- Email: [email protected]
Predicting potential winter wheat yield losses caused by multiple disease systems and climatic conditions
Crop Protection 99 (2017) 17e25
Contents lists available at ScienceDirect
Crop Protection journal homepage: www.elsevier.com/locate/cropro
Predicting potential winter wheat yield losses caused by multiple disease systems and climatic conditions a b Radivoje Jevti c a, *, Vesna Zupunski , Mirjana Lalosevi c a, Ljubica Zupunski a b
Institute of Field and Vegetable Crops, Maksima Gorkog 30, 21000 Novi Sad, Serbia University of Novi Sad, Faculty of Technical Sciences, Trg Dositeja Obradovica 6, 21000 Novi Sad, Serbia
a r t i c l e i n f o
a b s t r a c t
Article history: Received 5 December 2016 Received in revised form 11 April 2017 Accepted 3 May 2017
Yield losses in field crops are most commonly predicted by using regression models that include either biotic or abiotic factors as predictor variables. Knowing that yield loss is a complex trait, the potential capability of regression models for predicting yield losses by using models containing both biotic and abiotic factors as predictors were estimated in this study. Biotic factors considered in regression models were: leaf rust, powdery mildew, septoria tritici blotch and tan spot occurrence on the varieties Barbee and Durumko known to have various degrees of susceptibility to obligate parasites and leaf blotch diseases. Among abiotic factors, monthly averages of temperature, relative humidity and total rainfall taken from November to June for growing seasons 2006e2013 were used as predictors. In 2014, yellow rust became the predominant pathogen over leaf rust, thus 2014 and 2015 were excluded from regression models and analyzed separately. Since a high correlation was found between abiotic and biotic factors, partial least squares regression, stepwise regression and best subsets regression were applied. Best subsets regression revealed that models consisted of both biotic and abiotic factors were more precise in estimating regression coefficients and predicting future responses. The potential yield loss predictions, conducted using these models, were regressed with actual yield losses, and high coefficients of determination (R2 ¼ 79% for Barbee; and R2 ¼ 63% for Durumko) were obtained. It was also evident that using more predictors in regression models does not necessarily mean that the model would have a higher potential in making yield loss predictions. This study confirms that the relationship between a disease scoring scale and yield loss is not straightforward and higher potentials for yield loss predictions were given due to the regression models using abiotic and biotic predictor variables. © 2017 Elsevier Ltd. All rights reserved.
Keywords: Puccinia Blumeria Zymoseptoria Regression models Yield loss Winter wheat
1. Introduction Wheat (mainly common/soft wheat Triticum aestivum but also durum/hard wheat T. turgidum) is the world's largest crop in terms of harvested area and ranks third in the global annual production of commodities (FAO, 2012). Significant concerns have been raised by the scientific community about the impacts of climate change on future yield potentials of wheat. However, in spite of the fact that climatic changes have been emphasized to have impact on wheat yield and quality, not only directly but also due to the interactions with biotic factors, the effects of biotic factors on yield losses have been neglected in
* Corresponding author. E-mail addresses: [email protected] (R. Jevti c), vesna.zupunski@ ifvcns.ns.ac.rs (V. Zupunski), [email protected] (M. Lalosevi c), [email protected] (L. Zupunski). http://dx.doi.org/10.1016/j.cropro.2017.05.005 0261-2194/© 2017 Elsevier Ltd. All rights reserved.
recently reported studies (White et al., 2011; Juroszek and Von Tiedemann, 2013). Among the economically most important diseases affecting winter wheat are obligate parasites (Blumeria graminis f. sp. tritici, Puccinia graminis f. sp. tritici, Puccinia triticina, Puccina striiformis f. sp. tritici) and crop residue-borne necrotrophic pathogens (Pyrenophora tritici-repentis, Zymoseptoria tritici, Parastagonospora nodorum, Cochliobolus sativus, Fusarium species). Although chemical treatment is a very powerful disease-control tool, followed by increases in yield, the main imperative of integrated pest management is environmental protection and reduced fungicide input. Thus, many efforts have been directed on determining damage thresholds and developing mathematical models that can be used to forecast yield losses caused by pathogenic infection. Typically, these experiments are conducted in a few locations during a two or three - year period, but only some of them raise the question about disease dynamics and yield over longer time periods (Wiik, 2009).
18
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
In literature, adversarial reports appear regarding correlation between diseased leaf area or any disease scoring scale and yield. For instance, some authors reported linear regression as suitable for describing relationship between disease rating scale and yield loss (Wegulo et al., 2009; Green et al., 2014; Budka et al., 2015), whereas there are also reports that the relationship between the two is not straightforward (Duveiller et al., 2007). The objective of this study was to evaluate the potential of regression models for predicting yield losses in winter wheat if both biotic (leaf rust, powdery mildew, septoria tritici blotch, and tan spot) and abiotic factors (climatic conditions from November to June) are subjected to the same regression model. The data collected from 2006 to 2015 were analyzed and characterized in terms of agro-ecological conditions of Serbia. 2. Materials and methods Data were obtained from fungicide efficacy trials which were conducted in the locality of Rimski San cevi (Vojvodina, north province of Serbia) under the direction of the Institute of Field and Vegetable Crops, Novi Sad, Serbia, over the period of 2006e2015 using soft wheat variety Barbee (Triticum aestivum ssp. compactum), and hard wheat variety Durumko (Triticum turgidum subsp. durum). The Barbee variety has shown increased susceptibility to obligate parasites (Blumeria and Puccinia) while Durumko showed increased susceptibility to leaf blotch diseases (LBDs) such as Pyrenophora tritici-repentis, Zymoseptoria tritici, and Phaeosphaeria nodorum in agro-ecological conditions of Serbia. 2.1. Field trial Field trials were set up under naturally occurring inoculum and were arranged in a randomized block design comprising four replicates. The plot size of each replicate was 10 m2. A trial usually included 10 fungicide-sprayed and non-sprayed check treatments. Fungicides were applied at growth stage BBCH 36e37 (flag leaf just visible, rolled) and BBCH 51e59 (inflorescence emergence, heading). Different types of active ingredients such as amides, aromatics, azoles, benzimidazoles, morpholines, oxazoles, strobilurins, pyrazoles and pyridines were applied with recommended dosage rates using calibrated field crop sprayers with fan nozzles, at 300 kPa pressure and 200 L of water per hectare. Mean sowing date for winter wheat was 20 October (optimal time of sowing) and the mean harvest date was 30 June (range 25 June- 07 July). 2.2. Disease assessment Assessments of leaf disease severity were made at the growth stage 71e73 BBCH (kernel watery; early milk), known to be highly related to yield (Wegulo et al., 2009). Disease severity for leaf rust and powdery mildew was assessed using modified Cobb's scale (Peterson et al., 1948; Corazza and Islongo, 1987). Disease severity of septoria tritici blotch and tan spot were assessed using the disease rating keys devised by James 1971. The disease indices (%) of leaf rust, powdery mildew, septoria tritici blotch and tan spot were calculated by taking into consideration disease incidence and average disease severity (Cao et al., 2014). 2.3. Yield Yield was measured for each plot after harvest at 15% water content. Yield loss (%) was determined as yield reduction in untreated plots compared with yield response to fungicide treatment which provided the best control of wheat diseases (Eq. (1)).
Yð%Þ ¼ ððY1 Y2 Þ=Y1 Þ 100
(1)
Y1 - grain yield of fungicide treatment for the best wheat disease control. Y2 - grain yield of the non - sprayed check treatment.
2.4. Predictor variables for regression models In regression models, the biotic factors regarded as potential predictive variables were disease indices of leaf rust, powdery mildew, septoria tritici blotch and tan spot. On the other hand, abiotic factors used as predictors comprised monthly averages of: temperatures, relative humidity and total rainfall taken from November to June for growing seasons 2006e2013 (http://www. hidmet.gov.rs/). The shift in predominant rust pathogen occurred in 2014, thus a period of 2014 and 2015 was excluded from regression models and analyzed separately. 2.5. Statistical methods The effects of year, variety and fungicide treatment on yield were examined by analysis of variance (ANOVA). Further, multivariate regression models were used to estimate relationship between disease indices, abiotic factors and yield losses. Knowing that abiotic and biotic factors are correlated not just to the yield loss but also with each other (multicollinearity), partial least squares regression, stepwise regression and best subsets regression were applied to make predictions of yield losses in the Barbee and Durumko varieties. Multicollinearity is problematic because it can increase the variance of the regression coefficients, making it difficult to evaluate the individual impact that each of the correlated predictors has on the response. Partial least squares (PLS) is a biased regression procedure that reduces the number of predictors and extracts a set of components that describes maximum correlation among the predictors and response variables. The technique used is similar to principal component analysis because it gives the option of leave-one-out cross-validation, which is used to maximize the model's predictive ability. In addition, the method that can also be used to analyze correlated predictors is the stepwise regression model. It is used in the exploratory stages of model building to identify a useful subset of predictors. The process systematically adds the most significant variable or removes the least significant variable during each step until it identifies variables that explain the maximum variation in yield loss. Best subsets regression was performed to identify the best-fitting regression models with predictors of choice as well as to compare regression models obtained by PLS and stepwise regression. The general approach was to select the smallest subset that fulfills certain statistical criteria. The reason for using a subset of variables instead of a complete set is because the subset model might actually estimate the regression coefficients and predict future responses with smaller variance than the full model using all predictors. Regression models were followed with coefficient of determination (R2), coefficient of prediction (R2pred), variance inflation factor (VIF) and Mallows' Cp. Coefficient of determination (R2) is the percentage of variation in the response that is explained by the model. Coefficient of prediction (R2pred) determines how well the model predicts the response for new observations. VIF indicates how much the variance of an estimated regression coefficient increases if predictors are correlated. The VIFs will all be 1 if there is no correlation between factors. Mallows' Cp is used for
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
making comparisons between multiple regression models. It compares the precision and bias of the full model to models with a subset of predictors and strikes an important balance with the number of predictors in the model. A Mallows' Cp should be small and close to the number of predictors in the model plus the constant (p) value which is a number of explanatory variables. If Mallows' Cp is close to the number of predictors plus the constant, the model is relatively unbiased in estimating the true regression coefficients and predicting future responses. We determined the correlation coefficients between climatic factors, yield losses and disease indices using Spearman's coefficient of correlation. Minitab 17 (trial version) was used for entire analysis.
3. Results and discussion The yield of treated and untreated plots of the Barbee and Durumko varieties did not change linearly over the period of ten years. Average yield of variety Barbee was 5.1 t/ha on treated and 3.6 t/ha on untreated plots with an average yield loss of 30%. Average yield loss of variety Durumko was 10%, with 7.3 t/ha on treated and 6.5 t/ha on untreated plots. Yield of the Barbee and Durumko varieties differed significantly (P ¼ 0.000) and were influenced by year (P ¼ 0.000) and applied fungicide treatments (P ¼ 0.000). There were statistically significant interactions between year and cultivar (P ¼ 0.000); year and fungicide application (P ¼ 0.006) and cultivar and fungicide application (P ¼ 0.023). Yield losses varied considerably during the 2006e2015 period, and the Barbee variety suffered a much higher percentage yield loss (up to 45%) than the variety Durumko (up to 25%) (Fig. 1). The Barbee variety is a bearded, semi-dwarf soft white club winter wheat and is known for its susceptibility on powdery mildew and rusts of wheat. Because of its high susceptibility to obligate parasites it was used as susceptible check in field trails in many countries (Robe et al., 1996; Lorys et al., 2000; Chantret et al., 2001). Barbee variety has lower yield potentials (5.1 t/ha) compared with the Durumko variety (7.3 t/ha) and suffered higher average yield loss (30%). On average, disease index of leaf rust and powdery mildew on the Barbee variety for the period was 22% and 21%, respectively. However, in 2009 disease pressure of leaf rust (68%) and powdery mildew (30%) was very high causing yield loss of 44%. The average disease index of septoria tritici blotch on the Barbee variety was 7% for a ten-year period and it was of minor importance on yield losses comparing with influence of obligate parasites. According to Zhang et al. (2006), the occurrence of septoria tritici blotch, when combined with leaf rust and powdery mildew on the
Fig. 1. Yield losses of Barbee and Durumko varieties in the period 2006e2015.
19
same host, is highly influenced by variety itself; thus, five potential disease profiles were introduced based on different responses during the examination of the varieties produced in multiple disease systems. The Durumko variety is durum wheat which had increased susceptibility to leaf blotch diseases in agro-ecological conditions of Serbia, and suffered average yield losses of 10%. The yield loss of Durumko variety can be explained by low disease pressure of obligate parasites and septoria tritici blotch occurrence with an average disease index of 17%. The exception was in 2006 when septoria tritici blotch index reached 51% causing a yield loss of 20%. The low disease pressure of an obligate parasite and high disease pressure of septoria tritici blotch on Durumko variety can be explained by adaptation of predominant pathotypes on bread and durum wheat. It was reported that predominant and highly virulent populations of P. triticina on bread wheat are different and often avirulent on durum wheat and vice versa (Huerta-Espino and Roelfs, 1992). 3.1. Prediction of potential yield losses in Barbee and Durumko varieties using biotic and abiotic factors as predictors The potentials of regression models to predict yield losses using biotic together with abiotic factors as predictors were examined. Since climatic factors are highly correlated with disease indices, regression models such as partial least squares regression, stepwise regression and best subsets regression were applied. To estimate the most influencing factors on yield loss of varieties Barbee and Durumko, climatic factors and disease indices were subjected to partial least squares regression. The two-component regression model accounted for 71% of variation in yield loss of Barbee variety and adding more components did not significantly increase R2 (Fig. 2A). The partial least-squares method indicated that using only one predictor contributes in variation in yield loss of Durumko variety with an R2 of 57% (Fig. 2B). Among abiotic and biotic predictors, leaf rust and powdery mildew indices together with total rainfall in March had the highest absolute values of standardized coefficients and as a result these were considered to have the greatest impact on yield loss of the Barbee variety (Fig. 3). However, it should be noted that the value of standardized coefficients does not strictly imply a negative or positive correlation with yield loss because of multicollinearity. The factors which were considered to be highly influencing on yield loss of Durumko variety were temperatures in January and June, relative humidity in December and septoria tritici blotch index (Fig. 3). Stepwise regression, another model used to address multicollinearity, identified a subset of predictors based on their statistical significance. In stepwise regression, the leaf rust and powdery mildew indices as well as temperatures in April, proved to be the strongest predictors of yield loss in the Barbee variety, giving a coefficient of determination (R2) of 70% and coefficient of prediction (R2pred) of 54%. The stepwise regression model indicated that the greatest impact on yield loss of variety Durumko had septoria tritici blotch index and temperature in June giving the model with (R2) of 63% and (R2pred) of 57%. The most influential predictors on yield losses did not completely coincide with the application of the partial least squares and stepwise regression models on the Barbee or Durumko varieties. To compare regression models and examine how much variation of yield loss will be explained by the maximum R2ecriterion the best subsets regression was applied. The most influential factors determined by stepwise regression were subjected to best subsets regression together with climatic elements which were significantly moderately and highly correlated with yield losses and disease indices at P 0.05 and R2 50% (Table 1).
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
20
Fig. 2. Model analysis for yield loss prediction of Barbee (A) and Durumko (B) varieties obtained by partial least squares regression.
Fig. 3. Standardized coefficients of biotic and abiotic factors influencing yield loss of Barbee and Durumko varieties.
Table 1 The correlation between climatic factors, yield losses and disease indices related with varieties Barbee and Durumko.
Barbee
Leaf rust
December
March
April
May
Total rainfall (mm)
Total rainfall (mm)
Total rainfall (mm)
Temperature ( C)
r ¼ 0.741 (P ¼ 0.000) R2 ¼ 51.4%
r ¼ 0.857 (P ¼ 0.000) R2 ¼ 73.1%
r ¼ 0.570 (P ¼ 0.008) R2 ¼ 68.1%
Powdery mildew Yield loss Durumko
Septoria tritici blotch Yield loss
Climatic elements which were significantly correlated with yield losses and disease indices with R2 50% had higher values of Mallows’ Cp than those with R2 50% (data not shown); thus, they were not considered as the most influencing factors on yield loss of varieties. The best subsets regression revealed that temperature in April, leaf rust index and powdery mildew index formed the model with the smallest Mallows' Cp (2.1) which indicated that this model is relatively unbiased in estimating the true regression coefficients and predicting future yield losses of the Barbee variety (Table 2.)
r ¼ 0.739 (P ¼ 0.000) R2 ¼ 51.4%
The same set of predictors was identified by stepwise regression model giving R2 (70%) and R2pred (54%). The average temperatures in April ranged from 12.3 C to 14.6 C in Serbia's climate conditions and these temperatures are in accordance with the temperature range (12 C to 15 C) reported to be optimal for germination of leaf rust urediniospores (Junk et al., 2016). In addition, the minimal temperatures for leaf rust development starts from 8 C and are related with night temperatures (Junk et al., 2016). The average minimal temperature of 7 C was reached no earlier than April. Total rainfall and relative humidity in April were also significantly
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
21
Table 2 Best subsets regression for predicting yield loss in the Barbee variety.
related with leaf rust indices but were not selected as more significant than temperatures in April since optimal relative humidity > 60% is usually met in the spring in Serbia. The set of predictors identified by partial least squares regression (total rainfall in March, leaf rust index and powdery mildew index) had lower R2, R2pred and higher Mallows’Cp than the set of predictors chosen by stepwise regression. Additionally, best subsets regression indicated that using only biotic factors as predictors for yield loss predictions (in this case leaf rust index) would give lower R2 (55.6%), R2pred (49.7%) and much higher Mallows’Cp (8.8) than using it in combination with abiotic factors (Mallows’Cp decreased to 2.4). A high Mallows' Cp value (8.8) indicated that the model is not precise (has large variance) in estimating the true regression coefficients and predicting future responses. The combination of leaf rust and powdery mildew indices as predictors had high potential for predicting yield losses of the Barbee variety, but still with lower R2, R2pred and higher Mallows’Cp than using them in combination with T in April. It was also evident that using more predictors does not necessarily mean that the model would have a higher potential for making yield loss predictions. The best subsets regression revealed that the best model for yield loss prediction in variety Durumko was comprised of temperature in June and septoria tritici blotch index giving the smallest Mallows’ Cp (2.5) and optimal combination of R2 (63.2%) and R2pred (57.4%) (Table 3). Using only biotic factors as predictors for yield loss prediction (in this case septoria tritici blotch index) would give lower R2 (37.2%), R2pred (29.4%) and much higher Mallows’Cp (20.7) than
Table 3 Best subsets regression for yield loss prediction in variety Durumko.
using it in combination with abiotic factors. Low coefficient of determination (37.2%) obtained after regression of septroria tritici blotch index and yield loss of variety Durumko is in accordance with coefficients of determination of 39% and 44% obtained by Berraies et al. (2014) who estimated grain yield losses caused by septoria tritici blotch using 400 lines of durum wheat in two sowing seasons. The temperature in June was found to be highly influential on yield loss of the Durumko variety since environmental conditions in June are favorable for septoria tritici blotch pycnidia development. The day/night temperatures in June were on average 26 C day/15 C night which were found to be very conducive for pycnidia development causing higher severity of the disease. Chungu et al. (2001) reported that percent pycnidial coverage increased as incubation temperature increased from 15 C day/11 C night to 22 Cday/15 C night, which is in accordance with our study. In addition, the latent period is temperature dependent and decreased as temperature increased (Chungu et al., 2001; Shaw and Royle, 1992). It was also reported that the strongest relationship between disease index and yield includes the period from the end of anthesis (ZGS 69) to milk stage (ZGS 70e73) which coincides with the second half of May and June in Serbia's climate conditions. 3.2. Predicting yield losses of varieties Barbee and Durumko Since it was indicated that T in April, leaf rust index and powdery mildew index formed the best set of predictors for predicting yield loss of the Barbee variety, multiple regression was performed,
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
22
producing the following equation:
equation:
Y ¼ 86:0 4:72x1 5:05x2 þ 5:07x3 þ 0:392x1 x2 0:405x1
Y ¼ 84:0 3:76x1 þ 0:70x2 0:0210x1 x2
x3 þ 0:115x2 x3 þ 0:00766x1 x2 x3
Y e yield loss of variety Barbee x1e T in April x2 e Leaf rust index x3 e Powdery mildew index The P-value (0.000) for the regression model in the analysis of variance showed that the model estimated by the regression procedure is significant at an a-level of 0.05. The P-value for the estimated coefficient of leaf rust index (0.000) and T in April (0.046) indicated that these were significantly correlated to yield loss in the Barbee variety. The P-value for estimated interaction coefficients between T in April and leaf rust (0.046), as well as T in April and powdery mildew index (0.019) showed also significant correlation to yield loss at an a-level of 0.05. The VIFs indicated that regression coefficients were moderately correlated but still well estimated. The percentage of variation in yield loss explained by the model was R2 ¼ 79%, and the possibility for prediction of new observations was R2pred ¼ 61%. An analysis of residuals and a test for the possible occurrence of outliers showed that there were no outliers in the data. The normal probability plot evinced an approximately linear pattern which is consistent with a normal distribution. The plot of residuals versus the fitted values showed that residuals were randomly distributed (Fig. 4). The actual yield losses and predicted yield losses were regressed and high R2 value (79%) was obtained (Fig. 5). To make a prediction of yield losses of the Durumko variety, temperature in June and septoria tritici blotch index were subjected to multiple regression resulting in the following regression
Y eYield loss of variety Durumko x1 e T in June x2 e Septoria tritici blotch index Coefficients of determination (R2) and prediction (R2pred) were 63% and 55%, respectively. The P-value for the regression model (0.000) confirmed model significance at an a-level of 0.05. The Pvalues for the estimated coefficient of septoria tritici blotch index (0.001) and temperature in June (0.000) showed significant correlation with yield loss in the Durumko variety at an a-level of 0.05. The VIFs were less than 1.34, indicating that predictors were well estimated. A residual analysis indicated that residuals were distributed randomly and that there were no outliers in the data. The normal probability plot showed an approximately linear pattern that is consistent with a normal distribution (Fig. 6). The relation between actual yield losses of the Durumko variety and predicted yield losses was fitted to linear regression and high R2 value (63%) was obtained (Fig. 7). 3.3. The shift of predominant rust pathogens in Serbia Monitoring the occurrence of wheat diseases in the last ten years showed a shift in predominant rust pathogen in the wheat growing area. Yellow rust caused by Puccinia striiformis f. sp. tritici was first recognized in Serbia in 1997 (Jevti c et al., 1997) but until 2014 it was not considered as an economically important pathogen. In 2007, the average temperature in January (6.1 C) and February (5.8 C) exceeded ten-year temperature averages of 1.6 C and 2.3 C giving optimal temperature conditions for yellow rust outbreak, but total rainfall in April was 0 mm, inhibiting yellow rust occurrence (Fig. 8). In 2014, the average temperature in January
Fig. 4. Residual analysis in multiple regression model for predicting yield loss of the Barbee variety.
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
23
Fig. 5. Regression of actual and predicted yield losses of the Barbee variety.
Fig. 6. Residual analysis obtained by the multiple regression model related to the Durumko variety.
(4.2 C) and February (6.1 C) exceeded ten-year temperature averages and total rainfall in March (49.5 mm) and April (51.2 mm) were above average rainfall totals for the period favoring yellow rust occurrence in Serbia (Fig. 8). It was confirmed by Dr. Diane Saunders from John Innes Centre, UK, that the Warrior Race of wheat yellow rust became predominate in Serbia (personal communication). The Warrior Race rapidly spread throughout Europe from 2011 and became prevalent in many countries within one or very few years (Hovmøller et al., 2016). The yield loss caused by yellow rust in wheat production areas in Serbia was up to 20% with the disease index from 40 to 60%
(Jevti c et al., 2014). However, in field trials the Barbee variety suffered yield reduction up to 51%, with a maximum disease index of 60%. In the same year, a yield loss of the Durumko variety caused by yellow rust reached 20% with a maximum disease index of 11%. In this study, it was not possible to include the time period of 2014 and 2015 in the regression models for yield loss predictions because climatic conditions changed and favored the occurrence of yellow rust over leaf rust. Currently, yield loss prediction models have to be constructed carefully because climate change may modify the range of predominant wheat diseases, and pathogens currently considered economically less important may turn into
24
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
Fig. 7. Regression of actual and predicted yield losses of the Durumko variety.
Fig. 8. Climatic conditions in ten-year period in Serbia.
potential threats to wheat production. The results in this study also show that models for yield loss prediction should be constructed using data from more than a twoor three-year trial period, which is in accordance with those who suggested that there is a problem with models that rely solely on the quantification of visible disease symptoms without taking into account the variations in growing conditions that occur between seasons (Teng, 1985).
4. Conclusions 1. Considering the relationship between disease scoring systems and yield loss is not straightforward, an integrated approach, using abiotic and biotic factors in regression modeling would result in higher potentials for yield loss predictions. 2. Regression models for predicting potential yield losses, comprising both biotic and abiotic factors, gave higher R2 values
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
and R2pred and lower Mallows' Cp, compared to those comprising biotic factors only. 3. Climatic elements influencing yield loss and disease indices are correlated; thus, the impact of each should be analyzed as part of a complex environmental system. 4. Additionally, this study suggests that climate change can potentially cause shifts in predominant pathogens in wheat growing areas, jeopardizing the precision of yield loss prediction models. Acknowledgements This paper was realized as a part of the project TR 31066 Contemporary breeding of small grains for current and future needs financed by the Ministry of Education and Science of the Republic of Serbia. References Berraies, S., Gharbi, M.S., Rezgui, S., Yahyaoui, A., 2014. Estimating grain yield losses caused by septoria leaf blotch on durum wheat in Tunisia. ChileanJAR 74 (4), 432e437. http://dx.doi.org/10.4067/S0718-58392014000400009. Budka, A., Łacka, A., Gaj, R., Jajor, E., Korbas, K., 2015. Predicting winter wheat yields by comparing regression equations. Crop Prot. 78, 84e91. http://dx.doi.org/ 10.1016/j.cropro.2015.08.006. Cao, X., Yao, D., Duan, X., Liu, W., Fan, J., Ding, K., Zhou, Y., 2014. Effects of powdery mildew on 1 000-kernel weight, crude protein content and yield of winter wheat in three consecutive growing seasons. JIA 13 (7), 1530e1537. http:// dx.doi.org/10.1016/S2095e3119(14)60806-6. Chantret, N., Mingeot, D., Sourdille, P., Bernard, M., Jacquemin, J.M., Doussinault, G., 2001. A major QTL for powdery mildew resistance is stable over time and at two development stages in winter wheat. Theor. Appl. Genet. 103, 962e971. Chungu, C., Gilbert, J., Townley-Smith, F., 2001. Septoria tritici blotch development as affected by temperature, duration of leaf wetness, inoculum concentration, and host. Plant Dis. 85 (4), 430e435. Corazza, L., Islongo, M.T., 1987. Wheat and barley powdery mildew in Italy. In: Wolfe, M.S., Limpert, E. (Eds.), Integrated Control of Cereal Mildews: Monitoring the Pathogen. Springer. ISBN 978-90-247-3626-3. Duveiller, E., Singh, R.P., Nicol, J.M., 2007. The challenges of maintaining wheat productivity: pests, diseases, and potential epidemics. Euphytica 157, 417e430. http://dx.doi.org/10.1007/s10681-007-9380-z. FAO, 2012. Food and agriculture organisation of the United Nations, FAO STAT, FAO statistics division. http://faostat.fao.org/site/567/default.aspx#ancor.
25
Green, A., Berger, G., Griffey, C.A., Pitman, R., Thomason, W., Balota, M., 2014. Genetic resistance to and effect of leaf rust and powdery mildew on yield and its components in 50 soft red winter wheat cultivars. Crop Prot. 64, 177e186. Hovmøller, M.S., Walter, S., Bayles, R.A., Hubbard, A., Flath, K., Sommerfeldt, N., Leconte, M., Czembor, P., Rodriguez-Algaba, J., Thach, T., Hansen, J.G., Lassen, P., Justesen, A.F., Ali, S., de Vallavieille-Pope, C., 2016. Replacement of the European wheat yellow rust population by new races from the centre of diversity in the near-Himalayan region. Plant Pathol. 65, 402e411. Huerta-Espino, J., Roelfs, A.P., 1992. Leaf rust on durum wheats. Vortr. Pflanzenzuchtg 24, 100e102. James, W.C., 1971. An illustrated series of assessment keys for plant diseases, their preparation and usage. Can. Plant Dis. Surv. 51, 39e65. Jevti c, R., Jerkovi c, Z., Den ci c, S., Stojanovi c, S., 1997. Occurrence of yellow rust (Puccinia striiformis) on winter wheat in 1997. Plant Dr. 4, 455e458 (In Serbian). Jevti c, R., Lalosevi c, Mirjana, Jerkovi c, Z., Mladenov, N., Hristov, N., 2014. Yellow rust threatens to halve the wheat yield. Plant Dr. 1, 6e13 (In Serbian). Junk, J., Kouadio, L., Delfosse, P., El Jarroudi, M., 2016. Effects of regional climate change on brown rust disease in winter wheat. Clim. Change 135, 439e451. Juroszek, P., Von Tiedemann, A., 2013. Climate change and potential future risks through wheat diseases: a review. Eur. J. Plant Pathol. 136, 21e33. http:// dx.doi.org/10.1007/s10658-012-0144-9. Lorys, M.M.A., Lannou, V., Lannou, C., 2000. Selection for increased spore efficacy by host genetic backgroundin a wheat powdery mildew population. Phytopathology 90 (12), 1300e1306. Peterson, R.F., Campbell, A.B., Hannah, A.E., 1948. A diagrammatic scale for estimating rust intensity on leaves and stems of cereal. Can. J. Res. 26, 496e500. http://dx.doi.org/10.1139/cjr48ce033. Robe, P., Pavoine, Mt, Doussinault, G., 1996. Early assessment of adult plant reaction of wheat (Triticum aestivum L) to powdery mildew (Erysiphe graminis f sp tritici) at the five-leaf seedling stage. Agronomie. EDP Sci. 16 (7), 441e451. Shaw, M.W., Royle, D.J., 1992. Airborne inoculum as a major source of Septoria tritici (Mycosphaerella graminicola) infections in winter wheat crops in the UK. Plant Pathol. 42, 882e899. Teng, P.S., 1985. Construction of predictive models. II. Forecasting crop losses. Adv. Plant Pathol 3, 179e206. Wegulo, S.N., Breathnach, J.A., Baenziger, P.S., 2009. Effect of growth stage on the relationship between tan spot and spot blotch severity and yield in winter wheat. Crop Prot. 28, 696e702. http://dx.doi.org/10.1016/j.cropro.2009.04.003. Wiik, L., 2009. Yield and disease control in winter wheat in southern Sweden during 1977e2005. Crop Prot. 28, 82e89. http://dx.doi.org/10.1016/ j.cropro.2008.09.002. White, J.W., Hoogenboom, G., Kimball, B.A., Wall, G.W., 2011. Methodologies for simulating impacts of climate change on crop production. Field crop. Res. 124, 357e368. http://dx.doi.org/10.1016/j.fcr.2011.07.001. Zhang, X.Y., Loyce, C., Meynard, J.M., Savary, S., 2006. Characterization of multiple disease systems and cultivar susceptibilities for the analysis of yield losses in winter wheat. Crop Prot. 25, 1013e1023. http://dx.doi.org/10.1016/ j.cropro.2006.01.013.
Contents lists available at ScienceDirect
Crop Protection journal homepage: www.elsevier.com/locate/cropro
Predicting potential winter wheat yield losses caused by multiple disease systems and climatic conditions a b Radivoje Jevti c a, *, Vesna Zupunski , Mirjana Lalosevi c a, Ljubica Zupunski a b
Institute of Field and Vegetable Crops, Maksima Gorkog 30, 21000 Novi Sad, Serbia University of Novi Sad, Faculty of Technical Sciences, Trg Dositeja Obradovica 6, 21000 Novi Sad, Serbia
a r t i c l e i n f o
a b s t r a c t
Article history: Received 5 December 2016 Received in revised form 11 April 2017 Accepted 3 May 2017
Yield losses in field crops are most commonly predicted by using regression models that include either biotic or abiotic factors as predictor variables. Knowing that yield loss is a complex trait, the potential capability of regression models for predicting yield losses by using models containing both biotic and abiotic factors as predictors were estimated in this study. Biotic factors considered in regression models were: leaf rust, powdery mildew, septoria tritici blotch and tan spot occurrence on the varieties Barbee and Durumko known to have various degrees of susceptibility to obligate parasites and leaf blotch diseases. Among abiotic factors, monthly averages of temperature, relative humidity and total rainfall taken from November to June for growing seasons 2006e2013 were used as predictors. In 2014, yellow rust became the predominant pathogen over leaf rust, thus 2014 and 2015 were excluded from regression models and analyzed separately. Since a high correlation was found between abiotic and biotic factors, partial least squares regression, stepwise regression and best subsets regression were applied. Best subsets regression revealed that models consisted of both biotic and abiotic factors were more precise in estimating regression coefficients and predicting future responses. The potential yield loss predictions, conducted using these models, were regressed with actual yield losses, and high coefficients of determination (R2 ¼ 79% for Barbee; and R2 ¼ 63% for Durumko) were obtained. It was also evident that using more predictors in regression models does not necessarily mean that the model would have a higher potential in making yield loss predictions. This study confirms that the relationship between a disease scoring scale and yield loss is not straightforward and higher potentials for yield loss predictions were given due to the regression models using abiotic and biotic predictor variables. © 2017 Elsevier Ltd. All rights reserved.
Keywords: Puccinia Blumeria Zymoseptoria Regression models Yield loss Winter wheat
1. Introduction Wheat (mainly common/soft wheat Triticum aestivum but also durum/hard wheat T. turgidum) is the world's largest crop in terms of harvested area and ranks third in the global annual production of commodities (FAO, 2012). Significant concerns have been raised by the scientific community about the impacts of climate change on future yield potentials of wheat. However, in spite of the fact that climatic changes have been emphasized to have impact on wheat yield and quality, not only directly but also due to the interactions with biotic factors, the effects of biotic factors on yield losses have been neglected in
* Corresponding author. E-mail addresses: [email protected] (R. Jevti c), vesna.zupunski@ ifvcns.ns.ac.rs (V. Zupunski), [email protected] (M. Lalosevi c), [email protected] (L. Zupunski). http://dx.doi.org/10.1016/j.cropro.2017.05.005 0261-2194/© 2017 Elsevier Ltd. All rights reserved.
recently reported studies (White et al., 2011; Juroszek and Von Tiedemann, 2013). Among the economically most important diseases affecting winter wheat are obligate parasites (Blumeria graminis f. sp. tritici, Puccinia graminis f. sp. tritici, Puccinia triticina, Puccina striiformis f. sp. tritici) and crop residue-borne necrotrophic pathogens (Pyrenophora tritici-repentis, Zymoseptoria tritici, Parastagonospora nodorum, Cochliobolus sativus, Fusarium species). Although chemical treatment is a very powerful disease-control tool, followed by increases in yield, the main imperative of integrated pest management is environmental protection and reduced fungicide input. Thus, many efforts have been directed on determining damage thresholds and developing mathematical models that can be used to forecast yield losses caused by pathogenic infection. Typically, these experiments are conducted in a few locations during a two or three - year period, but only some of them raise the question about disease dynamics and yield over longer time periods (Wiik, 2009).
18
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
In literature, adversarial reports appear regarding correlation between diseased leaf area or any disease scoring scale and yield. For instance, some authors reported linear regression as suitable for describing relationship between disease rating scale and yield loss (Wegulo et al., 2009; Green et al., 2014; Budka et al., 2015), whereas there are also reports that the relationship between the two is not straightforward (Duveiller et al., 2007). The objective of this study was to evaluate the potential of regression models for predicting yield losses in winter wheat if both biotic (leaf rust, powdery mildew, septoria tritici blotch, and tan spot) and abiotic factors (climatic conditions from November to June) are subjected to the same regression model. The data collected from 2006 to 2015 were analyzed and characterized in terms of agro-ecological conditions of Serbia. 2. Materials and methods Data were obtained from fungicide efficacy trials which were conducted in the locality of Rimski San cevi (Vojvodina, north province of Serbia) under the direction of the Institute of Field and Vegetable Crops, Novi Sad, Serbia, over the period of 2006e2015 using soft wheat variety Barbee (Triticum aestivum ssp. compactum), and hard wheat variety Durumko (Triticum turgidum subsp. durum). The Barbee variety has shown increased susceptibility to obligate parasites (Blumeria and Puccinia) while Durumko showed increased susceptibility to leaf blotch diseases (LBDs) such as Pyrenophora tritici-repentis, Zymoseptoria tritici, and Phaeosphaeria nodorum in agro-ecological conditions of Serbia. 2.1. Field trial Field trials were set up under naturally occurring inoculum and were arranged in a randomized block design comprising four replicates. The plot size of each replicate was 10 m2. A trial usually included 10 fungicide-sprayed and non-sprayed check treatments. Fungicides were applied at growth stage BBCH 36e37 (flag leaf just visible, rolled) and BBCH 51e59 (inflorescence emergence, heading). Different types of active ingredients such as amides, aromatics, azoles, benzimidazoles, morpholines, oxazoles, strobilurins, pyrazoles and pyridines were applied with recommended dosage rates using calibrated field crop sprayers with fan nozzles, at 300 kPa pressure and 200 L of water per hectare. Mean sowing date for winter wheat was 20 October (optimal time of sowing) and the mean harvest date was 30 June (range 25 June- 07 July). 2.2. Disease assessment Assessments of leaf disease severity were made at the growth stage 71e73 BBCH (kernel watery; early milk), known to be highly related to yield (Wegulo et al., 2009). Disease severity for leaf rust and powdery mildew was assessed using modified Cobb's scale (Peterson et al., 1948; Corazza and Islongo, 1987). Disease severity of septoria tritici blotch and tan spot were assessed using the disease rating keys devised by James 1971. The disease indices (%) of leaf rust, powdery mildew, septoria tritici blotch and tan spot were calculated by taking into consideration disease incidence and average disease severity (Cao et al., 2014). 2.3. Yield Yield was measured for each plot after harvest at 15% water content. Yield loss (%) was determined as yield reduction in untreated plots compared with yield response to fungicide treatment which provided the best control of wheat diseases (Eq. (1)).
Yð%Þ ¼ ððY1 Y2 Þ=Y1 Þ 100
(1)
Y1 - grain yield of fungicide treatment for the best wheat disease control. Y2 - grain yield of the non - sprayed check treatment.
2.4. Predictor variables for regression models In regression models, the biotic factors regarded as potential predictive variables were disease indices of leaf rust, powdery mildew, septoria tritici blotch and tan spot. On the other hand, abiotic factors used as predictors comprised monthly averages of: temperatures, relative humidity and total rainfall taken from November to June for growing seasons 2006e2013 (http://www. hidmet.gov.rs/). The shift in predominant rust pathogen occurred in 2014, thus a period of 2014 and 2015 was excluded from regression models and analyzed separately. 2.5. Statistical methods The effects of year, variety and fungicide treatment on yield were examined by analysis of variance (ANOVA). Further, multivariate regression models were used to estimate relationship between disease indices, abiotic factors and yield losses. Knowing that abiotic and biotic factors are correlated not just to the yield loss but also with each other (multicollinearity), partial least squares regression, stepwise regression and best subsets regression were applied to make predictions of yield losses in the Barbee and Durumko varieties. Multicollinearity is problematic because it can increase the variance of the regression coefficients, making it difficult to evaluate the individual impact that each of the correlated predictors has on the response. Partial least squares (PLS) is a biased regression procedure that reduces the number of predictors and extracts a set of components that describes maximum correlation among the predictors and response variables. The technique used is similar to principal component analysis because it gives the option of leave-one-out cross-validation, which is used to maximize the model's predictive ability. In addition, the method that can also be used to analyze correlated predictors is the stepwise regression model. It is used in the exploratory stages of model building to identify a useful subset of predictors. The process systematically adds the most significant variable or removes the least significant variable during each step until it identifies variables that explain the maximum variation in yield loss. Best subsets regression was performed to identify the best-fitting regression models with predictors of choice as well as to compare regression models obtained by PLS and stepwise regression. The general approach was to select the smallest subset that fulfills certain statistical criteria. The reason for using a subset of variables instead of a complete set is because the subset model might actually estimate the regression coefficients and predict future responses with smaller variance than the full model using all predictors. Regression models were followed with coefficient of determination (R2), coefficient of prediction (R2pred), variance inflation factor (VIF) and Mallows' Cp. Coefficient of determination (R2) is the percentage of variation in the response that is explained by the model. Coefficient of prediction (R2pred) determines how well the model predicts the response for new observations. VIF indicates how much the variance of an estimated regression coefficient increases if predictors are correlated. The VIFs will all be 1 if there is no correlation between factors. Mallows' Cp is used for
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
making comparisons between multiple regression models. It compares the precision and bias of the full model to models with a subset of predictors and strikes an important balance with the number of predictors in the model. A Mallows' Cp should be small and close to the number of predictors in the model plus the constant (p) value which is a number of explanatory variables. If Mallows' Cp is close to the number of predictors plus the constant, the model is relatively unbiased in estimating the true regression coefficients and predicting future responses. We determined the correlation coefficients between climatic factors, yield losses and disease indices using Spearman's coefficient of correlation. Minitab 17 (trial version) was used for entire analysis.
3. Results and discussion The yield of treated and untreated plots of the Barbee and Durumko varieties did not change linearly over the period of ten years. Average yield of variety Barbee was 5.1 t/ha on treated and 3.6 t/ha on untreated plots with an average yield loss of 30%. Average yield loss of variety Durumko was 10%, with 7.3 t/ha on treated and 6.5 t/ha on untreated plots. Yield of the Barbee and Durumko varieties differed significantly (P ¼ 0.000) and were influenced by year (P ¼ 0.000) and applied fungicide treatments (P ¼ 0.000). There were statistically significant interactions between year and cultivar (P ¼ 0.000); year and fungicide application (P ¼ 0.006) and cultivar and fungicide application (P ¼ 0.023). Yield losses varied considerably during the 2006e2015 period, and the Barbee variety suffered a much higher percentage yield loss (up to 45%) than the variety Durumko (up to 25%) (Fig. 1). The Barbee variety is a bearded, semi-dwarf soft white club winter wheat and is known for its susceptibility on powdery mildew and rusts of wheat. Because of its high susceptibility to obligate parasites it was used as susceptible check in field trails in many countries (Robe et al., 1996; Lorys et al., 2000; Chantret et al., 2001). Barbee variety has lower yield potentials (5.1 t/ha) compared with the Durumko variety (7.3 t/ha) and suffered higher average yield loss (30%). On average, disease index of leaf rust and powdery mildew on the Barbee variety for the period was 22% and 21%, respectively. However, in 2009 disease pressure of leaf rust (68%) and powdery mildew (30%) was very high causing yield loss of 44%. The average disease index of septoria tritici blotch on the Barbee variety was 7% for a ten-year period and it was of minor importance on yield losses comparing with influence of obligate parasites. According to Zhang et al. (2006), the occurrence of septoria tritici blotch, when combined with leaf rust and powdery mildew on the
Fig. 1. Yield losses of Barbee and Durumko varieties in the period 2006e2015.
19
same host, is highly influenced by variety itself; thus, five potential disease profiles were introduced based on different responses during the examination of the varieties produced in multiple disease systems. The Durumko variety is durum wheat which had increased susceptibility to leaf blotch diseases in agro-ecological conditions of Serbia, and suffered average yield losses of 10%. The yield loss of Durumko variety can be explained by low disease pressure of obligate parasites and septoria tritici blotch occurrence with an average disease index of 17%. The exception was in 2006 when septoria tritici blotch index reached 51% causing a yield loss of 20%. The low disease pressure of an obligate parasite and high disease pressure of septoria tritici blotch on Durumko variety can be explained by adaptation of predominant pathotypes on bread and durum wheat. It was reported that predominant and highly virulent populations of P. triticina on bread wheat are different and often avirulent on durum wheat and vice versa (Huerta-Espino and Roelfs, 1992). 3.1. Prediction of potential yield losses in Barbee and Durumko varieties using biotic and abiotic factors as predictors The potentials of regression models to predict yield losses using biotic together with abiotic factors as predictors were examined. Since climatic factors are highly correlated with disease indices, regression models such as partial least squares regression, stepwise regression and best subsets regression were applied. To estimate the most influencing factors on yield loss of varieties Barbee and Durumko, climatic factors and disease indices were subjected to partial least squares regression. The two-component regression model accounted for 71% of variation in yield loss of Barbee variety and adding more components did not significantly increase R2 (Fig. 2A). The partial least-squares method indicated that using only one predictor contributes in variation in yield loss of Durumko variety with an R2 of 57% (Fig. 2B). Among abiotic and biotic predictors, leaf rust and powdery mildew indices together with total rainfall in March had the highest absolute values of standardized coefficients and as a result these were considered to have the greatest impact on yield loss of the Barbee variety (Fig. 3). However, it should be noted that the value of standardized coefficients does not strictly imply a negative or positive correlation with yield loss because of multicollinearity. The factors which were considered to be highly influencing on yield loss of Durumko variety were temperatures in January and June, relative humidity in December and septoria tritici blotch index (Fig. 3). Stepwise regression, another model used to address multicollinearity, identified a subset of predictors based on their statistical significance. In stepwise regression, the leaf rust and powdery mildew indices as well as temperatures in April, proved to be the strongest predictors of yield loss in the Barbee variety, giving a coefficient of determination (R2) of 70% and coefficient of prediction (R2pred) of 54%. The stepwise regression model indicated that the greatest impact on yield loss of variety Durumko had septoria tritici blotch index and temperature in June giving the model with (R2) of 63% and (R2pred) of 57%. The most influential predictors on yield losses did not completely coincide with the application of the partial least squares and stepwise regression models on the Barbee or Durumko varieties. To compare regression models and examine how much variation of yield loss will be explained by the maximum R2ecriterion the best subsets regression was applied. The most influential factors determined by stepwise regression were subjected to best subsets regression together with climatic elements which were significantly moderately and highly correlated with yield losses and disease indices at P 0.05 and R2 50% (Table 1).
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
20
Fig. 2. Model analysis for yield loss prediction of Barbee (A) and Durumko (B) varieties obtained by partial least squares regression.
Fig. 3. Standardized coefficients of biotic and abiotic factors influencing yield loss of Barbee and Durumko varieties.
Table 1 The correlation between climatic factors, yield losses and disease indices related with varieties Barbee and Durumko.
Barbee
Leaf rust
December
March
April
May
Total rainfall (mm)
Total rainfall (mm)
Total rainfall (mm)
Temperature ( C)
r ¼ 0.741 (P ¼ 0.000) R2 ¼ 51.4%
r ¼ 0.857 (P ¼ 0.000) R2 ¼ 73.1%
r ¼ 0.570 (P ¼ 0.008) R2 ¼ 68.1%
Powdery mildew Yield loss Durumko
Septoria tritici blotch Yield loss
Climatic elements which were significantly correlated with yield losses and disease indices with R2 50% had higher values of Mallows’ Cp than those with R2 50% (data not shown); thus, they were not considered as the most influencing factors on yield loss of varieties. The best subsets regression revealed that temperature in April, leaf rust index and powdery mildew index formed the model with the smallest Mallows' Cp (2.1) which indicated that this model is relatively unbiased in estimating the true regression coefficients and predicting future yield losses of the Barbee variety (Table 2.)
r ¼ 0.739 (P ¼ 0.000) R2 ¼ 51.4%
The same set of predictors was identified by stepwise regression model giving R2 (70%) and R2pred (54%). The average temperatures in April ranged from 12.3 C to 14.6 C in Serbia's climate conditions and these temperatures are in accordance with the temperature range (12 C to 15 C) reported to be optimal for germination of leaf rust urediniospores (Junk et al., 2016). In addition, the minimal temperatures for leaf rust development starts from 8 C and are related with night temperatures (Junk et al., 2016). The average minimal temperature of 7 C was reached no earlier than April. Total rainfall and relative humidity in April were also significantly
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
21
Table 2 Best subsets regression for predicting yield loss in the Barbee variety.
related with leaf rust indices but were not selected as more significant than temperatures in April since optimal relative humidity > 60% is usually met in the spring in Serbia. The set of predictors identified by partial least squares regression (total rainfall in March, leaf rust index and powdery mildew index) had lower R2, R2pred and higher Mallows’Cp than the set of predictors chosen by stepwise regression. Additionally, best subsets regression indicated that using only biotic factors as predictors for yield loss predictions (in this case leaf rust index) would give lower R2 (55.6%), R2pred (49.7%) and much higher Mallows’Cp (8.8) than using it in combination with abiotic factors (Mallows’Cp decreased to 2.4). A high Mallows' Cp value (8.8) indicated that the model is not precise (has large variance) in estimating the true regression coefficients and predicting future responses. The combination of leaf rust and powdery mildew indices as predictors had high potential for predicting yield losses of the Barbee variety, but still with lower R2, R2pred and higher Mallows’Cp than using them in combination with T in April. It was also evident that using more predictors does not necessarily mean that the model would have a higher potential for making yield loss predictions. The best subsets regression revealed that the best model for yield loss prediction in variety Durumko was comprised of temperature in June and septoria tritici blotch index giving the smallest Mallows’ Cp (2.5) and optimal combination of R2 (63.2%) and R2pred (57.4%) (Table 3). Using only biotic factors as predictors for yield loss prediction (in this case septoria tritici blotch index) would give lower R2 (37.2%), R2pred (29.4%) and much higher Mallows’Cp (20.7) than
Table 3 Best subsets regression for yield loss prediction in variety Durumko.
using it in combination with abiotic factors. Low coefficient of determination (37.2%) obtained after regression of septroria tritici blotch index and yield loss of variety Durumko is in accordance with coefficients of determination of 39% and 44% obtained by Berraies et al. (2014) who estimated grain yield losses caused by septoria tritici blotch using 400 lines of durum wheat in two sowing seasons. The temperature in June was found to be highly influential on yield loss of the Durumko variety since environmental conditions in June are favorable for septoria tritici blotch pycnidia development. The day/night temperatures in June were on average 26 C day/15 C night which were found to be very conducive for pycnidia development causing higher severity of the disease. Chungu et al. (2001) reported that percent pycnidial coverage increased as incubation temperature increased from 15 C day/11 C night to 22 Cday/15 C night, which is in accordance with our study. In addition, the latent period is temperature dependent and decreased as temperature increased (Chungu et al., 2001; Shaw and Royle, 1992). It was also reported that the strongest relationship between disease index and yield includes the period from the end of anthesis (ZGS 69) to milk stage (ZGS 70e73) which coincides with the second half of May and June in Serbia's climate conditions. 3.2. Predicting yield losses of varieties Barbee and Durumko Since it was indicated that T in April, leaf rust index and powdery mildew index formed the best set of predictors for predicting yield loss of the Barbee variety, multiple regression was performed,
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
22
producing the following equation:
equation:
Y ¼ 86:0 4:72x1 5:05x2 þ 5:07x3 þ 0:392x1 x2 0:405x1
Y ¼ 84:0 3:76x1 þ 0:70x2 0:0210x1 x2
x3 þ 0:115x2 x3 þ 0:00766x1 x2 x3
Y e yield loss of variety Barbee x1e T in April x2 e Leaf rust index x3 e Powdery mildew index The P-value (0.000) for the regression model in the analysis of variance showed that the model estimated by the regression procedure is significant at an a-level of 0.05. The P-value for the estimated coefficient of leaf rust index (0.000) and T in April (0.046) indicated that these were significantly correlated to yield loss in the Barbee variety. The P-value for estimated interaction coefficients between T in April and leaf rust (0.046), as well as T in April and powdery mildew index (0.019) showed also significant correlation to yield loss at an a-level of 0.05. The VIFs indicated that regression coefficients were moderately correlated but still well estimated. The percentage of variation in yield loss explained by the model was R2 ¼ 79%, and the possibility for prediction of new observations was R2pred ¼ 61%. An analysis of residuals and a test for the possible occurrence of outliers showed that there were no outliers in the data. The normal probability plot evinced an approximately linear pattern which is consistent with a normal distribution. The plot of residuals versus the fitted values showed that residuals were randomly distributed (Fig. 4). The actual yield losses and predicted yield losses were regressed and high R2 value (79%) was obtained (Fig. 5). To make a prediction of yield losses of the Durumko variety, temperature in June and septoria tritici blotch index were subjected to multiple regression resulting in the following regression
Y eYield loss of variety Durumko x1 e T in June x2 e Septoria tritici blotch index Coefficients of determination (R2) and prediction (R2pred) were 63% and 55%, respectively. The P-value for the regression model (0.000) confirmed model significance at an a-level of 0.05. The Pvalues for the estimated coefficient of septoria tritici blotch index (0.001) and temperature in June (0.000) showed significant correlation with yield loss in the Durumko variety at an a-level of 0.05. The VIFs were less than 1.34, indicating that predictors were well estimated. A residual analysis indicated that residuals were distributed randomly and that there were no outliers in the data. The normal probability plot showed an approximately linear pattern that is consistent with a normal distribution (Fig. 6). The relation between actual yield losses of the Durumko variety and predicted yield losses was fitted to linear regression and high R2 value (63%) was obtained (Fig. 7). 3.3. The shift of predominant rust pathogens in Serbia Monitoring the occurrence of wheat diseases in the last ten years showed a shift in predominant rust pathogen in the wheat growing area. Yellow rust caused by Puccinia striiformis f. sp. tritici was first recognized in Serbia in 1997 (Jevti c et al., 1997) but until 2014 it was not considered as an economically important pathogen. In 2007, the average temperature in January (6.1 C) and February (5.8 C) exceeded ten-year temperature averages of 1.6 C and 2.3 C giving optimal temperature conditions for yellow rust outbreak, but total rainfall in April was 0 mm, inhibiting yellow rust occurrence (Fig. 8). In 2014, the average temperature in January
Fig. 4. Residual analysis in multiple regression model for predicting yield loss of the Barbee variety.
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
23
Fig. 5. Regression of actual and predicted yield losses of the Barbee variety.
Fig. 6. Residual analysis obtained by the multiple regression model related to the Durumko variety.
(4.2 C) and February (6.1 C) exceeded ten-year temperature averages and total rainfall in March (49.5 mm) and April (51.2 mm) were above average rainfall totals for the period favoring yellow rust occurrence in Serbia (Fig. 8). It was confirmed by Dr. Diane Saunders from John Innes Centre, UK, that the Warrior Race of wheat yellow rust became predominate in Serbia (personal communication). The Warrior Race rapidly spread throughout Europe from 2011 and became prevalent in many countries within one or very few years (Hovmøller et al., 2016). The yield loss caused by yellow rust in wheat production areas in Serbia was up to 20% with the disease index from 40 to 60%
(Jevti c et al., 2014). However, in field trials the Barbee variety suffered yield reduction up to 51%, with a maximum disease index of 60%. In the same year, a yield loss of the Durumko variety caused by yellow rust reached 20% with a maximum disease index of 11%. In this study, it was not possible to include the time period of 2014 and 2015 in the regression models for yield loss predictions because climatic conditions changed and favored the occurrence of yellow rust over leaf rust. Currently, yield loss prediction models have to be constructed carefully because climate change may modify the range of predominant wheat diseases, and pathogens currently considered economically less important may turn into
24
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
Fig. 7. Regression of actual and predicted yield losses of the Durumko variety.
Fig. 8. Climatic conditions in ten-year period in Serbia.
potential threats to wheat production. The results in this study also show that models for yield loss prediction should be constructed using data from more than a twoor three-year trial period, which is in accordance with those who suggested that there is a problem with models that rely solely on the quantification of visible disease symptoms without taking into account the variations in growing conditions that occur between seasons (Teng, 1985).
4. Conclusions 1. Considering the relationship between disease scoring systems and yield loss is not straightforward, an integrated approach, using abiotic and biotic factors in regression modeling would result in higher potentials for yield loss predictions. 2. Regression models for predicting potential yield losses, comprising both biotic and abiotic factors, gave higher R2 values
R. Jevtic et al. / Crop Protection 99 (2017) 17e25
and R2pred and lower Mallows' Cp, compared to those comprising biotic factors only. 3. Climatic elements influencing yield loss and disease indices are correlated; thus, the impact of each should be analyzed as part of a complex environmental system. 4. Additionally, this study suggests that climate change can potentially cause shifts in predominant pathogens in wheat growing areas, jeopardizing the precision of yield loss prediction models. Acknowledgements This paper was realized as a part of the project TR 31066 Contemporary breeding of small grains for current and future needs financed by the Ministry of Education and Science of the Republic of Serbia. References Berraies, S., Gharbi, M.S., Rezgui, S., Yahyaoui, A., 2014. Estimating grain yield losses caused by septoria leaf blotch on durum wheat in Tunisia. ChileanJAR 74 (4), 432e437. http://dx.doi.org/10.4067/S0718-58392014000400009. Budka, A., Łacka, A., Gaj, R., Jajor, E., Korbas, K., 2015. Predicting winter wheat yields by comparing regression equations. Crop Prot. 78, 84e91. http://dx.doi.org/ 10.1016/j.cropro.2015.08.006. Cao, X., Yao, D., Duan, X., Liu, W., Fan, J., Ding, K., Zhou, Y., 2014. Effects of powdery mildew on 1 000-kernel weight, crude protein content and yield of winter wheat in three consecutive growing seasons. JIA 13 (7), 1530e1537. http:// dx.doi.org/10.1016/S2095e3119(14)60806-6. Chantret, N., Mingeot, D., Sourdille, P., Bernard, M., Jacquemin, J.M., Doussinault, G., 2001. A major QTL for powdery mildew resistance is stable over time and at two development stages in winter wheat. Theor. Appl. Genet. 103, 962e971. Chungu, C., Gilbert, J., Townley-Smith, F., 2001. Septoria tritici blotch development as affected by temperature, duration of leaf wetness, inoculum concentration, and host. Plant Dis. 85 (4), 430e435. Corazza, L., Islongo, M.T., 1987. Wheat and barley powdery mildew in Italy. In: Wolfe, M.S., Limpert, E. (Eds.), Integrated Control of Cereal Mildews: Monitoring the Pathogen. Springer. ISBN 978-90-247-3626-3. Duveiller, E., Singh, R.P., Nicol, J.M., 2007. The challenges of maintaining wheat productivity: pests, diseases, and potential epidemics. Euphytica 157, 417e430. http://dx.doi.org/10.1007/s10681-007-9380-z. FAO, 2012. Food and agriculture organisation of the United Nations, FAO STAT, FAO statistics division. http://faostat.fao.org/site/567/default.aspx#ancor.
25
Green, A., Berger, G., Griffey, C.A., Pitman, R., Thomason, W., Balota, M., 2014. Genetic resistance to and effect of leaf rust and powdery mildew on yield and its components in 50 soft red winter wheat cultivars. Crop Prot. 64, 177e186. Hovmøller, M.S., Walter, S., Bayles, R.A., Hubbard, A., Flath, K., Sommerfeldt, N., Leconte, M., Czembor, P., Rodriguez-Algaba, J., Thach, T., Hansen, J.G., Lassen, P., Justesen, A.F., Ali, S., de Vallavieille-Pope, C., 2016. Replacement of the European wheat yellow rust population by new races from the centre of diversity in the near-Himalayan region. Plant Pathol. 65, 402e411. Huerta-Espino, J., Roelfs, A.P., 1992. Leaf rust on durum wheats. Vortr. Pflanzenzuchtg 24, 100e102. James, W.C., 1971. An illustrated series of assessment keys for plant diseases, their preparation and usage. Can. Plant Dis. Surv. 51, 39e65. Jevti c, R., Jerkovi c, Z., Den ci c, S., Stojanovi c, S., 1997. Occurrence of yellow rust (Puccinia striiformis) on winter wheat in 1997. Plant Dr. 4, 455e458 (In Serbian). Jevti c, R., Lalosevi c, Mirjana, Jerkovi c, Z., Mladenov, N., Hristov, N., 2014. Yellow rust threatens to halve the wheat yield. Plant Dr. 1, 6e13 (In Serbian). Junk, J., Kouadio, L., Delfosse, P., El Jarroudi, M., 2016. Effects of regional climate change on brown rust disease in winter wheat. Clim. Change 135, 439e451. Juroszek, P., Von Tiedemann, A., 2013. Climate change and potential future risks through wheat diseases: a review. Eur. J. Plant Pathol. 136, 21e33. http:// dx.doi.org/10.1007/s10658-012-0144-9. Lorys, M.M.A., Lannou, V., Lannou, C., 2000. Selection for increased spore efficacy by host genetic backgroundin a wheat powdery mildew population. Phytopathology 90 (12), 1300e1306. Peterson, R.F., Campbell, A.B., Hannah, A.E., 1948. A diagrammatic scale for estimating rust intensity on leaves and stems of cereal. Can. J. Res. 26, 496e500. http://dx.doi.org/10.1139/cjr48ce033. Robe, P., Pavoine, Mt, Doussinault, G., 1996. Early assessment of adult plant reaction of wheat (Triticum aestivum L) to powdery mildew (Erysiphe graminis f sp tritici) at the five-leaf seedling stage. Agronomie. EDP Sci. 16 (7), 441e451. Shaw, M.W., Royle, D.J., 1992. Airborne inoculum as a major source of Septoria tritici (Mycosphaerella graminicola) infections in winter wheat crops in the UK. Plant Pathol. 42, 882e899. Teng, P.S., 1985. Construction of predictive models. II. Forecasting crop losses. Adv. Plant Pathol 3, 179e206. Wegulo, S.N., Breathnach, J.A., Baenziger, P.S., 2009. Effect of growth stage on the relationship between tan spot and spot blotch severity and yield in winter wheat. Crop Prot. 28, 696e702. http://dx.doi.org/10.1016/j.cropro.2009.04.003. Wiik, L., 2009. Yield and disease control in winter wheat in southern Sweden during 1977e2005. Crop Prot. 28, 82e89. http://dx.doi.org/10.1016/ j.cropro.2008.09.002. White, J.W., Hoogenboom, G., Kimball, B.A., Wall, G.W., 2011. Methodologies for simulating impacts of climate change on crop production. Field crop. Res. 124, 357e368. http://dx.doi.org/10.1016/j.fcr.2011.07.001. Zhang, X.Y., Loyce, C., Meynard, J.M., Savary, S., 2006. Characterization of multiple disease systems and cultivar susceptibilities for the analysis of yield losses in winter wheat. Crop Prot. 25, 1013e1023. http://dx.doi.org/10.1016/ j.cropro.2006.01.013.