter 2.

Spatial spread of potato late blight within a field of potatoes by continuous leaf

and stem infection is rapid and limited options are available to growers once the crop

l4

canopy has become infected. Making appropriate management recommendations in such
a situation is, therefore, difﬁcult as symptoms develop latently, up to 7 days after initial
infection and spread of mycelium through the foliar tissue. The importance of weather
characteristics during the growing season and associated risk estimates are
unquestionably linked to development of late blight epidemics, but the spatial and
temporal dynamics of infection have not been quantiﬁed. Neither has the relationship
between those dynamics been understood in relation to weather-based explanatory

variables.

RESEARCH QUESTIONS

This research focuses on understanding the influence of weather and climate on
potato late blight dynamics in Michigan. Models are developed to further understand the
Spatial Spread of late blight and the increase in magnitude of foliar infection in a ﬁeld
situation with respect to physical, weather-based characteristics of the canopy. In
addition, historical analysis quantiﬁes trends in the effects of these weather variables over
the long term throughout the Michigan region. The primary research question explored
is: In what ways can potato late blight epidemic dynamics and risk be better understood,
and recommendations improved, through spatio-temporal modeling of ﬁeld infections
and historical weather data analysis? The main research focus is expanded more
specifically at various spatial and temporal scales, as follows:

a) historical assessment of late blight risk in the greater Michigan region

15

> Historically, how much has potato late blight risk varied spatially and temporally

in the greater Michigan region?

> What are the trends in the last 50 years in late blight severity value
accumulations?
b) epidemiology of late blight spread within a field during an epidemic
> When and where does the leading edge of late blight expand following
introduction of initial inoculum?

> What is the influence of weather on the spread of a late blight within a ﬁeld?

OBJECTIVES

> Assess the impact of location on late blight risk within Michigan and analyze trends
at these locations to determine how risk has changed over time.

> Create an algorithm for estimating the location of the leading edge of late blight
including conﬁdence bounds.

> Characterize the Spread of potato late blight in a confined (research plot) area as a

function of weather related explanatory variables.

This dissertation consists of three chapters. Following this introduction, the second
chapter analyzes historical weather conditions in the Upper Great Lakes region and the
implications of trends for the severity of potato late blight epidemics in space and time.

The third chapter describes the movement of the leading edge of infection following

16

inoculation and quantifies the relationship between this movement and explanatory

variables involving time and weather conditions.

17

CHAPTER 2
HISTORICAL TREND ANALYSIS OF REGIONAL

POTATO LATE BLIGHT RISK
INTRODUCTION

Potato late blight is a temporally sporadic disease that occurs only when
microclimate conditions within the canopy are favorable and inoculum is present (Lacy
and Hammerschmidt, 1995). Leaf wetness duration and in-canopy relative humidity are
critical variables in determining the relative risk of late blight development. AS a result,
changes in meteorological variables throughout the growing season that inﬂuence the
amount of in-canopy moisture and vapor pressure could significantly impact subsequent
disease pressure. Precipitation and dew point temperature are two variables of this nature
that can be easily quantified.

This chapter describes the identification of weather-based trends that have
inﬂuenced potato late blight disease risk in the Upper Great Lakes region of the US.
Only the inﬂuence of climate on disease risk as calculated by modified Wallin disease
severity values (Wallin, 1962) is considered, regardless of irrigation, other cultural
practices, or pathogen biotype changes that may have impacted late blight risk. This
historical perspective for potato late blight risk characterizes temporal trends in the
greater Michigan region from 1948-1999. The analyses also explore spatial similarities in
trends by statistically comparing the magnitude and timing of late blight risk between

locations.

18

METHODS AND MATERIALS
Data Characteristics

Historical air and dew point temperatures were extracted from the National
Climatic Data Center’s (NCDC) Surface Airways data-set for seven ﬁrst order National
Weather Service (NW S) stations in the greater Michigan region (Alpena - APN, Grand
Rapids - GRR, Green Bay - GRB, Muskegon - MKG, Toledo - TOL, Traverse City -
TVC, and Sault Ste Marie - Y62) (EarthInfo, 1948-1999). Throughout this chapter,
locations are listed in tables and ﬁgures in order of latitude, from northernmost to
southernmost. For each location, Shown in Figure 3, years with missing values for more
than seven days of the growing season (May 1 to September 30) were not used. Station
record lengths ranged from 35 to 49 years for the 52-year period from 1948-1999. The
initial year in the period of record for each station was as follows: Y62, 1948; APN,
1964; TVC, 1949; GRB, 1950; MKG, 1948; GRR, 1964; TOL, 1955.

Possible discontinuities in the data series were of concern because of the length of
the series and the fact that instrument updates, site modiﬁcations and location changes
occurred at each of the stations independently. The magnitude of change until 1990, as
identiﬁed in previous studies is low, especially in view of the 20°C temperature window
used in this study to derive secondary variables (Robinson, 2000). For example,
Robinson (2000) found that pre-l990 changes altered temperature measurements less
than 1°C. Therefore, the data series was judged to be appropriate for the Study. Of

continued concern, however, is the gradual shift beginning in the 1990’s to the

19

 

 

 

 

50 0 50100150 km
1:21:32

 

Figure 3. National Weather Service (NWS) stations used in
historical trend analysis.

20

Automated Surface Observing System (ASOS) instrumentation at airports to standardize

observational methodology, as the impact on data continuity after this change is unclear.

Data Handling Methods

Relative Humidity Thresholds

Potato late blight disease severity values (DSV) were calculated each day, May 1
through September 30, at each location every year. DSV were based on the Wallin
method (as described in Chapter 1) as modified by MSU Late Blight Lab and described
in Baker et al. (2000) and below. Relative humidity was obtained with hourly air and
dew point temperatures. Each hour above 80 percent relative humidity was assigned to its
respective Wallin-based temperature category and category hours were summed for each
day. Hourly values were considered conducive for late blight if the associated air
temperature ranged from 7.2 to 27°C. Hours that were both above the relative humidity
threshold and within temperature ranges from 7.20 —11.69, 11.70-14.99, and 15.00-27.00
for a requisite number of hours (detailed explanation in Table l) were assigned the
corresponding DSV. DSV range from 0 to 4, where 0 were not conducive, and 4 were
highly conducive to late blight development. 24-hour daily periods were set at 12:00
noon through 11:59 am, to maximize single day estimation of duration of the leaf
wetness period.

Although a 90 percent relative humidity ambient air threshold is often used as the

best approximation of leaf wetness, an 80 percent relative humidity threshold was used in

21

this study for the following reasons. Given the dense structure of a potato canopy,
relative humidity is typically higher in the canopy than the ambient air. Figure 4 shows
the relationship between canopy and ambient relative humidity. Hourly relative humidity
data were recorded at Entrican, MI during the growing season in 1997, 1998 and 1999
from the time of canopy closure (LAI=3) to beginning of canopy senescence (LAI=2).
Mean and standard deviation of the ambient relative humidity per one percent intervals of
canopy relative humidity are Shown with respect to the 90 percent criterion for estimating
leaf wetness and the line of 1:1 relationship between canopy and ambient relative
humidity is included for clarity. Approximately 125 ambient relative humidity values
were included for each canOpy interval. Ambient percent relative humidity values were
about 5-7 percent lower than those recorded in the canopy air, with standard deviations
ranging from 3-5 percent. At 80 percent ambient relative humidity, the canopy relative
humidity was within one standard deviation of 90 percent and at 83 percent ambient
relative humidity, the canopy mean relative humidity was 90 percent. Given these results,
a large percentage of hours when the canopy relative humidity was near 90 percent would
not be counted if the ambient threshold for DSV accumulation was greater than 80
percent.

Comparison of disease severity accumulations in the canopy and ambient air at
different relative humidity thresholds also indicate that the 90 percent ambient relative
humidity threshold is insufﬁcient for estimating canopy conditions (Figure 5). The 90
percent criterion has been shown in previous studies to underestimate leaf wetness
duration in ground mom and to be less than optimal for disease alert systems (Thompson,

1981; Giesler et al., 1996). The 80 percent ambient threshold, however, consistently

22

 

100 - -.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

:2 -
E
3
I
“>’
:: 90‘
0)
O:
E
(D
9.
it”
> 80-:
D.
O
C
(U
0
704

 

 

 

l l I I I

70 75 80 85 90 95

Ambient Percent Relative Humidity

 

0 Mean ambient relative humidity per one percent interval
of canopy relative humidity
— Line of 1:1 relationship between canopy and ambient relative humiditL

 

 

 

Figure 4. Comparison of in-canopy and ambient air relative humidity from
a potato canopy in Entrican, MI during the growing seasons of 1997, 1998,
and 1999. Mean and standard deviation of ambient relative humidity are
shown for each one percent interval of canopy air relative humidity
(approximately 125 ambient values per canopy percentage). Thel :1
relationship line is shown for clarity, and the 90% canopy relative humidity
is Shown as the common criteria for estimating leaf wetness.

23

Accumulated DSV’s

 

100
1997

80"

60‘ can!

40‘ o 9'

20‘ W
04

 

 

 

80

 

 

 

 

 

 

0 20 40 60 100
50
1 998 .
o
40 « o '
M v
M v
30 ‘ o M
we
I'm,
20 ‘ 0..
V 0
fr..- '"w °
' U—
10 4 . v' m
o o'
0 20 40 60 80 100
100
1 999
80 -l 'd.’

60‘

40-

20'I

 

 

 

 

0 20 40 60

Days in Accumulation Period

Figure 5. Disease severity value accumulations
for 1997, 1998, 1999 at Entrican, Michigan.
DSV were based on relative humidity thresholds
of 90 percent in the canopy and both 80 and 90

100

 

 

O

V

80% RH - Ambient
90% RH - Ambient
90% RH - Canopy

 

percent in the ambient air. DSV were accumulated in days after potato
emergence and billing, when sensors were placed in the canopy

(approximately May 15-30).

'24

 

overestimates the 90 percent canopy threshold in terms of disease risk. Given the
closeness of disease severity value accumulations between 80 percent ambient and 90
percent canopy on day 100 in 1998, and day 51 in 1999, a lesser ambient threshold would
be prone to occasional but serious underestimation in potato late blight risk, especially
during periods of strongly increasing DSV accumulation. Therefore, an 80 percent
ambient air relative humidity threshold was used for the calculation of potato late blight

disease severity values throughout this analysis.

Hourly and Three-Hour Data

Archived NWS weather data alternated between hourly and three-hour values.
Three-hour values were most frequently used from 1965-1972, and from 1978 to 1981.
When hourly values were not available, three-hour values were used to estimate hourly
values with linear interpolation. Hourly ambient air temperature values, dew point
temperature, and relative humidity hourly values were compared with those interpolated
from three-hour values and analyzed with respect to the timing and magnitude of

differences.
Trend Amalysis
Non-parametric statistical methods were appropriate for trend analysis since the

underlying distributions of the data were generally unknown. These methods are also not

affected by missing values and data errors that resulted from measurement changes and

25

missing hourly values that may have impacted estimation of correlation and slope
parameters. The possibility of a large number of outliers in the risk indicators derived as
secondary variables from ambient and dew point temperature also made non-parametric
statistics a better choice than Similar parametric estimates. Trends in the timing and
accumulation of disease severity values were quantiﬁed using a non-parametric slope
estimator (Sen, 1968). Unlike linear regression, the trend magnitude statistic (B) is not
signiﬁcantly impacted by lack of normality, missing data values and errors that are
unavoidable in derivative data. This method was used in similar studies of hydrological
and climatological time series trend analyses (Andresen et al., 2001). When n is the

number of observation in a series, B is calculated as:

B = median{D,.j} Equation 1

where for all possible pairs (x,~ ,Xj), 1 _< i < j sn, D,-,- is the difference in a risk associated

value through time when:

._ (x). _ xi)
j—i
and i and j are years included in the analysis for a single location while x is the risk

D

ij

Equation 2

related estimate for that speciﬁc year.

The probability of detecting a statistically signiﬁcant difference between risk
indicators, derived as secondary variables from weather data, was estimated at the p=0.05
signiﬁcance level using Kendall’s tau b non-parametric correlation coefﬁcient (SAS
statistical software package). Both the magnitude of the correlation coefﬁcient and the
number of years (n) included for a given location were used in the determination of

Si gniﬁcance. Kruskall-Wallis one way analysis of variance on ranks was also used at the

26

 

 

Ill

=0.05 level to test for statistically significant differences between potato late blight risk
indicator estimates at various locations. Trend analysis was performed using time series
of the DSV derived variables for each growing season from 1948-1999 at each location.

These risk indicators (Table 2) are described in detail below.
Seasonal DS V Accumulation

The growing season included the 153 days from May 1 through September 30 for
each year. The seasonal accumulation of DSV, 2331’ or 2v , is the sum of all 153
individual values when v is the individual disease severity value from 0 to 4 of days
within a Single growing season. Trend analysis was performed on both the 253v time
series including all available years, and on a temporal subset that excluded years (1') after

153
1990 ( 20 V199,”). This subset limited the study to pre-ASOS instrumentation in order

to assume the 1°C maximum series bias found by Robinson (Robinson, 2000).

DS V Day Type Analysis

Different disease severity values (0,1,2,3,4) indicated different late blight risk

related conditions, or a different type of disease risk day. When x, is a growing season
153

day with a speciﬁc DSV, y, 0 xy (or 2129) represents the sum of days of this type in a

particular growing season, e. g. Xx). represents x days in a particular growing season with

disease severity value of y. The number of days with each speciﬁc DSV value, 2x). and

the number of days with a value greater than zero, 21x,” , were compared to determine if

27

Table 2. Indicators of potato late blight risk estimated for each growing season and
location included in the 1948-1999 analyses and derived as secondary variables from
ambient air temperature, dew point temperature, and time in days during the growing
season (from May 1 through September 30).

 

 

Derived Simplified Description
Variables
v the individual DSVl per day within a Single growing
season2
x a growing season day with a particular DSV assigned
y the DSV assigned to a particular growing season day
2153‘) 2v number DSV accumulated per growing season2
0
Z 153v Evjslggo number DSV accumulated per growing season for growing
0 151990 seasons prior to the year 1991, where j is the year
2153 x 2x, total number of days with a particular DSV in a single
0 1' growing season
t t2v=18 time in days (I) until the growing season accumulation of
(X ”'8’ DSV is equal to 18
t Isl-=30 time in days (I) until the growing season accumulation of
(2 ”‘30) DSV is equal to 30
,M number of ﬁve day periods (tint-+5) in a single growing
2 , V 2 10 2v210 season when the accumulated DSV (for those ﬁve days) is

equal to at least 10

2 Enddate v 30-day 2v number DSV accumulated during a particular period
Standate speciﬁed by start date and end date during a single growing
season

 

'Potato late blight disease severity values (DSV) were calculated using the modified
Wallin method explained in this chapter.

2 Growing season included May 1 through September 30 (153 days) in all years ( 1949 -
1999) in analyses.

28

trends were related to increasing frequency of a specific type of risk condition. The
distribution of Expo and the change in 2x, distribution were also compared at each station

location.

Threshold Analysis

The time, t, in days until the accumulation of DSV, 2v, reached a threshold value,
was calculated for fungicide Spray triggers at 18 and 30 accumulated DSV. The 18 DSV
05:18) accumulation corresponded to the traditional Wallin model threshold initiation of
fungicide sprays(Wallin and Schuster, 1960; Wallin, 1962; MacKenzie, 1981), while 30
03:30) was used in the MSU modiﬁed system as a key 2v that signaled a point to
increase fungicide application rate from minimum to maximum manufacturer’s
recommended application rate. In the MSU system, applications of fungicides were
recommended to start soon after emergence regardless of conditions, or iv, in order to

prevent initiation of epidemics from late blight contaminated seed.
Analysis of Highly Conducive Periods
Consecutive rain days and other relatively short time periods with large increases
DSV accumulation are especially important for management decisions related to late

blight control. The number of times during each growing season when the accumulated

late blight risk, v, during any ﬁve day time period, t,;_,-+ 5, was greater than 10 ( 2:5'1! 2 10

29

or Zv210) was used to estimate the frequency of canopy conditions that were highly

conducive to late blight.
Timing of Change in Risk

To determine the timing in the changes in potato late blight risk accumulation

over the 50 year period, trends In 30- day accumulated values ( 2% v) were assessed

Standute
to determine which periods within the growing season were most inﬂuenced by changing
weather patterns. Since no previous research has established an optimal time scale within
the growing season for late blight risk, or deﬁned the beginning and end of risk periods
seasonally, DSV accumulations, v, were started on the ﬁrst and fifteenth of each month
(2::::lv ,szjvv "2:?“ to ensure complete coverage of the entire growing season.
These 30-day periods were then associated with potato canopy leaf area index (LAI)
estimates for the growing region (Allen and Scott, 1992). LAI monthly estimates were

important to the understanding of how canopy conditions relate to those in the ambient

air during analyzed time intervals.
Environmental Variable Trends
Similarly, to determine relationships between trends in late blight risk values and

the physical environmental variables from which they were derived, monthly dew point

temperature(Td) and ambient temperature (Ta) were also analyzed for the presence of

2mm], "2.80130 T
trends ( Ma)l ’ . Sept]

30

RESULTS

Comparison of Hourly and Three-Hour Data

Differences between hourly and interpolated values in air and dew point
temperature and associated relative humidity percentages were greatest at sunrise and
sunset, as would be expected given the Sinusoidal nature of the diurnal temperature curve,
and the linear nature of the interpolation method. In 1985 at Grand Rapids, for example,
the greatest difference in both ambient temperatures and relative humidity occurred just
after sunrise and before sunset, or 8:00 and 21 :00 respectively, and Standard deviations
were greatest from 21:00 to 8:00. Temperature differences were less than 1°C and the
mean difference in relative humidity values was about 1' 1-2 percent with a mean standard

deviation 3.8 percent.

Seasonal DSV Accumulation

Median value lines were overlaid on time series of annual accumulations at each
location (Figure 6(a-g)). Across the region, DSV accumulations, Ev, tended to be greater
in 1960-62, 1978-79 and 1994-96. Boxplots and accompanying summary statistics were
used to visualize the distribution of iv at each location during the growing seasons from
1948-1999 (Figure 7). The greatest 2v ranges, as well as the highest maximum Xv

values, were found at Muskegon, Grand Rapids and Toledo. Alpena, Traverse City, and

31

 

18° a. Y62

 

 

 

 

 

 

 

 

DSV Accumulation

 

 

 

 

 

180
150 - C. TVC

140 -

120‘“ ‘
100- .

.
80" ‘ A A ‘
60"
40“ . 5 ‘ A.

20q . A A

 

 

 

 

1950 1960 1970 1980 1990 2000

Year

Figure 6 (a-g). DSV accumulation (2v) by year for: a) Sault Ste Marie (Y62),
b) Alpena (APN), c) Traverse City (TV C), (1) Green Bay (GRB), e) Muskegon
(MKG), f) Grand Rapids (GR), and g) Toledo (TOL). Values are overlaid
with location-speciﬁc median value line and smoothed trend line.

32

DSV Accumulation

 

 

180
160'-
140'-
120.. A
100-* -
80-‘ g
60- A g

40'-
20-*

 

‘5

 

 

 

180
160-‘
140'-
120'-

 

 

 

 

 

 

 

180
160-‘
140 -
120-1
100"
80

60 -
40“
20 r

f. GFIR

 

 

 

 

 

 

180

 

160‘-
140‘-
120'-
100-‘
80'-
60"
40 a
20“
0 I I I

 

 

 

 

33

2000

DSV Accumulation

 

180

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

159
154
160 ‘ 151 O
O
O O
140 -
O
118 120 .
120 - , o '
107 108 9
—- 107
0 8 103 100
100 - . o ' 97 - 95
92 .
O
81 _c_ 80 82
80 - '
72 73 76
. 66 .
64 64 67 67
60 -
f 51 o 7"
51 54 __
I9 - = -
40 a , o
o 38 o 39
0 I
20 4 I . I 28 27
23
21 :
14
O I I I I T I I
Y62 APN TVC GRB MKG GRR TOL
NW S Station Locations

Figure 7. Accumulated potato late blight DSV (2v) from May 1

to September 30 at seven locations in the Upper Great Lakes region,
locations described in Figure 6. Summary statistics, from top to bottom
of each boxplot, include maximum (above maximum point), 75%
quartile (right of upper whisker), median (box midline), 25% quartile
(right of lower whisker) and minimum values (below minimum).

34

 

Sault Ste Marie were less variable, and had lower Xv values than the other stations. Green
Bay had a Similar mean iv to the southern Stations, but fewer outliers.

The results from the analysis of 2v, including the median DSV accumulated and
the rate of change in iv for both the total data set and the pre-ASOS temporal subset are
shown in Table 3. The median iv for the entire 52-year period ranged between 51 at
Traverse City to 90 at Toledo. Alpena and Sault Ste Marie were not statistically different
from Traverse City. Green Bay, Muskegon and Grand Rapids were not signiﬁcantly
different from Toledo. However, the three northernmost locations were signiﬁcantly
different from the four southernmost locations. All 2v were found to increase between
1948 and 1999 and the increases were signiﬁcantly greater than zero at all sites except
Toledo and Traverse City. Increases in 2v at Sault Ste Marie, Green Bay, Muskegon and
Toledo were of similar magnitude, between 0.54 and 0.61 DSV per year from 1948-1999.
The magnitude of the increases in 2v at Alpena and Grand Rapids, 1.00 and 1.10
respectively, was nearly double those at the other locations.

When years after 1990 were excluded from the analysis, similarities and
differences between locations became more complex. The median 2:31-51“) at Green
Bay and Toledo were greater than stations with a median lower than 65, but not
signiﬁcantly different from Muskegon, Grand Rapid, or one another. The median values
of 2Vj51990 at Muskegon and Grand Rapids were not signiﬁcantly different from that at
Sault Ste Marie. Grand Rapids and Traverse City had median values signiﬁcantly
different from one another. Alpena and Sault Ste Marie median 2Vj51990 were not
signiﬁcantly different from those at Grand Rapids and Traverse City. The estimated rates

of increase over time remained significantly greater than zero after the removal of post-

35

Table 3. Analysis of accumulated DSV per growing season, May 1 through September
30, at seven locations in the greater Michigan region from 1948-1999 and from 1948-

1990 including a) the median DSV accumulated per growing season and b) the rate of
change in DSV accumulated per growing season.

 

a) Median number DSV accumulated per growing season1

(25%)

 

 

1948-1999 1948-1990
NW5 stationz N Median N Median
Y62 47 64.0 b3 38 63.5 bcd
APN 35 64.0 b 26 63.5 cd
TVC 49 51.0 b 40 49.0 d
GRB 36 92.5 a 27 89.0 a
MKG 45 80.0 a 36 76.5 ab
GRR 36 82.0 a 27 80.0 abc
TOL 41 95.0 a 32 93.5 a

 

b) N on-paramentric trend (rate of change in v per year) for
DSV accumulated per growing season

 

 

NW S station 1948-1999 1948-1990
Y62 0.55 *4 0.23
APN 1.00 *** 1.12 *
TVC 0.25 0.10
GRB 0.56 ** 0.43
MKG 0.54 * 0.54
GRR 1.10 * 2.21 *
TOL 0.61 0.86

 

‘ Growing season included May 1 through September 30 (153 days).

2National Weather Service (NWS) station locations: Y62 - Sault Ste Marie; APN -
Alpena; TVC - Traverse City; GRB - Green Bay, WI; MKG - Muskegon, MI; GRR -
Grand Rapids, MI; TOL - Toledo, OH.

3 Within a single column, values followed by the same letter are not Signiﬁcantly
different at P = 0.05 (Kruskal—Wallis One Way Analysis of Variance on Ranks).

4 Rate of change is significantly greater than zero at P = 0.05 (*), P = 0.01 (**), or P =
0.001 (***) (Kendall Tau b).

36

1990 data at only Alpena and Grand Rapids. In comparison to the analysis of iv, the rate
of change in XVI-gem increased at Toledo, remained constant at Muskegon, and decreased

at all other locations.

DSV Day Type Analysis

Median values and trend statistics for the number of different disease risk day
types each growing season are shown in Table 4. The number of days with a DSV of 0,

153

0 x0

, at Alpena and Traverse City were significantly different from the four
southernmost locations, but not from one another or from Sault Ste Marie. The median
value at Sault Ste Marie was not signiﬁcantly different from that of Grand Rapids.
2:310, decreased at all locations over the 50-years in the region. This decrease was
statistically significantly at Sault Ste Marie, Alpena, Green Bay and Muskegon.

The number of days that the DSV was greater than zero (Expo ) are shown in
Figure 8 (a-g) for all NW S locations, broken down by Speciﬁc values of y (2xy ). The
median 2.x, were more frequent at all stations than Ex), 2x2, or 2x3 (Table 4). 232,213 and
2x4 were less frequent at all locations, and increased in range as station latitude
decreased.

The trends in the number of days with each Expo were not signiﬁcantly different
between one another at the four southernmost stations (Table 4). Neither were Expo
signiﬁcantly different from one another at the three northern stations. 2x, at Sault Ste

Marie, Traverse City, and Alpena were not Si gniﬁcantly different from those at Grand

Rapids. 2x; at Sault Ste Marie, Alpena, Muskegon and Grand Rapids were not

37

Table 4. Analysis of speciﬁc DSV value day types per growing season, May 1 through
September 30, at seven locations in the greater MiChigan region from 1948-1999: a)
median number of days with each value per growing season and b) rate of change in
number of days of each value per growing season.

a) Median number of days per growing season1 with speciﬁc DSV value (y)
153

from0-4(20 xy)

 

 

 

 

 

 

EXEDZ Z £53150 2 (1)53 x, (1)53 x2 (1)53 x3 :3 x4

Y62 122.0 ab3 12.0 be 9.0 be 5.0 ab 4.0 bc

APN 123.0 a 12.0 c 8.0 bc 5.0 ab 5.0 abc

TVC 125.0 a 12.0 be 7.0 c 3.0 b 4.0 c

GRB 109.0 c 18.0 a 13.0 a 6.5 a 7.0 a

MKG 113.0 c 15.0 ab 11.0 ab 6.0 a 5.0 ab

GRR 115.0 be 14.5 abc 11.0 ab 6.5 a 5.0 abc

TOL 106.0 c 18.0 a 14.0 a 7 .0 a 5.0 ab
b) Non-paramentric trend (rate of change in x per year) in number of days
per growing season with Speciﬁc DSV value (y) from 0 through 4

3x3“ 2 (1)53 x0 2 (1)53 x1 2 L53 x2 (1)53 x3 53 x4

Y62 -0.22 *4 0.03 0.08 * 0.02 0.04 *

APN -0.44 ** 0.14 0.16 ** 0.11 ** 0.07

TVC -0.10 0.00 0.00 0.30 0.00

GRB -0. 19 * 0.00 0.09 * 0.06 * * 0.03

MKG -O.30 ** 0.09 0.09 * 0.04 0.04

GRR -0.48 0.15 0.22 ** 0.00 0.13 *

TOL -0.25 0.06 0.13 * 0.00 0.08

‘ Growing season included May 1 through September 30 (153 days) in all years (1949 -

1999) in analyses.

2National Weather Service (NWS) station locations: Y62 - Sault Ste Marie; APN -
Alpena; TVC - Traverse City; GRB - Green Bay, WI; MKG - Muskegon, MI; GRR -
Grand Rapids, MI; TOL - Toledo, OH.

3Within a single column, values followed by the same letter are not signiﬁcantly
different at P = 0.05 (Kruskal-Wallis One Way Analysis of Variance on Ranks).

4 Rate of change is signiﬁcantly greater than zero at P = 0.05 (*) or P = 0.01 (**)
(Kendall Tau b).

38

DSV Accumulation

 

80

 

 

 

 

 

 

a.Y62
60-
40-1 I I I II II
I | -. lgi‘ 1:! |.|‘~5
1" ulli it .1. ‘5 1| "1 1111:
20‘ --1'-“"‘" *""‘-""" “1'1 *' H 1- 111B
ii; ”1111'”'1i,!1 .. 1“ El 111
0..
8° b APN
60"
40-
I | ? I I! II III: II
:1 g I
201 .l -1 I'ilIHII “ll-1111
. 'Hi 1 LI! illl'iIH II III 13*
0 I I E
80
c.TVC
60"
40-
"iilri i- I I |--§|*'
I I "' I l . a ’!1’ t . g:
20- ‘33.: .t- . __ n i I 11- "I
5.53:; Edi :i “a” 11_1!g (”I i_ 11 .
- v - i 3 , 1
DJ
1950 1960 1970 1980 1990
. Year
Figure 8 (a-g). Number of days per growmg season of each
DSV value (l,2,3,4) stacked to Show proportion of overall
annual DSV accumulation for: a) Sault Ste Marie (Y62), - DSV =1
b) Alpena (APN), c) Traverse City (T VC), (1) Green Bay ‘ DSV =2
(GRB), e) Muskegon (MKG), 0 Grand Rapids (GR), [:1 DSV =3
and g) Toledo (TOL). Total growing season days were m DSV =4

153 in this analysis; DSV=0 each year was 153-Expo.

39

 

 

 

2000

 

 

 

 

 

 

 

d. GRB
e. MKG

 

 

 

 

f. GRR

 

 

 

80
80

1990 2000

1980

Year

1970
40

1960

g. TOL

1950

 

 

 

co=m_:E:oo< >mo

60 -
40 -

significantly different from one another. 209 at Sault Ste Marie and Alpena were not
signiﬁcantly different from the four southern stations. At Muskegon, Grand Rapids,
Toledo, Sault Ste Marie, and Alpena, 2x4 were not statistically different, but Traverse
City was significantly different from Green Bay, Muskegon and Toledo.

While 2x0 decreased from 1948 to 1999 at every location, temporal trends for
every location and every Expo were either positive or zero. The decrease in 2x0 was
statistically different at Sault Ste Marie, Alpena, Green Bay and Muskegon. No location
showed a statistically Significant change in Ex] , but all locations except Traverse City
had a signiﬁcant increase in in. Alpena and Grand Rapids had the greatest rate of
change per year in Ex, and 21:2. Alpena and Green Bay had Significant increases in 2x3
and Grand Rapids and Sault Ste Marie had a signiﬁcant increases in 2x4 over the 50 year

time period.
Threshold Analysis

The number of days from the beginning of the growing season until the
accumulation of 18 DSV (tzvzlg) and 30 DSV (tgvz3o) is Shown in Table 5. Traditional
Wallin models initiated fungicide sprays at t$x=13 , and the MSU model used tzx=3o as a
point to increase fungicide application rate. For both sz=13 and tzx=3o the number of days
was greatest at Traverse City and Traverse City was signiﬁcantly different from the four
southernmost stations, but not from Sault Ste Marie or Alpena. Sault Ste Marie and
Alpena [Ev-:18 were not Si gniﬁcantly different from Muskegon lzv=13. Alpena lzv=13 was

also not signiﬁcantly different from those at Green Bay, Muskegon and Grand Rapids.

41

Table 5. Analysis of the accumulation of DSV to thresholds of 18 and 30 per growing
season, May 1 through September 30, at seven locations in the greater Michigan region
from 1948-1999 including a) the median DSV accumulated per growing season and b)
the rate of change in DSV accumulated per growing season.

a) Median number of days (t) to DSV accumulation

thresholds (2 v) per growing seasonl

 

 

NW S station2

 

t(2 9:18) IQ: v=30)
Y62 81.0 ab3 95.0 ab
APN 81.0 abc 99.0 ab
Tvc 83.0 a 107.0 a
GRB 69.5 cd 84.0 c
MKG 68.0 bcd 86.0 bc
GRR 61.5 cd 87.0 bc
TOL 57.0 d 76.0 ' c

 

b) Non-paramentric trend in number of days (I)
(rate of change in t per year) to DSV accumulation
thresholds per growing season

 

 

NWS station ti v=18) t(2 0:30)
Y62 -O.48 *4 -0.45 *
APN -0.83 -1.08 **
TVC -0.08 -0-18

GRB -014 -0-21

MKG -032 -0-33

GRR -0.64 * -1.09 ***
TOL -0.39 -0.47 *

 

‘ Growing season included May 1 through September 30 in all years (1949 - 1999).
2National Weather Service (NWS) station locations: Y62 - Sault Ste Marie; APN -
Alpena; TVC - Traverse City; GRB - Green Bay, WI; MKG - Muskegon, MI; GRR -
Grand Rapids, MI; TOL - Toledo, OH.

3Within a Single column, values followed by the same letter are not Si gniﬁcantly
different at P = 0.05 (Kruskal-Wallis One Way Analysis of Variance on Ranks).

4 Rate of change is significantly greater than zero at P = 0.05 (*), P = 0.01 (**), or P =
0.001 (***) (Kendall Tau b).

42

Sault Ste Marie, Alpena, Muskegon and Grand Rapids were not significantly different
from one another with respect to t2v=30.

The time to these spray thresholds decreased over the 50-year period at all
locations (Table 5). The time in days to both thresholds signiﬁcantly decreased at Grand
Rapids and Sault Ste Marie, and 126:” decreased signiﬁcantly at Alpena and Toledo,
from 1948-1999. Alpena and Grand Rapids had the greatest decrease over time in the

number of days to the thresholds.
Analysis of Highlinonducive Periods

The numbers of periods highly conducive to late blight within a growing season
for each location are shown in Figure 9(a-g). Conditions were considered to be highly
conducive if 10 DSV or more were accumulated in a five-day period (2v210 ). Years
when the total number of periods was zero are Shown with a black box at the origin,
while years without a bar or origin point were not included in the analysis. The peak
number of Evz 10 occurred from 1978-80 and 1994-99 for most locations. Maximum
sz 10 increased with decreasing latitude. A decreasing frequency of zero values over
time was noted for Traverse City, Grand Rapids, and Toledo.

Median values of Evz 10 and rate of change in the number of periods for each
location from 1948-1999 are shown in Table 6. 2v210 at the three northern stations were
not signiﬁcantly different from one another, nor were the southern stations signiﬁcantly
different from one another. At Sault Ste Marie, Alpena, Muskegon, Grand Rapids and

Toledo 2v210 was not statistically different between locations. The increase in the

43

.—
I

Number of Five-Day Intervals with 210 DSV Accumulation

 

15-

“3*

a. Y62

 

 

15~

10-

b. APN

 

 

 

 

 

15*

10-

c. TVC

m
U!- w
..A.uu- n.-
.—
uLca .ll‘r’k"l J's”
alum.
.-
uw .2
.-
:-
mw-l'm mac-p.-
-——v-———-—-_
mam-um
«Inc—ow.“ a.
Ann I
..
n -. ..-
nu". 'uuca,‘-.—- -‘-=1—-
-n

 

 

 

W.
“M m
m“
I-‘ >Hu 3.1-._
. l“ —- . - v . u
.a—An-a-I
m
W

- ul- W au- 1.
~ :1: n\. .a—J-u-In-— . .
run-avi- ma» ”0"-
:Mmu-m-n --

a - u . .-

0”. cm

.-
mp1! “.n '-r 0...

mm
W

 

 

 

 

I F [ I j

1950 1960 1970 1980 1990 2000
Year
Figure 9 (a-g). Number of ﬁve-day intervals with disease severity value
accumulation of ten or more per growing season for: a) Sault Ste Marie
(Y62), b) Alpena (APN), c) Traverse City (TVC), (1) Green Bay (GRB),
e) Muskegon (MKG), f) Grand Rapids (GR), and g) Toledo (TOL).
Note change of scale in 9g.Years when the total umber of high
accumulation periods was zero are shown with a black box at the origin,
while years without a bar or origin point were not included in the analysis.

44

Table 6. Analysis of the number of ﬁve day periods with an accumulation of ten or more
DSV per growing season, May 1 through September 30, at seven locations in the greater
Michigan region from 1948-1999 and from 1948-1990 including a) the median DSV
accumulated per growing season and b) the rate of change in DSV accumulated per
growing season.

 

a) Median number of ﬁve days periods (ti to tm)
with an accumulation of ten DSV or greater

N W ' l r 5 .
S station ( 2 I” v 2 10) per growrng season2

 

Y62 2.0 bc
APN 2.0 bc
TVC 1.0 c

GRB 5.0 a

MKG 3.0 ab
GRR 4.0 ab
TOL 4.0 ab

 

b) Non-paramentric trend in the number of ﬁve days
periods (ti to n+5) with an accumulation of ten DSV

NWS station or greater per growing season (rate of change in
periods per year)

Y62 0.00
APN 0.00
TVC 0.25 *4
GRB 0.05
MKG 0.07
GRR 0.14 *
TOL 0.13 **

 

' National Weather Service (NWS) station locations: Y62 - Sault Ste Marie; APN -
Alpena; TVC - Traverse City; GRB - Green Bay, WI; MKG - Muskegon, MI; GRR -
Grand Rapids, MI; TOL - Toledo, OH.

2Growing season included May 1 through September 30 (153 days) in all years (1949 -
1999) in analyses.

3 Within a Single column, values followed by the same letter are not signiﬁcantly
different at P = 0.05 (Kruskal-Wallis One Way Analysis of Variance on Ranks).

4 Rate of change is signiﬁcantly greater than zero at P = 0.05 (*) or P = 0.01 (**)
(Kendall Tau b).

46

number of 2v210 intervals from 1948 to 1999 at Grand Rapids, Toledo and Traverse
City was significant statistically. The greatest increase in sz 10 was measured at
Traverse City. Increases in live 10 at Green Bay and Muskegon were also positive, but

there was no change in the number of Evz 10 intervals at the two northernmost stations,

Sault Ste Marie and Alpena (Table 6).
Timing of Chgge in Risk

The analysis of DSV accumulation during 30-day periods throughout the growing
season, 23:3," , started on the ﬁrst and ﬁfteenth of each month, is Shown in Table 7.
Median 30—day 2v were consistently lowest at Traverse City, except during the period
beginning June 15, when Alpena was slightly lower. Highest median 30-day 2v
generally occurred at Toledo for all periods. On only one occasion (Julyl-July3l, Grand
Rapids) were the medians of any of the four southernmost locations of Green Bay,
Muskegon, Grand Rapids and Toledo statistically different from one another. The three
northernmost locations of Sault Ste Marie, Alpena and Traverse City were not
statistically different throughout the season. From June 15 onward, the median 30-day 2v
at Grand Rapids was not Signiﬁcantly different from the medians at the three
northernmost stations.

All trends of 30-day 2v Since 1948 were positive with the exception of Muskegon
and Grand Rapids during September. Greatest increases occurred during periods that

started at June 15 or later. The increase in 30-day iv was statistically signiﬁcant at Sault

Ste Marie, Alpena, Muskegon and Grand Rapids for periods beginning on both June 15

47

.3 =3. =22.on Aid 5d u m do AL mdd n m «a 83 :9: 83on based?!» ﬂ omega do 2mm 4

.Amxcem :0 35:3, do warmed. an? 0:0 m=_e>»-_8_m=§v mod n m De :8de €5.85ch .0: 93 8:2 2:3. 05 .3 330:8 $29» 6528 Bwﬁm a 55:5 m
.EO .8209 - 15.—.32 .mEdeM 2.80 - «do

:2 .cowoxmsz - 0M2 a3 sham 520 - mac 55 85.65. - U>F Heed? - 2% SEE 2m 23am - do». $5582 :23; $37: 833m 3533 3:262 N
.mombmcm E Adda - add: 980» =a E Aged mnC dm dons—83m 5385 d .32 dons—2: common wEBEO _

 

 

 

 

 

ddd dmd Hmd Ed wdd 3d a. ddd 3d mdd 40H

dd .d- odd ._. mdd i. wmd .1. dmd ... dmd 3 .d 3 .d ddd NED

wdd- 2 .d ndd 3d a. 2 .d ... 2 .0 Odd wdd ddd 0M2

L _d nod _ _d ddd wdd d dd wdd ddd ddd mmo

dd 2 .d dd .d add ddd odd ddd ddd ddd UPH

2d a. mmd ndd mmd 3. mmd .2. de 2d ddd cod 2%

wdd hdd m _ .d 2 .d ._. E .d a. m: .d odd ddd dd 8%

endow 238m omega. 2m: 3A3 23% ones: 2.6.3 582 :o. m

> 23m. N > mdma< N > ~w=< N 32b: N > 135 N 3223‘. N > 1.5:. N > 2.82 N 3 79: N WM?

.533 wEBBw 5m zaddm waist 35 :odﬂsﬁsoom >mQ E Cam» god $8 5 owcano do 88d use: 03258982 3

a ddd a d.md a odd an odd an ddd a ddd m dd_ m ddd m dd 40%

pm 0: pm mdd an ddd one mdd on We— ﬁ 03 a WQ 5 md on dd ~30

an d.m_ a ddd e ddd pa ddd an ddd pm 3: an d.: one dd on. dd 0M2

an d.§ m odd a mdd a mdd a m._d a QB one D: an dd an d.m mMO

3 odd d d.2 9 odd o 93 o d.: a QB on dd on dd 0 dd U>H

on Q? an ddd an odd on d.w_ on dd— n dd 0 dd on dd on dd 2%

am dd“ an d._d an ddd one ddd on dd~ an d.: on dd 0 dd no dd 8%
cease. 23% 8 mad 2.5;. 3%: as a... a e

.. saw an... N a a... N anaew a sew ....__...Hw shew .mﬁw .mhw emnm

 

.538 waged cod mdotod haddm macaw QNV 2233833 >mQ 5602 Q

 

downed. @5on Hon mdotom haddm war—2o code—2:38 >mD E omSEo do 88 3 due
common wEBEw cog adored meddm meted dad cede—scusooa >mQ 5608 2d? ”wEEUE add—ddg 88d 5&2 :awEomE 830% 05
E 28:82 co>om “a .dm BBEoEom smacks“ d >22 d083, wEBoew eon adored Quidm macs—o cone—2960a >mQ do imbued. d 033.

48

and July 1. At Grand Rapids, 30—day 2v increases were statistically signiﬁcant starting
on July 15 and August 1. The increase in 30-day 2v at Alpena starting August 15 was
also signiﬁcant. Finally, 30-day 2v at Toledo increases were signiﬁcant starting June 1.
Mean 30-day iv for the each location’s full growing season are shown in Figure
10. 30-day 2v increased at all locations from May until August, then decreased to the end
of the season. The change in environmental conditions favorable to late blight closely
mirrored the seasonal pattern of potato canopy leaf area index (LAI), also estimated in
Figure 10. While the exact timing and magnitude of LAI varied with variety, planting
date and growing season, it typically increased from emergence until about August, when
the canopy senesced (Allen and Scott, 1992). Mean 30-day 2v values circled in Figure
10 increased signiﬁcantly in 30-day 2v from 1948-1999 (extracted from Table 7). In
most cases, signiﬁcant increases in 30-day 2v occurred during June and July, earlier in

the season than the more typical climatological late blight risk peak in August.

Environmental Variable Trends

Rate of change statistics for monthly mean dew point and ambient air temperature
at the seven locations from 1948-1999 are shown in Table 8. For every month, Alpena
had the greatest increase in dew point temperature of any station. The 50 year increase in
dew point temperature at Alpena was signiﬁcantly greater than zero during July and
August, and also for the growing season as a whole. Increases only slightly less than
those at Alpena occurred at the two southernmost stations of Grand Rapids and Toledo

during June, July and August. The increase was signiﬁcantly greater than zero at Grand

49

 

35

30-

251

20-

15-

104

Mean 30-Day DSV Accumulation

 

 

 

 

O p=0.05
o Y62
o APN ’
v TVC
v GRB I
- MKG ' 3
n GRR I ©
0 TOL v a
. © 8
© 0
0
@ ' ©
2 e . 9
V
9? © 0
O 9 O v .
0 v
3 a
o
a O
a U
9

 

Canopy
LAI

<1<1-2 1-4 2-5 3-6 3-6 4-6 2-5 2-4

 

 

 

r I I I T T— I I I

May1 May 15 June1 June 15 July1 July 15 Aug1 Aug 15 Sept1

Starting Date for 30-Day Mean

Figure 10. Mean 30—day DSV accumulation with starting dates at 15-day
intervals throughout the growing season. Locations where the increase in
30-day DSV accumulation during 1948-1999 was signiﬁcantly greater

than zero at the p=0.0S level are circled. Associated 30-day accumulation

rates of change and signiﬁcantly different were extracted from Table 6.
Estimates of potato canopy leaf area index (LAD for northern growing regions

(e. g. latitude 42N+) are included for each 30-day period (Allen and Scott,
1992). LAI = 3 is considered full canopy.

50

.3 amp =8=§V $188de .0 5...: So u m .5 mod n m a 83 55 seem €58.25: 2 omega .6 2mm.

.30 .8208 - 40H MH2 523% 2.80 - ~30 m2 5095332 .. 0M2 55

..3m :85 - mam ”.96 8.822% - 0?“. Hand? - 77?. ”232 3m :smm - do». “£5932 533% $375 858m 3533 3:23 Z N
.8932“ E 83— - ddd: was?» :3 5 33d mm: dm BdEoEom swsobd _ .32 32:2: .838 wEBBO _

 

 

 

 

 

5d ddd ddd mdd ddd ddd 40,—.

3d ddd ddd ddd ddd mdd ~30

3d ddd- ddd ddd “dd- 5d- 0M2

ddd ddd _d.d ddd ddd mdd 55

3d ddd 8d ... ddd ddd mdd U>H

* ddd mdd ddd ddd mdd mdd 2%

3d 8d add 5d ddd mdd dd>
aegm eonﬁoaom 6:33 .33 0:3. .32 coca; mgz “M

533 336% 5d .3 838388 :3 E Amman—U do 835 25.5 092589.82 3

ddd ddd ddd * ddd ddd ddd 40...

dd 5d- ._. mdd mdd mdd 5d ”20

ddd 5d. 5d 5d ddd ddd- 0&2

* ddd mdd ddd mdd ddd 5d mMO

ddd ddd ddd ddd ddd ddd D:

.22.. mdd mdd * mdd 1* ddd mdd odd Zm<

3d 3d 5d 5d 5d ddd dd>

:8an donEoEow saws/5. b3. 2:; .32 Naodﬂm mBZ

 

.538 953% Sn 888382 .53 Bed 5 Ema—3:0 do 33.: 28: 030883952 3

23823.3 :3 3 98 83823.8 “Began—U 3 no“. dddded— 8o: :2on
@3522 532w 2: E 20:32 55% 3 .dm BdEoEom cwsoﬁa ~ .32 .2038 mEBEw Hod owSEo do 88 do 3.33.4. .w 2de

 

 

Rapids in August and at Toledo in July. The seasonal total increase in dew point
temperature was also signiﬁcantly greater than zero at Green Bay. While the greatest
increase in dew point temperature occurred in May at Sault Ste Marie and Alpena, all
other locations had higher increases later in the season.

In contrast to the later season dew point increases, the greatest increases in
ambient air temperatures occurred in May at most locations. Greatest increases again
occurred at Alpena, except in July when the increase was slightly higher at Toledo. The
mean air temperature increase was signiﬁcant at Alpena for the seasonal mean and at
Traverse City for July. The only air temperature decreases occurred at Muskegon from

1948-1999 during May, June and September.

DISCUSSION

Environmental conditions between 1948 and 1999 became more conducive for
development of potato late blight throughout the Upper Great Lakes region, within the
parameters of what is currently known about the inﬂuence of temperature and relative
humidity on late blight development. Risk of infection has also been increasing with the
advent of metalaxyl-insensitive strains of P. infestans and tightening fungicide
regulations. Host resistance does not seem to be a viable solution to late blight risk in the
immediate future. Taken together, these trends may mean that potato production is
becoming relatively more difﬁcult through time in the Upper Great Lakes region.

The increase in late blight risk, as measured by at least one of the risk indicators

derived from DSV estimates, was signiﬁcantly greater than zero at every location. For

52

each tested risk indicator, the increase in risk was statistically signiﬁcant for at least one
of the northern locations (Traverse City, Alpena, Sault Ste Marie) and one of the southern
locations (Toledo, Grand Rapids, Muskegon, Green Bay).

Conditions at Alpena and Grand Rapids became significantly more conducive for
the development of late blight over the period from 1948 to 1999 in the greatest variety of
tested indicators used to estimate risk. Risk indicators such as 23v, 2x2 , and 30-day 21v
in June and July also showed statistically significant increase at Sault Ste Marie and
Muskegon. The lowest rate of change over the 50-year period of any location for nearly
all measures of change in risk was found at Traverse City. The only risk related indicator
with a trend that increased signiﬁcantly greater than zero at Traverse City was the
number of ﬁve-day intervals with an accumulation of at least ten disease severity values.

When 1991 through 1999 were removed from the data set, rates of change were
no longer significant for Xv at Sault Ste Marie, Green Bay, or Muskegon. This result
could be related to the implementation of A808 instrumentation, but because these
changes most likely would not have such a broad impact, it may point to increasingly
conducive meteorological conditions in the 1990’s.

2v per growing season, used in this study as an indicator of overall growing
season late blight risk, increased at all stations from 1948-1999. At all locations there was
a decrease in the number of days with no late blight risk (21x0). In addition, the number of
days of each day type indicated that late blight risk from minimal (21x1) through severe
(21x4) levels increased or at least remained constant at each location.

Fungicide applications are often initiated or increased following specific DSV

related thresholds. This research has shown that these thresholds corresponded closely to

53

   

the time of ‘full’ potato canopy (LAI=3). Environmental conditions that inﬂuence late
blight risk also closely mirror the estimated LAI of the potato crop. Knowledge of the
canopy environment, therefore, may be more valuable for management decision making
than Wallin-based DSV systems for monitoring late blight risk in terms of making
fungicide recommendations. A DSV-type system may be most useful with respect to
identifying extremely conducive periods in the growing season (e. g. ﬁve day periods with
at least 10 DSV’s). However, it is difﬁcult for growers to respond to these periods as
high precipitation totals that favor late blight development also make ﬁeld work (such as
spraying) difﬁcult. Because of the high precipitation and humidity during these periods,
they would be fairly easily identiﬁed without a DSV system. Future research should
focus on the usefulness of DSV-type late blight risk systems.

As shown in Figure 10, ambient environmental conditions most favorable for late
blight occurred during August coincident with peak leaf area index (LAD in the potato
canopy. This period is crucial because when LAI 23 canopy conditions are increasingly
more conducive to late blight than ambient conditions might suggest. Both air flow and
light penetration decrease as LAI increases, resulting in a greater likelihood of late blight
development because of higher relative humidity percentages, cooler temperatures and
longer leaf drying times in the canopy. Signiﬁcant increases in conducive conditions as
measured by 30-day 2v occurred during June and July, prior to the August peak in late
blight risk. This result is important in terms of foliar fungicide spray timing early in the
season, e. g. shortly after emergence, especially in June. The period of increasing risk
spans the period of canopy development, from just after emergence in June to the August

peak in measured late blight risk indicators, indicating that from 1948 to 1999 higher late

54

 

blight risk occurred more quickly after emergence and remained higher for longer periods
of time, e.g. during times of full canopy development.

When 30-day trends in the environmental variables of dew point and ambient air
temperature were analyzed, ambient air temperature increased most frequently in May.
This result suggests an earlier spring warm up and a possible lengthening of the growing
season. Growth of the potato canopy occurs with development and expansion of
individual leaves and the rate of development and expansion increases with increasing
temperatures. Warmer temperatures in May lead to higher LAI earlier in the season,
creating a canopy microclimate more conducive to late blight development at the same
.time as larger scale weather patterns are also increasing late blight risk.

Dew point temperatures, by contrast, increased less in May than during the rest of
the growing season. This result would suggest an increase in growing season moisture
throughout the Upper Great Lake region in June, July, August and September. Because
the rate of change in late blight risk was strongest in June and July, when dew point
temperatures are increasing as opposed to ambient air temperatures, increasing moisture
is most likely impacting the accumulation of DSV more than change in temperature. The
accumulation of fungicide spray threshold values earlier in the season, is likely also being
impacted more by the increase in dew point temperatures in June and July than the earlier
spring warm up. Median time to threshold values across locations corresponded to
estimates of full canopy conditions.

Increases in dew point and air temperature are supported by the research of Karl
and Knight’s on trends in precipitation in the United States, which estimated a 10 percent

increase in the annual precipitation as a national average in the last 80 years (1998).

55

 

During this time, heavy and extreme precipitation events began to contribute a greater
portion of the total precipitation in both single and multiple day precipitation events. The
increasing frequency of these events is likely one of the factors inﬂuencing the increase
of highly conducive ﬁve-day intervals. The annual number of days with precipitation
also increased in the contiguous United States (Karl and Knight, 1998).

In the Great Lakes region, precipitation amounts have risen approximately 0.4
mm yr'l since 1895, with a significantly greater number of wet days following wet day
(Andresen et al., 2001). These increases in annual precipitation, number of days with
precipitation, and the number of wet days following wet days are each expected to have

increased risk of potato late blight infection and subsequent yield and economic losses.
CONCLUSIONS

Throughout the Upper Great Lakes region, the risk of potato late blight has been
increasing due to increasingly favorable environmental conditions. The most consistent
indicators of change in estimated risk of late blight development across the greater
Michigan region were: a) the increase in total DSV (2v) per growing season, b) the
decrease in days with a DSV equal to zero (conditions not conducive to late blight
development), and c) the increase in the number of days per season with a DSV equal to
2 (conditions moderately conducive to late blight development). These results may
indicate that the resurgence of potato late blight reported in Fry and Goodwin (1997) may

be due not only to metalaxyl-insensitive genotypes, but may also be related to an increase

56

in the frequency and duration of environmental conditions conducive to late blight and to
earlier canopy development.

Future research should also explore the long-term trends of other weather-based
plant pathogen and host systems. Results from this and similar trend analysis can then be
used as a baseline to compare the affects of various climate change scenarios on pathogen
systems. New data sources, such as NEXRAD precipitation data, also will provide new
research opportunities for estimating potato late blight risk on ﬁner spatial scales. With
respect to P. infestans, the usefulness of DSV-type late blight risk systems should be
analyzed and the threshold values for both relative humidity and temperature should be

re-evaluated.

57

CHAPTER 3
CHARACTERIZIN G THE LEADING EDGE OF A POTATO

LATE BLIGHT FIELD INFECTION
INTRODUCTION

Potato late blight, caused by Phytophthora infestans Mont. de Bary, is the most
important disease of the potato in all potato growing regions throughout the world (GILB,
1999) and, economically, the most important potato disease in North America (Inglis et
al., 1996). Late blight spreads in potato canopies by continuous leaf and stem infection
when inoculum is present and environmental conditions are favorable, typically moist
cool weather (Lacy and Hammerschmidt, 1995). Potato late blight risk has been
estimated with for several decades Wallin daily disease severity values, which are
calculated based on temperature and relative humidity conditions (W allin, 1962;
MacKenzie, 1981). Disease severity values (DSV) range from O to 4, with 0 representing
conditions unsuitable for infection and 4 representing extremely conducive conditions
(Wallin, 1962).

Cultural practices, such as timing of irrigation and harvest, as well as chemical
control through intensive fungicide use in combination with the desiccation of infected
plants, are often used to as a means of limiting epidemics. Analyses of spatial and
temporal dynamics of late blight and other Phytophthora species have not focused
speciﬁcally on understanding in-ﬁeld epidemics (J aime-Garcia, 1998; Zwankhuizen et

al., 1998; Ristaino and Gumpertz, 2000). The spread of potato late blight and the percent

58

of the foliage infected through time and space as the epidemic progresses are important to
growers who must make management decisions to limit the spread of late blight. No
fungicides currently in use completely arrest epidemic development, so desiccation of
infected plants is a critical component of risk management (Kirk et al., 2000; Mayton et
al., 2001). Early detection and conﬁdence in the desiccation zone is especially important
given that infected plants reach 100 percent foliar infection approximately 20 days after
initial inoculation (Baker et al., 2000; Mayton et al., 200]).

The research described in this chapter characterizes the spread of the leading edge
of a potato late blight field infection after a point source inoculation. Late blight
epidemics in Michigan are often initiated from infected seed and infection spreads from
this single initial point source. Linear regression techniques that incorporate time and
weather-based independent variables are used to empirically describe the rate and
magnitude of the spread. These models may then be used within deﬁned conﬁdence
bounds in practical, post-infection situations. Although late blight can also spread from
outside sources such as cull piles and nearby infected ﬁelds, it is not possible to

determine exact location of inoculum source for modeling purposes.

Spatial Methods in Plant Pathology

Spatial statistical methods tend to focus on the structure of data sets. Space, time
and attribute characteristics then deﬁne the applicability of point pattern, zonal or

geostatistical methods. The prominence of geostatistics (e. g. semivariogram and kriging

analysis) in plant epidemiology spatial applications in recent years no doubt stems from

59

this traditional perspective. With the growing popularity of methodologies that
incorporate geographic information systems (GIS), a considerable number of studies have
been published recently that both use GIS and spatial analysis for epidemiological
research (Headrick and Pataky, 1988; Lecoustre et al., 1989; Lannou and Savary, 1991;
Orum et al., 1997) and that advocate its use in epidemiological studies (Chellemi et al.,
1988; Rossi et al., 1992; Liebhold et al., 1993; Nelson et al., 1994; Madden and Hughes,
1995).

The methods used in this analysis deviate from traditional methods for
determining spatial structure. Chrisman (1997) made the case that of the space, time and
attribute aspects of an analysis, one must be ﬁxed, one controlled, and one measured (e. g.
time held constant, space divided in consistent blocks, severity of infection estimated).
Although this may be true in a traditional multidisciplinary geographic setting, the
method and data source used in more non—traditional research ventures should
incorporate assumptions of ﬂuidity that are not accounted for in the generally accepted
‘ﬁxed, controlled, measured’ scheme. Field observations, although occasionally obtained
mechanically, are generally gathered by trained ﬁeld technicians, where human error and
bias may result. Time of assessment varies from day to day, placement of initial inoculum
varies from plot to plot, and space is controlled very weakly as it is impossible to assess
ﬁeld plots daily in unchanging zones. In addition, confounding variables such as
variability in plant density, canopy structure, and differences in fertilization and irrigation
at a micro-scale are impossible to be accounted for in a way that would make most

traditional methods usable.

60

In epidemic systems that develop spatially and temporally without the direct
impact of physical—based explanatory variables such as weather, a theoretical model of
disease progress in space and time can be relatively simply established by combining
deﬁned mathematical relationships (Olanya and Campbell, 1990). Straightforward
stochastic techniques using neighbor relations have been defined for viruses (Gibson,
1997) and Spread of fungal pathogens not linked to weather variables (Goleniewski and

Newton, 1994; Xu and Ridout, 1998; Xu and Ridout, 2000).

Objectives

Because potato late blight spreads rapidly, limited options are available to
growers once infection occurs. Infected seed tubers often initiate epidemics at a single
point in the canopy in regions, such as the Upper Great Lakes region, where P. infestans
does not typically over-winter as mycelial fragments or spores in the soil. This chapter
describes the analysis of late blight spread within an experimental plot and speciﬁcally
associates the movement of the leading edge of infection outward from the point of initial
inoculum introduction.

The analysis follows the movement of late blight symptoms through the potato
canopy using the lesion farthest from the point of initial inoculation both along-row and
across-row as an indicator of movement through the canopy. Amount and rate of spread
of late blight varied due to differences in climate and host susceptibility. Therefore, it
was important that this study encompassed several years and was highly replicated within

years. The primary objective of this analysis was to measure the movement of the leading

61

edge of a potato late blight ﬁeld locus in space and time based on prior knowledge of
location and time of inoculation. Linear regression models that used: a) time in days after
inoculation as the independent variable, b) Wallin’s disease severity values estimates, and
c) hourly microclimate characteristics of the canopy, to predict location and rate of

appearance of lesions distal to the initial point of inoculation (LD) were calculated.

METHODS

Field plots

Previous experiments at Michigan State University (MSU) have identified potato
cultivars and ABL with different responses to foliar late blight. Snowden has consistently
been one of the most susceptible (Douches et al., 1997). In all ﬁeld experiments, cut and
whole seed pieces (75-150g) of potatoes, cv. Snowden, were planted with 27 cm between
seed in north-south rows 86 cm apart at the Michigan State University Muck Soils
Experimental Station, Bath, MI (Houghton Muck; Euic, mesic Typic Haplosaprists).
Soils were plowed to 20 cm depth during October following harvest of preceding crops.
Soils were prepared for planting with a mechanical cultivator in early May and fertilizer
applied during ﬁnal bed preparation on the day of planting. Fertilizers were applied in
accordance with results from soil testing carried out in the spring of each year and about
250 kg N/ha (total N) was applied in two equal doses at planting and hilling. Additional
micronutrients were applied according to petiole sampling recommendations and in all
years; approximately 0.2, 0.3 and 0.2 kg/ha boron, manganese and magnesium,

respectively, were applied as chelated formulations.

62

Plots were irrigated as necessary to maintain canopy and soil moisture conditions
conducive for development of foliar late blight (Madden and Hughes, 1995) with turbine
rotary garden sprinklers (Gilmour Group, Somerset, PA, U.S.A.) at 1055 1 H20 ha/hr and
managed under standard potato agronomic practices. Weeds were controlled by hilling
and with metolachlor at 2.3 l/ha 10 days after planting (DAP), bentazon salt at 2.3 l/ha,
20 and 40 DAP and sethoxydim at 1.8 l/ha, 58 - 60 DAP. Insects were controlled with
imidacloprid at 1.4 kg/ha at planting, carbaryl at 1.4 kg/ha, 31 and 55 DAP, endosulfan at
2.7 l/ha, 65 and 87 DAP and pennethrin at 0.56 kg/ha, 48 DAP. The dates of application
were similar for all years.

Zoospore suspensions were made from P. infestans cultures of a single isolate,
[MI 95-7, U88 genotype, insensitive to mefenoxam/metalaxyl, A2 mating type ], the
predominant biotype present in the major potato growing regions of North America (Fry
and Goodwin, 1997), grown on rye agar plates for 14 days in the dark at 15°C. Sporangia
were harvested from the rye agar plates by rinsing the mycelial/sporangial mat in cold
(4°C) sterile, distilled water and scraping the mycelial/Sporangia] mat from the agar
surface with a rubber policeman. The mycelial/sporangial suspension was stirred with a
magnetic stirrer for 1 hour. The suspension was strained through four layers of
cheesecloth and the concentration of Sporangia was adjusted to about 1 x 103
Sporangia/ml using a hemacytometer. Sporangia] cultures were incubated for 2-3 hours
at 4°C to stimulate zoospore release. The suspension was sprayed on greenhouse grown,
potted potato plants at a rate intended to expose all potato foliage to inoculum of P.
infestans. After inoculation, plants were covered in plastic to prolong leaf wetness and

moved to growth chambers held at a constant temperature of 15°C. Single inoculated

63

plants were moved to predetermined positions within the research plots after lesions
developed and Sporangia were visible on leaves and stems. Research plots were
untreated with any fungicide during the last week of July in each year of the study.
Fourteen research plots were included in the analysis; 8m x 26m plots were inoculated at
the center point of their longest edge in 1997 (3 plots) and 1998 (5 plots); 16m x 16m
plots were inoculated at their center points (8m from the edge along the ninth row) in
1999 (2 plots) and 2001 (4 plots),

Mean percent blighted foliar area was measured by subjective visual estimation
from the time the plots were inoculated until complete defoliation of the canopy. Plots
were assessed throughout the epidemic at 2-meter intervals along-row in each row of the
plots from 1997-1999. Assessments occurred every 3-5 days. In 2001, daily assessments
of percent blighted foliar area were made at 1m intervals along-row in each row from the
time lesions could be visually detected until they reached the outer edges of the plot.
When lesions reached the outer edge, 1m measurements were continued on a 3-5 day

schedule.

Regression Analysis

Following data collection, each plot was subdivided into quadrants from the point
of inoculation. Rectangular plots were divided into two quadrants, because the
inoculation point was in the center of the longest edges, and square plots were divided
into four, because the inoculation point was at the plot’s center point. For each year, half

of the quadrants were randomly selected for use in model estimation (set A). The other

64

quadrants were used for model testing (set B). The lesion distal to the point of initial
inoculation was determined for each quadrant using the percent blighted foliar area
estimates and became the derived variable, LD.

Least squares linear regression and multiple regression were used to quantify the
association between independent variables and location of the leading edge of late blight.
Variables that were not signiﬁcant in the regression, or were strongly collinear with other
independent variables, were not included in the analyses. Multicollinearity was
considered a problem if tolerance [as computed in Statistical Analysis Software (SAS—
Institute, 1999)] of the explanatory variable on the other variables was less than 0.].

Individual variables were tested for association with the dependent variable using
the full data set (P=0.05). Regression equations using these variables were then
reanalyzed including dummy slope and intercept variables to test for significant
differences between data sets A and B. The dummy intercept value was held constant at
zero for data set A and l for data set B. If the dummy intercept was associated (P=0.05)
with LD, then the intercept of data set A and data set B were signiﬁcantly different from
one another. The dummy slope value was held constant at zero for data set A and given
the value of the associated independent variable for data set B. If the dummy slope was
associated (P=0.05) with the dependent variable, then the slope of the relationship
between data sets A and B were considered signiﬁcantly different from one another with
respect to the dependent variable.

If the dummy variables were not signiﬁcantly greater than zero for the randomly
selected data sets, then the model was considered usable in ﬁeld situations. If the dummy

variables were collinear with one another, only the dummy slope was tested (P=0.05). In

65

order to isolate the movement of late blight following the start of the epidemic, data
values equal to zero, representing absence of lesion development, were removed.

For each quadrant and assessment date, the distribution of lesions was analyzed to
result in both along—row and across-row parameters. Both splash and aerial dispersal
impact the spread of Sporangia and late blight of P. infestans. Consequently, new foci
develop and overlap both within and across rows (Ristaino and Gumpertz, 2000).
Therefore, this analysis does not rely strictly on Euclidean distance of disease from
inoculation point. Although this change of assumptions is unusual, we feel that
incorporating separate along-row and across-row components is justiﬁed, as the
characteristics of disease spread along and across rows have not been established and
may be related to differences in plant canopy structure and density and the interaction of
two dispersal mechanisms (Ristaino and Gumpertz, 2000). In addition, recommendations
may be implemented in ﬁeld situations where equipment limitations and ease of use call
for a rectangular desiccation zone.

In order to determine if spread was isotropic, the distance of the lesion distal to
the initial point of inoculation (LD) both along—row and across-row was computed not as a
standard Euclidean distance, but as the greatest number of 1m assessment units
north/south and east/west. For the example quadrant shown in Figure 11, both the along-
row and across-row distance were recorded as 5m although standard Euclidean distance
measurements would indicate 6.4m as the maximum spread (given i as the inoculation
point). The resulting model thus describes rectangular disease spread, as opposed to the

more traditional circular model of disease expansion.

66

 

 

 

 

 

 

 

 

 

 

Along—row

 

 

 

 

 

Across-row

Figure 11. Example of along and across-row distance components
using an example quadrant from the analysis. Instead of estimating
a circular disease expansion radius in Euclidean distance from the
inoculation (i) point (6.4 for this quadrant), the leading edge
expansion model uses the maximum distance from the inoculation
point of both an along-row and across-row component (5, in both

cases).

67

The time dependent regression model included no explanatory variables other
than distance from inoculation point (LD) and days after inoculation (DAI). This model
was used with dummy variables to test for a signiﬁcant difference between across-row
and along-row spread for data set A. Speciﬁc environmental variables, as well as
indicators of potato late blight risk derived as secondary variables from hourly ambient
air temperature and dew point temperature from day of initial inoculation to day of
assessment are described in Table 9.

Similar to the time dependent approach, two more detailed regression models
were also developed using explanatory variables in addition to space and time. The
models relied on disease severity value (DSV) and weather characteristics from time of
inoculation through assessment date for L9. Both risk estimates and weather
characteristics were derived as secondary variables from hourly weather data from an
automated weather station in East Lansing MI, approximately 15 km from the ﬁeld plots.
This station is part of the Michigan Automated Weather Network (MAWN) and is the
closest station with hourly temperature and relative humidity data consistent over the
ﬁve-year period of the study.

An approach that may be described as driven by the physical environment used
traditional weather variables that inﬂuence risk of late blight spread, speciﬁcally
temperature (T) and relative humidity (RH). Included in the analysis were the number of
hours from initial inoculation with: a) each of the relative humidity/temperature
categories used in modiﬁed-Wallin DSV calculation (2W1, RH>80% and 7°CsT<11°C;

2W2, RH>80% and 11°CsTs 15°C ; 2W3, RH>80% and Ts 15°C; ZWALL, RH>80% and

7°C sTs 15°C); b) each 10 percent humidity category from 50 percent upwards (2RH50,

68

Table 9. Environmental variables and indicators of potato late blight risk derived as
secondary variables from hourly ambient air temperature and dew point temperature
from day of initial inoculation to day of assessment.

 

 

Variables Description
DA] days after inoculation
DS V potato late blight disease severity values, calculated for each day
after inoculation using the modiﬁed Wallin method explained in
Chapter 2.
T temperature in °C
RH relative humidity
2T)“ sum of hours since inoculation with temperature greater than 15°C
ERHW sum of hours since inoculation with 90% SRH s 100%
ZRHBO sum of hours since inoculation with 80% SRH <90%
2RH70 sum of hours since inoculation with 70% SRH <80%
ERHM sum of hours since inoculation with 60% SRH <70%
ERH” sum of hours since inoculation with 50% sRH <60%
BRI-Im sum of hours since inoculation with 70% SRH s100%
2W, sum of hours since inoculation with 80% SRH $100% and 7°C
sT<11°C
2W2 sum of hours since inoculation with 80% SRH s 100% and 11°C
sT<15°C
23WJ sum of hours since inoculation with 80% SRH s 100% and 15°C 5T
327°C
EWALL sum of hours since inoculation with 80% sRH 5100% and 7°C ST
327°C
2x, total number of days with DSV =1 since inoculation
2x2 total number of days with DSV =2 since inoculation
2x3 total number of days with DSV =3 since inoculation
2x, total number of days with DSV =4 since inoculation
Expo total number of days with DSV>0 since inoculation
2v sum of all consecutive DSV per day since inoculation

 

69

50%sRH<60%; ERH60, 60%sRH<70%; 2RH70, 70%sRH<80%; ERHgo,
80%sRH<90%; ERHgo, 90%sRH_<.100%); and c) accumulated hours at 15°C and above
(2T>15, T215°C) and at a relative humidity of 70% and above (2RH>70, RH270%) (Table
9).

A risk-based approach utilized DSV estimates and their accumulation by day
from inoculation to day of assessment. Derived variables included: total DSV from
inoculation to assessment date (2v), the number of days with late blight risk from
inoculation to assessment date (2x,,>0), and the number of days with each unique DSV

value, 1 through 4, from inoculation to assessment date (2x1, 2x2, 2x3, 2x4).

Application of Leading Edge Regression

For the time dependent model, linear regression was used to model the
relationship between time in days after inoculation and the expansion of the leading edge
from the inoculation point to test for directional differences in data set A. The
implications of this model for use in estimating in-ﬁeld desiccation zone parameters was
explored. Predicted values and the 99 percent conﬁdence bound for those values were

calculated in SigmaStat (SPSS-Inc, 1997).

70

RESULTS

Time Dependent Variables

Linear regression results for the time dependent model are shown in Table 10
(Equation 10.01). Distance of farthest infected foliage from inoculation (LD) was used as
the dependent variable, explained only by days after inoculation. Dummy variables were
not signiﬁcant in the regression model, indicating that the slope and intercept of the
relationship between time as DAI and distance as LD were not signiﬁcantly different
between data sets A and B. The adjusted R-squared value was 0.624 with a error mean
square (MSE), an unbiased estimator of variance for the regression model, of 3.6.

Adjusted R-squared and MSE were similar for the comparison of along-row and
across row spread of late blight using only data set A (Equation 10.02). Predicted
distance of along and across row spread were not signiﬁcantly different from one another.
Adjusted R-squared was slightly higher and MSE was slightly lower than those of the
regression incorporating the full data set, as would be expected given the smaller data set.
Because directional variation was not statistically signiﬁcantly different at P=0.05, the
along-row and across-row measured and predicted distances were combined for all

further analyses. A scatter plot of DAI (from 0 to 30) and LD is shown in Figure 12a.

71

 

' ‘ ' ' ' d' I o'noculation
LeSIon distal to inoculation (Lo) L990" 'Sta ‘ ' (LO)

Lesion distal to inoculation (L0)

 

 

 

 

 

 

 

 

 

 

 

 

 

a.
12 «
10 ~
84
6 4
4 .1
2 .1
o . :::::::.:: e, 3 .
O 5 10 15 20 25 30
Days after inoculation (DAI)
14
12 .
o o o o
O 0
1o « o
8 « o o 00. o o
00 000 to
6 « o 00 00 0
on O 9 °
4 -1 d .0
O O: I .
8: o .
2 ‘ no“. 3 '
.00“. O. o
0 I I I I V I
o 50 100 150 200 250 300 350
Hours since inoculation with RH >90
14
C.
12 .
o o o 0
1o« 0 ° '
o
8-1 0 o o o o
O O O O
6- O O O
o o O O 0
4-1 . . . .
i ’ 3
o
2‘ o O O I I
O O O O O o
o T I r r T I I ‘1' 1
o 2 4 6 a 10 12 14 16 18 20

Days since inoculation with DSV > O (Expo)

Figure 12. Regression results for lesion distal to inoculation
(LD) and: a) days after inoculation, b) hours since inoculation
with relative humidity greater than 90 percent, and c) days since
inoculation with DSV > 0.

72

Weather Variables

Results of linear models incorporating single weather-based independent variables
are also shown in Table 10. The number of hours in the lowest two modiﬁed-Wallin
DSV estimation category (2W,, 2W2) were very weakly correlated with LD (Eq. 10.3 and
10.4). MSE’s were nearly twice those of the regressions incorporating days after
inoculation, indicating considerably more prediction error. Number of hours in the third
modiﬁed-Wallin category (2W3) had a stronger relationship with distance from
inoculation point, with an adjusted R-squared of 0.558 (Eq. 10.05). The sum of all
modiﬁed-Wallin hours, EWALL, (Eq. 10.06) had a slightly higher adjusted R-squared and
lower MSE than Equation 10.05.

A regression equation with hours of temperature greater than 15°C (2R5) as the
independent variable (Eq. 10.07) had adjusted R-squared and MSE between those of the
2W3 (Eq. 10.05) and the sum of all modiﬁed-Wallin hours (EWALL).

Adjusted R-squared values of regression models incorporating hours accumulated
since inoculation in various relative humidity categories ranged from 0417-0603, and
MSE ranged from 3.82-5.6. When the independent variable was hours accumulated
between 70 and 80 percent relative humidity (ERHyo) the regression model yielded the
highest adjusted R-squared value and the lowest MSE. Hours between 50 and 60, 60 and
70, and 90 and 100 (2RH50, 2RH60, ERHgo) had slightly weaker relationships. Relative
humidity hours between 80 and 90 (ERHgo) percent had the weakest relationship with
distance of the farthest lesion from the inoculation point of any tested relative humidity

category. The relationship between 2RH90 and L0 is shown in Figure 12b.

73

Table 10. Linear regression results for time and weather independent variable models.
Dummy slope and intercept variables were included to test signiﬁcant differences
between the models created from two randomly chosen data sets (A and B). Dummy
variables were identiﬁed as signiﬁcant (<0.0001) or not signiﬁcantly different (A=B).

 

 

 

 

 

Eq. Adjusted Error Mean Dummy
# Regression Model R - squared Square (MSE) Intercept Slope
Time Dependent
Data sets A and B
10.01 y=-2.73 + 0.47(DAI) 0.624 3.618 A = B A = B
Along Row vs Across Row
10.02 y = -1.91 + 0.43(DAI) 0.638 3.010 A = B
Weather Dependent
' 10.03 y = 1.60 + 0.24(2w,) 0.335 6.390 = A = B
10.04 y = 3.40 + 0.07(2W2) 0.373 6.027 A = B A = B
10.05 y = -3.17 + 0.04(2W,) 0.558 4.250 =
10.06 y = -l .83 + 0.03(EWALL) 0.597 3.870 A = B
10.07 y = -1.23 + 0.03(ET>,5) 0.569 4.150 A = B A = B
10.08 y = 0.07 + 0.04(2RH90) 0.556 4.270 A = B A = B
10.09 y = -249 + 0.10(2RH80) 0.417 5.610 A = B
10.10 y = -1.67 + 0.13(ERH70) 0.581 4.030 A = B A = B
10.11 y = -1 . 14 + 0.16(ERH60) 0.549 4.340 A = B A = B
10.12 y = -3.49 + 0.18(2RH50) 0.541 4.410 A = B
10.13 y = -l.66 + 0.03(ERH<70) 0.603 3.820 A = B A = B

 

74

Dummy slope and intercept indicated that data sets A and B would not result in
statistically different relationships between these weather variables and location of the

leading edge of infection.
Disease Severity Value Variables

The DSV based regression models utilized input data derived from the standard
meteorological variables in the form of modiﬁed-Wallin DSV (Table 11). DSV
dependent linear regression models each had considerably lower adjusted R-squared
values and higher MSE than the weather or time dependent models. The sum of DSV
from inoculation to observation (2v) as the independent produced a model (Eq. 1 1.1) with
an adjusted R-squared of 0.405. A slightly stronger relationship existed for the sum of
days with DSV>0 (Expo), and the relationship between Expo and LD is shown in Figure
12c. Because the dummy slope was statistically signiﬁcant in this model, the data sets A
and B resulted in models with signiﬁcantly different slopes. Number of days with
individual modified-Wallin severity values from 1 through 4 resulted in much lower test
statistics. Days with DSV=4 (Ex,)had the strongest relationship, with adjusted R-squared
of 0.337 (Eq. 11.6). The dummy slope variable was signiﬁcant when DSV=1 was the
independent variable, indicating that with this model, data from the sets A and B resulted

in signiﬁcantly different slopes.

75

Table l 1. Linear regression results for DSV independent variable models. Dummy
slope and intercept variables were included to test signiﬁcant differences between the
models created from two randomly chosen data sets (A and B). Dummy variables were
identiﬁed as signiﬁcant (<0.0001) or not signiﬁcantly different (A=B).

 

 

 

Eq. Adjusted Error Mean Dummy

# Regression Model R - squared Square (MSE) Intercept Slope

Weather Dependent

11.] y = 1.60 + 0.24(2W,) 0.405 5.722 A = B A = B
11.2 y = 3.40 + 0.07(2W2) 0.419 5.607 <0.0001
11.3 y = -3.17 + 0.04(2W,) 0.162 8.060 <0.0001
11.4 y = -1.83 + 0.03am...) 0.063 9.015 A = B
11.5 y = -1.83 +'o.o3(2w,u) 0.226 7.440 A = B A = B
11.6 y = -1.23 + 0.03(2T>,5) 0.337 6.370 A = B A = B

 

76

Combined Variables

Multiple regression models combining variables that were not collinear and
retained their significance in combination with one another are shown in Table 12. The
model including the hours accumulated in each of the three modified-Wallin categories,
EWALL (Eq. 12.1), had similar statistics to those of the time dependent model, DAI (Eq.
8.1). A model that included relative humidity over 90 percent (ERILgo) and between 70
and 80 percent (2RH70) as independent variables resulted in the only relationship stronger
than those including time in days. after inoculation (Eq. 12.2). Adjusted R-squared was
0.645 and MSE was 3.416. Even in combination, the days with speciﬁc DSV (Eq. 12.3)
were not as strongly related to LD as were the weather dependent variables. The dummy
slope was significant when both DSV=1 and DSV=4 were used as independent variables,
indicating that the models from data sets A and B resulted in signiﬁcantly different rates
of movement of LD through time.

The ﬁnal two combined models (Eq. 12.4 and 12.5) included days after
inoculation (DAI) with DSV variables, and these models had the strongest relationship
with farthest late blight lesion from initial inoculation point of any models tested and the
lowest MSE. Data sets A and B resulted in models that were not signiﬁcantly different,
as the dummy slope and intercepts were not statistically signiﬁcant in the regression.
When DAI as independent variables was combined with: a) DSV=4 (29:4,) and b) both
DSV>O and DSV=4 (2x4, 2%.”) , the models resulted in adjusted R-squared values of

0.673 and 0.717 respectively.

77

 

 

 

 

 

 

 

m u < m u < 82 :8 $963 + $593.? 9559c + wen. u a we

m u < m n < 2; . ﬂed 3866 + c<€$d + 42. .1. a 4.2
682 8:555

586v 23. $3 34662 + 3.8mm; + mom- .1. x 02
252839 >mQ

m u < m u < 63% 26.0 $5686 + @5886 + $4- 0 A «.2

m u < £2 Rod 3:836 + 33886 + 3:82.? + 8. - n A <2

“concomoa 8583
one—m 2.8.3.:— AmmEv tong—om . ~— 35: aﬁmmouwoy— #
.2555 onus—em :32 Spam 13.6.33... .3—

 

80$ saute 32565:? a: .o cooodvv 28%? a 85262
203 9.0—nut? >82:an Am can ,3 8m 8% 2: no 322 95 05 50253 oogocmcwmm $8 8 33:05 203 83a§> 30885
new 32m 48:55 .2258 22%.? acoucoaocﬁ 3:5an >mQ was $533 .08: H8 8:58 coimoewe 30:5 .2 2an

78

Application of Regression Model

As a practical application of model results, the time dependent model developed
from data set A to test the differences between along and across row distances was used
to estimate desiccation zone parameters. The combined data set and resulting regression
line (Eq. 10.02), as well as 99 percent confidence bounds for the model, are shown in
Figure 13. Predicted values obtained from regression equations for data set A and B were
not significantly different (as seen from dummy variable signiﬁcance results of Equation
10.01) and so the equation estimated from set A was retained for use in the estimation of
desiccation zone parameters. The 99 percent upper conﬁdence bound for the model,
where days after late blight inoculation is the independent variable, follows the quadratic

equation:

99%C0nﬁdence = —2.485 + 0.744DAI + 0.003850Al 2

and was used as an approximation of the observable leading edge of potato late blight for
desiccation zone recommendations. Because no lesions were found prior to eight days
after the introduction of inoculum, the bound of late blight expansion at eight days with
99 percent conﬁdence was considered the distance of non-visible expansion of P.
infestans infection. From the above equation, calculating for 8 DAI, the 99 percent
conﬁcence bound was 3.71m. This result was added to the overall confidence bound for
recommendations as a best estimate based on empirical observation for the impact of the

eight-day latent period (delay in expression of symptoms) on late blight spread. The

79

LD, Distance from inoculation (m)

 

25

204

 

y = -1 .952 + 0.436x
0 Distance vs DAI, 1997-2001 (set A)
Confidence Bounds: 99%

 

 

—L
01
l

.1.
O
l

 

 

 

 

5 10

I I I

15 20 25 30

Days after inoculation (DAI)

Figure 13. Observed values and predicted linear regression
line. Distance from initial inoculation point in meters is
predicted using days after inoculation.

80

formula for the expansion of potato late blight with 99 percent confidence from a point

source inoculation both along-row and across-row was therefore estimated as:

99%C0nﬁdence = 1.23 + 0.744x + 0.0038512

Leading edge results from data sets both A and B are shown in relation to this
curve in Figure 14. The size in square meters of the recommended desiccation zone with
respect to time since inoculation, equal to total area with 99 percent confidence, is shown
in Figure 15. Estimated percent foliar infection, interpolated from assessment zone
centroids, for a single ﬁeld plot thirteen days after inoculation in 2001 is shown as an
example (Figure 16). Predicted observable infection is 2.41 m from the point of
inoculation. Observable infection distance with 99 percent conﬁdence is 7.8 m and total

infection distance with 99 percent confidence is 11.5 m from inoculation.

DISCUSSION

Linear regression modeling was used to characterize spread of late blight within a
field plot. While this technique is not common within plant pathology literature, linear
regression was a straightforward method for determining basic association between
spread and independent variables.

The time in days after inoculation (DAI) had the strongest relationship with
distance of farthest lesion from the point of initial inoculation. Number of hours over 70
percent relative humidity (ERHW) had the strongest relationship of the weather
dependent variables. These two independent variables were also highly correlated with

one another, partially because they are both accumulated variables. Although the

81

LD, Distance from inoculation (m)

 

60

50‘

Desiccation zone recommendation
o All data (sets A & B)

 

 

 

30-

20-

10-

 

 

 

 

I I I I I I

1 0 20 30 40 50 60

Days after inoculation (DAI)

Figure 14. All data from sets A and B and the
recommended desiccation zone in meters both along and
across-row from the inoculation point in days after
inoculation.

82

 

 

 

 

 

 

 

— Predicted dsease area (actual)
Precicted total disease area (99% conﬁdence)
— — Predicted observable disease area (99% conﬁdence)
- Late blight bwomes visible in field
6000*
102‘
9
a)
E
92 40004 /
‘3 /
Si /
I ' /
a -' //
< 2000~ .- .- /
.- /
/
.-' " //
04___.;.-;_:.-Jngr_:
0 10 20 30 40

Days after inoculation (DAI)

Figure 15. The increase in the area of predicted observable
disease, the predicted observable disease, and the total
predicted disease. Late blight expansion becomes visible in
the ﬁeld approximately 8 days after inoculation.

83

 

 

 

 

 

 

Observable dseese -percent foliar infection
-o.1.o2 -°-5-1

0.2 - 0.3 =1 - 2

- 0.3 - 0.5 2 - 4

— Predicted observable disease

211% Observable deeese - 99% conﬁdence
5 Desiccation zone

 

 

 

Figure 16. Observable disease from an example 2001

research plot after 13 days after inoculation with the model

predicted observable disease, the predicted observable
disease 99% conﬁdence bound, and the total predicted
diseased area with 99% conﬁdence (recommended
desiccation zone).

84

 

occurrence of potato late blight epidemics has been associated with conducive canopy
weather conditions, the results of this study did not directly support that canopy
conditions traditionally thought to promote late blight are strongly related to the spread of
the leading edge of late blight lesions in the ﬁeld. Several aspects of the analysis could
have impacted this result. In each year, initial inoculum was placed in the canopy at
approximately the same time and canopy conditions, including both canopy development
and diurnal temperatures, were highly favorable for epidemic development. This
similarity in the time of initial inoculum could have impacted the results from other
independent variables. The ﬁrst two modiﬁed-Wallin categories (Eq. 10.03 and 10.04),
for instance, were infrequent during the epidemic period in comparison to the other
weather variables used for independent variables. Because temperatures were relatively
consistent between years during the epidemic, as well, it was not unexpected that relative
humidity categories, without regard to temperature, were strongly related to distance
from inoculation.

These same factors may have impacted the rate of along and across row late blight
spread between replicates and years. While differences in along-row and across-row
spread have been noted in the ﬁeld in some instances (Baker et al., 2000), these
differences may relate to plant density and canopy structure that are related to row
direction in some years, and plots, but not in others. In general it would be expected that
along-row plants would be more likely to have a more similar microclimate and more
likely to be in physical contact with one another. However, it was observed in some ﬁeld
seasons during the course of this study that even relatively slight differences in drainage,

nutrient availability, and/or soil characteristics could impact the structure and density of

85

the potato canopy. Potato late blight spread was observed to be closely associated with
these slight changes in canopy density, which corresponded to row direction in some (but
not all) cases. Further research should, perhaps, incorporate some measure of canopy
characteristics in the analysis.

The weaker relationship exhibited between LD and disease severity value
variables also may relate to the overall suitability of canopy conditions for epidemic
development. These regression results were also impacted by the fact that DSV are not
ratio scale data. 2v assumes that DSV is an interval scale, which may or may not be true.
The analysis of individual values of DSV assume only nominal scale data. Although the
nominal scale is likely a safer assumption with regards to DSV in the independent
variable, counts of these day types are also only interval scale data, which does not match
standard regression assumptions. Probably for similar reasons, all cases where the
dummy slope variable was signiﬁcantly different between data set A and B occurred in
regression models incorporating DSV variables.

Of the combined models, both the combined modified-Wallin category model
(Eq. 12.1) and the model including Expo (Eq. 12.5) included negative coefﬁcients. In the
former, the negative value probably resulted from the fact that nearly all hours were
within one of the three categories. In this case, conducive hours in a lower temperature
class represented the inverse of the hours in the higher temperature classes which were
more conducive for late blight (less time needed in the category for infection to occur).
In the case of the combined model incorporating Expo, the negative coefﬁcient for this
variable was counterintuitive and was related to interaction between Expo and the other

independent variables.

86

Additionally confounding for all model results was the fact that both the
dependent and independent variables were accumulated over time. As yet, the
relationship between potato late blight and the number of days prior to infection that are
crucial to the understanding of risk remain undeﬁned. Since the course of an epidemic is
assumed to be affected to some extent by conditions over the period from inoculation
onward, both distance and environmental conditions were accumulated for each date
from the day of inoculation. Because each daily and hourly value after inoculation is
incorporated into every disease risk estimate, this accumulation results in noise in the
regression models.

The increase in area of the desiccation zone with time continues to make
management of in-ﬁeld potato late blight infections challenging. Because this model
expresses time as days after inoculation (DAI), inoculation date would need to be
estimated using weather conditions, or perhaps highest foliar infection rates. Difﬁculties
may also arise during practical implementation of the model in determining the
inoculation point, as this may not be the most blighted region of the ﬁeld. Canopy
characteristics, such as LAI, affected by factors such as nutrient availability, time in the
growing season of late blight development, and defoliation as a result of symptoms each
could impact the observed location of highest infected foliage. The inﬂuence of canopy

structure and density is also of utmost importance to ﬁeld applications of this research.

CONCLUSION

87

This chapter adds further doubts, in addition to those expressed in Chapter 2, as to
the usefulness of Wallin-type risk values in the estimation and prediction of late blight
risk. Again it seems that canopy conditions are more strongly associated with late blight
infection, as well as epidemic spread in the ﬁeld. Furthermore, the accumulations of
these DSV should be used with caution in any situation, given that they are not clearly
deﬁned as an interval scale numeric variable. Instead, the DSV may serve the purpose of
a day-typin g mechanism. Addition/subtraction and other numerical operations on ‘day-
types’ then become theoretically questionable. Further research is necessary to clarify
the relationship between ambient air condition, canopy environment and spread of potato
late blight.

Further research should also include verifying spread models through extensive
field tests. Modeling the complete late blight epidemic from inoculation to complete
canopy defoliation would also be useful from the standpoint of understanding ﬁeld
infection dynamics. Continued research should also focus on the interplay between
cultivar resistance and fungicide regimes, in addition to environmental conditions. The
implications of this type of research for potato breeders might suggest a focus on slowing

disease spread through the production of varieties which decrease sporulation rates.

88

REFERENCES

Agrios, G.N., 1997. Plant Pathology. Harcourt Brace and Company, San Diego.

Allen, B. and Scott, R., 1992. Principles of agronomy and their application in the potato
industry. In: P. Harris (Editor), The Potato Crop. Chapman & Hall, London, pp.
816-81.

Andresen, J ., Alagarswamy, G., Rotz, C., Ritchie, J. and LeBaron, A., 2001. Weather
impacts on maize, soybean, and alfalfa production in the Great Lakes region,
1895-1996. Agronomy Journal, 93: 1059-70.

Andresen, J ., and Edson, C. 2000. Spatial variability of leaf wetness. SW Michigan
Research and Extension Center Annual Report, MSU Extension/Michigan
Agricultural Experiment Station. East Lansing, MI, 168-170.

Armstrong, R., Barthakur, N. and Norris, E., 1993. A comparative study of three leaf
wetness sensors. International Journal of Biometeorology, 37: 7-10.

Atzema, A.J., 1998. Determination of quality of forecast moisture content of cereals at
harvesting time using forecast weather elements. J. Agric. Engng Res., 71: 119-
125.

Baker, K.M., Andresen, J.A., Kirk, W.W. and Stein, J .M., 2000. Crop disease mitigation:
daily risk modeling for Michigan potato growers, 4th International Conference on
Integrating GIS and Environmental Modeling (GIS/EM4): Problems, Prospect
and Research Needs, Banff, Alberta, Canada.

Barr, A. and Brown, D., 1995. Estimating forage yield and quality changes during ﬁeld
drying for hay. 2. Model of forage drying. Agricultural and Forest Meteorology,
76: 107-127.

Beaumont, A., 1947. The dependence on the weather of the dates of outbreak of potato
late blight epidemics. Transactions of the British Mycological Society, 31: 45-53.

Chellemi, D.O., Rohrbach, K.B., Yost, RS. and Sonoda, R.M., 1988. Analysis of the
spatial pattern of plant pathogens and diseased plants using geostatistics.
Phytopatholog)’, 78(2): 221 —226.

Deshpande, R., Hubbard, K., Coyne, D., Steadman, J. and Parkhurst, A., 1995.

Estimating leaf wetness in dry bean canopies as a prerequisite to evaluating white
mold disease. Agronomy Journal, 87: 613-619.

89

Dorozhkin, N.A., 1971. The inﬂuence of the microclimate of potato fields on the
development of the Phytophthora infestans (Mont) de Bary. Academy of
Sciences of Belorussian SSR, 15(11): 1050-3.

Doster, MA. and Fry, W.E., 1991. Evaluation by computer simulation of strategies to
time metalazyl applications for improved control of potato late blight. Crop
Protection, 10: 209-214.

Douches, D., Kirk, W., Jastrzebski, K., Long, C. and Hammerschmidt, R., 1997.
Susceptibility of potato varieties and advanced breeding lines (Solanum
tuberosum L) to Phytophthora infestans (Mont) de Bary in greenhouse
screenings. American Potato Journal, 74(2): 75-86.

Downing, T.E., Harrison, P.A., Butterfield, RE. and Lonsdale, K.G. (Editors), 2000.
Climate Change, Climatic Variability and Agriculture in Europe: An Integrated
Assessment. Environmental Change Institute, University of Oxford, Oxford.

EarthInfo, 1948-1999. Surface Airways, Boulder, CO.

Fry, W.E. and Goodwin, SB, 1997. Resurgence of the Irish potato famine fungus.
BioScience, 47(June 1997): 363-371.

Gibson, G., 1997. Investigating mechanisms of spatiotemporal epicemic spread using
stochastic models. Phytopathology, 87(2): 139-146.

Giesler, L., Horst, G. and Yuen, G., 1996. A site-specific sensor for measuring leaf
wetness duration within turfgrass canopies. Agricultural and Forest Meteorology,
81: 145-56.

GILB, 1999. Global Initiative on Late Blight. Late Blight: A Threat to Global Food
Security. In: L. Crissman and C. Lizarraga (Editors), Late Blight: A Threat to
Global Food Security, Quito, Ecuador.

Gillespie, T. and Kidd, G., 1978. Sensing duration of leaf moisture retention using
electrical impedance grids. Can. J. Plant Sci., 58: 179-87.

Gillespie, T.J., Srivastava, B. and Pitblado, RE, 1993. Using operational weather data to
schedule fungicide sprays on tomatoes in southern Ontario, Canada. Journal of
Applied Meteorology, 32: 567—573.

Gleason, M., Taylor, S., Loughin, T. and Koehler, K., 1994. Development and validation
of an empirical model to estimate the duration of dew periods. Plant Disease, 78:
1011-1016.

Goleniewski, G. and Newton, A., 1994. Modelling the spread of fungal diseases using a

nearest neighbour approach: effect of geometrical arrangement. Plant Pathology,
43: 63 1 -643.

90

Harris, P. (Editor), 1992. The potato crop: the scientific basis for improvement. Chapman
and Hall, London.

Harrison, J ., 1992. Effects of the aerial environment on late blight of potato foliage -- a
review. Plant Pathology, 41: 384-416.

Harrison, J. and Lowe, R., 1989. Effects of humidity and air speed on sporulation of
Phytophthora infestans on potato leaves. Plant Pathology, 38: 585-91.

Headrick, J .M. and Pataky, J.K., 1988. Spatial and temporal development of common rust
in susceptible and partially resistant sweet corn hybrids. Phytopathology, 78(2):
227-233.

Hijmans, R., Grunwald, N ., van Haren, R., MacKerron, D. and Sherrn, H., 2000. Potato
late blight simulation for global change research, Global initiative on late blight,

pp. 1-2.

Inglis, D.N., Johnson, D.A., Legard, D.E., Fry, W.E. and Hamm, PB, 1996. Relative
resistance of potato clones in response to new and old populations of
Phytophthora infestans. Plant Disease, 80: 575-578.

Jaime-Garcia, R., 1998. The Spatial and Temporal Analysis of Phytophthora infestans
Genetic Diversity and its Inﬂuence on Late Blight Epidemics at a Regional Scale,
University of Arizona, 164 pp.

Jeffries, R. and Heilbronn, T., 1995. Simulating effects of water-stress on growth and
yield of potato. In: P. Kabat, B. Marshall, B. van den Broek, J. Vos and H. van
Keulen (Editors), Modelling and Pararneterization of the Soil-Plant-Atmosphere
System. Wageningen Pers, Wageningen, pp. 213-26.

Johnson, D.A., 1998. Expansion of potato late blight forecasting models for the Columbia
Basin of Washington and Oregon. Plant Disease, 82(6): 642-645.

Kadish, D. and Cohen, Y., 1992. Late blight (Phyophthora infestans (Mont) De Bary)
development from potato seed-pieces treated with fungicides. Pesticide Science,
55: 1151-58.

Karl, T. and Knight, R., 1998. Secular trends of precipitation amount, frequency, and
intensity in the United States. Bulletin of the American Meteorological Society,
79(2): 231-241.

Kirk, W.W., Davies, H.V. and Marshall, B., 1985. The effect of temperature on the

initiation of leaf primordia in developing potato sprouts. Journal of Experimental
Botany, 36(171): 1634-1643.

91

Kirk, W.W. et al., 2000. Combining varietal resistance with managed fungicide
applications for the control of potato late blight. African Journal of Crop Science,
5: 329-335.

Lacy, M.L. and Hammerschmidt, R., 1995. Diseases of potato: late blight. E1802,
Extension Bulletin Michigan State University, East Lansing, MI.

Lambert, D., Currier, A. and Olanya, M., 1998. Transmission of Phytophthora infestans
in cut potato seed. Amer J of Potato Res, 75(257-263).

Lannou, C. and Savary, S., 1991. The spatial structure of spontaneous epidemics of
different diseases in a groundnut plot. Netherlands Journal of Plant Pathology, 97:
355-368.

Lecoustre, R., Fargette, D., Fauquet, C. and de Reffye, P., 1989. Analysis and mapping of
the spatial spread of African cassava mosaic virus using geostatistics and the
kriging technique. Phytopathology, 79(9): 913-920.

Liebhold, A.M., Rossi, RE. and Kemp, WP, 1993. Geostatistics and geographic
informations systems in applied insect ecology. Annual Review of Entomology,
38: 303-327.

Lu, R. and Siebenmorgen, T.J., 1994. Modeling rice ﬁeld moisture content during the
harvest season. Transactions of the ASAE, 37(2): 553-560.

MacKenzie, DR, 1981. Scheduling fungicide applications for potato late blight with
BLIGHTCAST. Plant Disease, 65(5): 394-399.

Madden, L.V. and Hughes, G., 1995. Plant disease incidence: distributions,
heterogeneity, and temporal analysis. Annual Review of Phytopathology, 33: 529-
564.

MASS, 1999-2000. Michigan Agricultural Statistics, Michigan Department of
Agriculture, Lansing.

Mayton, H., Forbes, G., Mizubuti, E. and Fry, W., 2001. The roles of three fungicides in
the epidemiology of potato late blight. Plant Disease, 85(9): 1006-1012.

Monroe, J ., Santini, J. and Latin, R., 1997. A model deﬁning the relationship between
temperature and leaf wetness duration, and infection of watermelon by

Colletotrichum orbiculare. Plant Disease, 81: 739-742.

MPIC, 2000. Michigan Potato Pest Survey, Michigan Potato Industry Commission
Potato Growers of Michigan.

92

Nelson, M.R., Felix-Gastelum, R., Orum, T.V., Stowell, L]. and Myers, DE, 1994.
Geographic information systems and geostatistics in the design and validataion of
regional plant virus management programs. Phytopathology, 84(9): 898-905.

NPC, 1999. Potato Statistical Yearbook, National Potato Council, Englewood, CO.
NPC, 2001. Potato Statistical Yearbook, National Potato Council, Englewood, CO.

Nutter, F.W., Jr and MacHardy, W.E., 1980. Selection of components for a potato late
blight forecasting and fungicidal control program. Plant disease, 64: 1103-1105.

Olanya, O. and Campbell, C., 1990. Analysis of epidemics of Leptosphaerulina leaf spots
on alfalfa and white clover in time and space. Phytopathology, 80: 1341-47.

Orum, T.V., Bigelow, D.M., Nelson, M.R., Howell, DR. and Cotty, P.J., 1997. Spatial
and temporal pattern of Aspergillusﬂavus strain composition and propagule
density in Yuma County, Arizona, soils. Plant Disease, 81(8): 911-916.

Penrose, L. and Nicol, H., 1996. aspects of microclimate variation within apple tree
canopies and between sites in relation to potential Venturia inaequalis infection.
New Zealand Journal of Crop and Horticultural Science, 24: 259-66.

Raposo, R., Wilks, D8. and Fry, W.E., 1993. Evaluation of potato late blight forecasts
modified to include weather forecasts: a simulation analysis. Phytopathology,
83(1): 103-108.

Ristaino, J. and Gumpertz, M., 2000. New frontiers in the study of dispersal and spatial
analysis of epidemics caused by species in the genus Phytophthora. Annual
Review of Phytopathology, 38: 541-76.

Robinson, P., 2000. Temporal trends in United States dew point temperatures.
International Journal of Climatology, 20: 985-1002.

Rossi, R.E., Mulla, D.J., Joumel, AG. and Franz, EH, 1992. Geostatistical tools for
modeling and interpreting ecological spatial dependence. Ecological Monographs,
62(2): 277-314.

SAS-Institute, 1999. Statistical Analysis Software, Cary, NC.

Sen, P., 1968. Estimates of the regression coefﬁcient based on Kendall’s tau. Journal of
the American Statistical Association, 63: 1379-1389.

Shtienberg, D. and Elad, Y., 1997. Incorporation of weather forecasting in integrated,
biological-chemical management of Botrytis cinerea. Phytopathology, 87: 332-
340.

93

Shtienberg, D. and Fry, W.E., 1990a. Field and computer simulation evluation of spray-
scheduling methods for control of early and late blight of potato. Phytopathology,
80: 772-777.

Shtienberg, D. and Fry, W.E., 1990b. Quantitative analysis of host resistance, fungicide
and weather effects on potato early and late blight using computer simulation
models. American Potato Journal, 67: 277—286.

SPSS-Inc, 1997. SigmaStat, Chicago, IL.

Thompson, N., 1981. The duration of leaf wetness. Meteorological Magazine, 110(1302):
1-12.

Twengstrom, E., Sigvald, R., Svensson, C. and Yuen, J ., 1998. Forecasting Sclerotinia
stem rot in spring sown oilseed rape. Crop Protection, 17(5): 405-411.

Vincelli, RC. and Lorbeer, J .W., 1989. BLIGHT-ALERT:a weather-based predictive
system for timing fungicide applications on onion before infection periods of
Botrytis squamosa. Phytopathology, 79(4): 493-498.

Wallin, J .R., 1962. Summary of recent progress in predicting late blight epidemics in
United States and Canada. American Potato Journal, 39: 306-312.

Wallin, J .R. and Schuster, M.L., 1960. Forcasting potato late blight in western Nebraska.
' Plant Disease Reporter, 44(12): 896-900.

Wilks, D.S., Pitt, RE. and Pick, G.W., 1993. Modeling optimal alfalfa harvest scheduling
using short-range weather forecasts. Agricultural Systems, 42(3): 277-305.

Xu, X. and Ridout, M., 1998. Effects of initial epidemic conditions, sporulation rate, and
spore dispersal gradient on the spatio-temporal dynamics of plant disease
epidemics. Phytopathology, 88: 1000-12.

Xu, X. and Ridout, M., 2000. Stochastic simulation of the spread of race-speciﬁc and
race-nonspeciﬁc aerial fungal pathogens in cultivar mixtures. Plant Pathology, 49:
207-218.

Zehr, E. and Shepard, D., 1996. Bacterial spot of peach as inﬂuenced by wter congestion,
leaf wetness duration, and temperature. Plant Disease, 80: 339-341.

Zwankhuizen, M.J., Govers, F. and Zadoks, J .C., 1998. Development of potato late blight

epidemics: disease foci, disease gradients, and infection sources. Phytopathology,
88(8): 754-763.

94

'-

   

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

lllllllllﬂllllllllilllllﬂllllllll