Study on the influence of weather periods on the occurrence of leaf rust and powdery mildew in winter wheat using an interval-based correlation approach

Untersuchung des Einflusses der Witterung auf das Auftreten von Braunrost und Echtem Mehltau an Winterweizen mit Hilfe eines intervallbasierten Korrelationsansatzes

Material and methods

Because the maximum infestation level of fungal diseases is the best indicator of disease-related damage, only data from the beginning of anthesis until early ripening (Feekes 16 or BBCH 60 to 70) was included in the analyses. If more than one measurement was performed during anthesis, the value from the last monitoring date was used. The mean monitoring date for leaf rust (Fig. 2) and powdery mildew (not shown) was June 16^th to 17^th.

Fig. 2. Monitoring day as “day of the year” (1976 to 2010) for winter wheat at monitoring sites for leaf rust: The mean value (red dot), median (crossbar in box), inter­quartile range (box height), minimum and maximum (whiskers) are shown. Overall mean values are shown in the upper left box.

Monitoring sites (Fig. 1) were connected with weather data from the corresponding weather stations using Thiessen polygons calculated with the ArcGIS software package. Briefly, the monitoring sites were linked to weather data from the nearest weather station by calculating Voronoi fields through inverse distance weighting to interpolate the climatic data between weather stations (Shepard, 1968). Because monitoring sites from 1976 to 1990 could not be linked to municipalities, district capitals of the former German Democratic Republic were taken as substitute monitoring sites. From 1991 to 2010, the municipalities linked to the monitoring sites were known, and the capitals of the municipalities were defined as the monitoring sites.

Fig. 1. Monitoring sites for leaf rust and powdery mil­dew on winter wheat and weather stations with corresponding Thiessen polygons in Saxony-Anhalt between 1976 and 2010.

Time intervals with the highest correlation coefficients were calculated using the “Window Pane” algorithm proposed by Coakley and colleagues (Coakley et al., 1988; Coakley, 1989). In this study, the algorithm was applied with time intervals ranging from day 1 to day 300 before disease monitoring, and window lengths of 5 to 300 days. The results were presented as correlograms (Goldwin, 1982), which are perfectly suited for presenting the results of the “Window Pane” algorithm in its entirety.

In contrast to the aforementioned publications describing the “Window Pane” and Goldwin’s correlogram methods, Kendall correlation coefficients were calculated to derive the correlation matrices. Moreover, Kendall’s correlation coefficient was used to account for non-normality of the pathogen data, and non-parametric tests were performed to derive distributionally independent results (Hartung, 2009).

Results

Fig. 3. Correlograms of significant correlation (α = 0.05; Kendall’s coefficient) between a) mean temperature, b) precipitation sum, c) sunshine duration, d) relative humidity, and e) wind speed and leaf rust severity on winter wheat in Saxony-Anhalt from 1976 to 2010. Legend elements represent the median of the in­terval of correlation coefficients. Day zero represents the monitoring day.

Fig. 4. Correlograms of significant correlation (α = 0.05; Kendall’s coefficient) between a) number of precipita­tion days, b) number of freezing days and c) number of days with snowfall and leaf rust severity on winter wheat in Saxony-Anhalt from 1976 to 2010. Legend elements represent the median of the interval of cor­relation coefficients. Day zero represents the monitoring day.

Correlations between precipitation sums and leaf rust infestation showed rather short- to medium-term relationships (Fig. 3b). Positive correlations were observed from days 90 to 150 (mid-March to mid-January), around days 230 and 295, and in the first 20 to 40 days before monitoring. Short-term negative correlations were found around 60, 170 and 210 days before field monitoring.

Our correlation analyses of sunshine duration and leaf rust infestation revealed some interesting findings (Fig. 3c). First, sunshine duration had a significantly positive influence on leaf rust severity when averaged over 300 days. In contrast to Fig. 3a, correlations on shorter time scales were higher than those for longer time periods. Furthermore, there were significant positive correlations from days 30 to 60 (mid-May to mid-April) and 160 to 270 before monitoring (end of January to mid-September). No significant positive correlations were found from early spring to mid-winter (day 80 to 150), and significant negative correlations were observed between days 90 and 120 (March to February). There were no significant correlations during the first 30 days before disease monitoring. Between days 290 and 300, significant negative correlations with leaf rust infestation were observed.

Correlations between leaf rust infestation and relative humidity showed mainly short-term relationships, except during the period from day 140 to 270 (end of January to mid-September), when significant negative mid- to long-term correlations were observed (Fig. 3d). From days 30 to 60 (mid-May to mid-April), another noteworthy negative correlated period was observed.

Analysis of correlations between wind speed and leaf rust infestation showed nearly exclusively negative relationships (Fig. 3e). Significant negative correlations were found for mean wind speed in autumn (days 240 to 270 before monitoring), late autumn and winter (days 150 to 180), and during the first 60 days before monitoring (spring to early summer).

Analysis of correlations between leaf rust infestation and the number of days with precipitation (Fig. 4a) showed significant positive correlations from day 90 to day 150 (March to January), day 190 to day 240 (end of November to mid-October), and day 290 to day 300. Significant negative correlations occurred during the interval from days 35 to 60 and around day 170 before monitoring.

The influence of freezing days on leaf rust infestation in spring was mainly negative (Fig. 4b), especially during the winter to early spring, from the beginning of November (day 225) until mid-March (day 90). Our analysis of correlations between the number of days with snowfall and leaf rust occurrence revealed the most negative correlation coefficients in the interval between days 90 and 210 (mid-March to mid-November), and further negative correlations during shorter time periods around day 210 (Fig. 4c).

Fig. 5. Correlograms of significant correlation (α = 0.05; Kendall’s coefficient) between a) mean temperature, b) minimum temperature, c) precipitation sum, d) relative humidity, e) sunshine duration, and f) maxi­mum temperature and powdery mildew severity on winter wheat in Saxony-Anhalt from 1976 to 2010. Legend elements represent the median of the interval of correlation coefficients. Day zero represents the monitoring day.

Fig. 6. Correlograms of significant correlation (α = 0.05; Kendall’s coefficient) between a) number of precipita­tion days, b) number of freezing days, and c) number of days with temperatures between 17 and 23°C and powdery mildew severity on winter wheat in Saxony-Anhalt from 1976 to 2010. Legend elements repre­sent the median of the interval of correlation coefficients. Day zero represents the monitoring day.

Regarding relationships between minimum temperatures and powdery mildew infestation levels, there were significant positive correlations in the period from day 70 to day 110 (beginning of April until end of February) and during the winter months (February to December) (Fig. 5b). Conversely, significant negative correlations were observed from mid-November to mid-October.

Two distinct patterns could be seen when looking at the correlograms between precipitation sums and powdery mildew occurrence (Fig. 5c). Firstly, precipitation sums from February to September and from days 20 to 40 before monitoring were negatively correlated with disease severity. Secondly, time windows around 300 days before monitoring and during the first 10 to 15 days before monitoring showed significant positive correlations.

Analysis of correlations between relative humidity and disease infestation level showed significant negative correlations from days 90 to 230 (mid-March to the beginning of November) and positive relationships in April and during the first 10 to 15 days before disease monitoring.

There were significant negative correlations between sunshine duration and disease occurrence for the time window from December until the monitoring date (Fig. 5e). The strongest correlations were observed in the first 10 to 15 days before monitoring.

The correlogram for maximum temperatures and severity of B. graminis f. sp. tritici showed significant negative correlations in late summer (end of August) and from the beginning of February until monitoring (Fig. 5f). Between the end of January and the beginning of December, there were significant positive correlations during interval lengths of 10 to 20 days.

In late spring (days 20 to 60) and especially in autumn (days 210 to 270), the number of days with precipitation had a significantly negative effect on powdery mildew occurrence at anthesis (Fig. 6a). Significant positive correlations were observed during the first 20 days before monitoring.

Regarding correlations between freezing days and powdery mildew occurrence (Fig. 6b), significantly positive correlations were observed from mid-October to the end of November (days 240 to 200).

Furthermore, we observed significant positive correlations between the number of days with mean temperatures of 17 to 23°C and powdery mildew occurrence for periods in beginning April, from days 70 to 80 and around mid-October. There were significant negative correlations around 15 and 300 days before monitoring (Fig. 6c).

Analyses of daily weather variables provide evidence of important time intervals during the growing period. However, the identification of periods of greatest influence is not possible using daily weather data because the explanatory power of correlations between daily weather variables and disease data is very limited and may contain many spurious relationships. Furthermore, there is no evidence that daily weather a few months before the monitoring date for a fungal disease could possibly influence disease severity similarly every year. How often days with favorable or unfavorable conditions occurred in a specific time interval and how favorable or unfavorable those conditions were is much more important (Burleigh et al., 1972a). Therefore, daily weather data must be accumulated on different time scales to gain further insight into disease-weather relationships.

The “Window Pane” algorithm produced reasonably good results and is a useful tool for investigating disease-weather relationships because all time intervals that could possibly explain variations in disease levels can be evaluated rather than only a few selected intervals. This method can be used for preliminary variable selection before regression modeling when a large number of possible explanatory variables must be processed. The explanatory variables must then be condensed to those explaining the biggest part of variability of the predictand. Calvero Jr. et al. (1996), Pietravalle et al. (2003) and Te Beest et al. (2008, 2009) mentioned only the “best” correlations for each variable but provided no overview of the calculated results.

The Goldwin correlogram (Goldwin, 1982) is an ex­cellent method for displaying the results of the variable selection process, obtaining an impression of all significantly correlated time windows, and comprehending the basis on which some correlations are selected and others dropped. The use of correlograms allowed us to show not only significance or non-significance, but also the strength of correlation between infestation data and weather parameters for over 350,000 correlations analyzed per disease.

However, the suitability of different correlation coef­ficients to analyze data with heavily skewed probability distributions must be further investigated to improve the results. Partial correlation coefficients could produce better variable selection results than simple correlation coefficients, but are much more costly in terms of computing capacity and the expenditure of time. Furthermore, proper interpretation of the correlograms is not possible without a solid background in phytopathology. Definition of the minimum window length to extract the “best” correlations requires a deep understanding of the subject matter as well. Therefore, new statistical models must be developed to gain further insight into disease-weather relationships. These models should make it possible to run climate change scenarios for future leaf rust and powdery mildew development and to perform climate impact assessments for future agricultural production and plant disease management based on these plant pathogen scenarios.

The clearly positive influence of higher temperatures on all developmental stages of leaf rust on winter wheat and on disease severity at anthesis found in our analyses agree with the findings of laboratory studies (Chester, 1946; Hassebrauk, 1959; Zadoks, 1965) and statistical analyses of monitoring data (Chester, 1946; Hogg, 1969; Burleigh et al., 1972a, b; Daamen et al., 1992; Jahn et al., 1996; Eversmeyer and Kramer, 1998; Moschini and Pérez, 1999; Wiik and Ewaldz, 2009). Especially in winter and spring, temperatures play a major role in determining the amount of urediniospores managing to overwinter and increasing the amount of initial inoculum available for epidemic development on host plants in spring (Burleigh et al., 1969; Eversmeyer and Kramer, 1994, 1995, 1996, 1998).

Higher spring temperatures can accelerate uredia and urediniospore development and shorten latency and incubation times, thus increasing the number of leaf rust generations until anthesis of wheat. Lower temperatures slow down the development of leaf rust and decrease the amount of disease cycles until anthesis (Roelfs and Bushnell, 1985; Eversmeyer and Kramer, 1996). The identified correlations show the importance of air temperatures during the transition from autumn to winter and from winter to spring, when temperatures are at the lower limit of leaf rust development (Chester, 1946).

The influence of precipitation on the development of leaf rust on winter wheat has already been the topic of many studies (Daamen et al., 1992; Heitefuss et al., 1993; Wiik and Ewaldz, 2009). As an important source of humidity, precipitation is essential for urediospore germination and infection. This becomes obvious when looking at the positive correlation periods in August, October and May. The findings in August agree with those of Wiik and Ewaldz (2009). In late summer, humidity is an important prerequisite for leaf rust survival throughout the crop-free period as it provides moisture for volunteers. Furthermore, precipitation in late summer results in lower insolation, which makes it easier for leaf rust to over-summer (Zadoks, 1965; Eversmeyer and Kramer, 1998).

In autumn, leaf rust infects newly sown winter wheat, and its population size and spatial distribution can expand if weather conditions are good enough. Sufficient precipitation is needed as a source of moisture for leaf wetness, a major determinant of optimal infection and germination. These conditions seem similarly important in spring, especially in May, when leaf rust is in the crucial phase that determines disease severity during anthesis. The wetter and warmer the conditions during this time, the shorter the latency and infection periods and the more generations produced (Hogg, 1969). The above-mentioned findings contradict those of Jahn et al. (1996), who found that leaf rust infestation on winter wheat increases under drier future conditions.

The positive relationships between precipitation in January and February and leaf rust severity cannot be completely explained as a result of the presence of the necessary humidity, but rather with the isolation effect of snowfall at lower temperatures in contrast to freezing temperatures without snowfall. However, this isolation effect was not seen in the correlogram for snowfall days. Our results showed the exact opposite of an isolation effect by snow cover. Snow-free periods during winter are needed for re-infection of host material and are thus important for leaf rust survival (Eversmeyer and Kramer, 1998). Caution is advised when interpreting findings on the influence of snow cover on urediniospores because the results only show correlograms for snowfall and disease severity. The fact that no snow is falling does not necessarily mean that there is no snow cover. Additionally, the calculated variable snowfall seem to be dominated by low temperatures, and the expected precipitation signal is mainly absent. Further studies are needed to determine whether this method for calculating days with snowfall makes sense or must be changed.

The effects of sunshine duration and radiation on the development of fungal pathogens are known (De Vallavieille-Pope et al., 2002). This knowledge is also valid for leaf rust on winter wheat where urediniospores are exposed to intense solar radiation during transport (Sache, 2000), germination, infection, reproduction and sporulation. As expected, correlations between sunlight hours and leaf rust severity show patterns similar to those between disease severity and mean temperatures over long periods of time. During winter, the correlation structures for sunlight duration are contrary to those for mean temperature. This can be explained by the decoupling of radiation and temperature during winter months. Sunny days and high radiation are common during winter weather situations with persistent high-pressure systems, which generally bring relatively cold rather than warm temperatures to Central Europe. However, it is surprising that sunshine duration shows no significant correlations regarding the conditions in March and early April. The negative correlated period around day 290 corresponds with the findings of the correlograms for precipitation. If there were a positive correlation for precipitation sums in this interval, sunshine duration would naturally show a negative relationship because more precipitation would mean more cloudiness and less sunshine hours, and that, in turn, would increase the possibility of leaf rust urediniospores over-summering during crop-free periods by using a “green bridge” during this time (Roelfs and Bushnell, 1985).

Wind speed is often used to explain how exogenous inoculum is transported to regions, thus strengthening leaf rust epidemics or making them possible (Hirst and Stedman, 1967; Sache, 2000). Contrary to the findings in literature, our correlogram for wind speed shows mainly negative correlations. This suggests that wind speed can be too high for urediniospores to adhere to leaf surfaces before appressorium development, but it is unlikely that this effect could hinder the development of leaf rust for a long period of time. Thus, the strongest effect of wind speed on leaf rust severity might be accelerated drying of leaf surfaces at higher wind speeds. This might lower leaf wetness below the levels required for infection and germination of urediniospores (Stuckey and Zadoks, 1989). Especially in late summer, early autumn, at the time of winter wheat sowing, and in spring, leaf wetness accelerates the development of leaf rust. Therefore, long periods of leaf dryness can result in massive obstruction or even long-term interruption of the developmental cycle of P. triticina (De Vallavieille-Pope et al., 1995).

Powdery mildew is likely to capitalize from warmer temperatures in winter. Mild winters allow more conidia to survive the winter, resulting in more initial inoculum at the beginning of spring (Boland et al., 2004). Our findings support this theory as they show significant positive relationships between minimum temperatures and powdery mildew severity at the beginning of February until early December. Minimum temperatures seemed to be around the lower limit for powdery mildew development from March to December because minimum temperatures during this time frame were positively correlated with disease severity.

The negative correlations between mid-October and late November found in this study are not discussed in literature, and no biological process explaining this phenomenon can be found (temperatures in late autumn are usually at the lower limit for disease development, so higher temperatures cannot impede the development of powdery mildew). The same pattern can also be seen as positive correlations in the correlogram for freezing days and disease severity.

Temperature sensitivity of powdery mildew was observed during the three months before the monitoring date and in late August of the preceding year. This effect was also documented by Aust (1981a) and Stephan (1957). It seems that before anthesis and during late summer, when powdery mildew conidia are struggling to survive and capitalize on the “green bridge”, temperatures can be too high for optimal development.

An interesting difference in patterns was seen in March and April between correlograms for minimum temperatures and disease severity and maximum temperatures and disease severity. In the correlograms for these months, minimum temperatures seemed to be around the lower limit, as indicated by the positive correlations, and maximum temperatures seemed to be around the upper limit, as indicated by negative correlations. This suggests that the optimal temperature range for powdery mildew on winter wheat is extremely small. Moreover, this contradicts the findings of Te Beest et al. (2008), who did not find maximum temperatures in spring and early summer approaching the upper temperature limit for powdery mildew. According to their data, the warmer temperatures are from spring onward, the better for powdery mildew development. Stephan (1957) showed that powdery mildews do in fact have a rather small window of optimal temperature for overall development. However, our data suggest that temperatures in March and April are critical for powdery mildew infection, penetration, germination and sporulation. This was also emphasized by Morgounov et al. (2011). It seems to be analog to Chester’s findings on the “critical month” of leaf rust development (Chester, 1946). Warmer conditions during this period can give powdery mildew a good start into the new vegetation season by shortening latency and infection periods and by amplifying the number of generations until anthesis. This was observed in the analysis of correlations between minimum temperatures and disease severity and of correlations between disease severity and the number of days with temperatures between 17 and 23°C, which are approximately the optimal temperatures for powdery mildew development (Prigge et al., 2005; Yarwood et al., 1954).

The negative influence of very high temperatures on powdery mildew severity in the weeks before the monitoring date and in late summer goes hand in hand with high radiation periods with a significantly negative impact on disease severity. This corresponds to the findings of Spencer (1978) and Martin (1975), who found that high insolation conditions and therefore higher leaf surface temperatures (Te Beest et al., 2008) can hamper germination of powdery mildew conidia.

The correlogram for precipitation sums and powdery mildew severity displays two patterns. Firstly, precipitation averaged over periods of at least one month during the whole time frame considered had a significantly negative effect on disease severity, especially in autumn. Secondly, precipitation sums averaged over the first 20 days before monitoring showed significant positive correlations with disease severity.

The first pattern is in agreement with the findings of Spencer (1978) and Boughey (1949), who found that powdery mildew is more dangerous under drier conditions. Laboratory experiments have shown that the conidia of powdery mildew have a higher moisture content than, for example, urediniospores of leaf rust, and that they do not have to rely on precipitation events for moisture (Yarwood et al., 1954). The second pattern shows that the additional moisture content is not sufficient at higher temperatures in summer. Then, precipitation events must provide the moisture needed, in particular, for infection and germination of conidia. High relative humidity is a well known prerequisite for those developmental processes (Hassebrauk, 1959; Prigge et al., 2005; Friedrich, 1994; Merchán Vargas, 1984).