Yellow (stripe) rust, caused by Puccinia striiformis Westlend. f. sp. tritici Eriksson, has been known as a destructive disease in wheat farming systems worldwide. The development of yellow rust in susceptible wheat cultivars could be intensified by appropriate agronomic and environmental conditions as documented earlier (Naseri & Marefat, 2019). Plot-scale experiments in Iran evidenced that the area under disease progress curve (AUDPC) of yellow rust significantly corresponded to the timing of disease-onset and cultivar maturity, number of days with minimum temperature within 5-12˚C and maximum relative humidity (RH) above 60% over autumn-winter-spring months, resistance index and planting date (Naseri & Marefat, 2019). The next step was attempting to characterize the within and over-season progression of yellow rust on wheat crops following the description of specific and easy-to-use disease progress curve elements (Naseri & Kazemi, 2020). Based on a factor analysis, AUDPC, disease-onset time, Gaussian parameters, maximum disease incidence, and severity predictors justified 80% of the variability in yellow rust development across different wheat cultivars planted at various dates (Naseri & Kazemi, 2020). In another attempt, the spatial and temporal patterns of yellow rust epidemics in irrigated wheat at large scale were predicted according to the mean maximum RH for Aban (from October 23 to November 21), mean minimum temperature for Esfand (from February 20 to March 20), the number of icy days over autumn-winter-spring months, rainy days, days with minimum temperature within 7-8˚C and maximum RH > 60%, and the number of periods involving consecutive days with minimum temperature within 6-9˚C and maximum RH > 60% (Naseri & Sharifi, 2019). Whereas, over the same period (2009-2018) and regions, the two predictors of yellow rust epidemics in rain-fed wheat differed from the disease predictors described for irrigated wheat as mentioned earlier (Naseri, 2022). Although all the above-mentioned studies have potentially updated the best predictors of yellow rust development in wheat, it is still desired to understand how strongly such influential agronomic, climatic, and disease predictors are linked to wheat productivity. Moreover, we should know the complex interactions among planting date, resistance level, yellow rust progress, wheat maturation, and weather. The interaction of yellow rust with limited combinations of these agroecological factors has been studied earlier to estimate wheat yield (Murray et al., 1994; Gladders et al., 2007; Sharma-Poudyal & Chen, 2011; Lavilla, 2022). All these agroecological factors affect yellow rust and wheat yield when they interact with each other during the growing season at the field scale. Thus, we need an advanced estimation of wheat production and breeding programs according to wisely selected descriptors of agroecosystems.

In Iran, the effectiveness of various levels of race-specific and slow rusting resistance was evaluated in field plots under severe yellow rust epidemics that occurred during 2011-2013. Then, the wheat yield was predicted according to the AUDPC, apparent infection rate, final rust severity, and coefficient of infection determined for 16 wheat cultivars and lines (Safavi, 2015). The other yield estimations in wheat crops affected by yellow rust have been conducted using such common disease progress descriptors. Such yield estimates have been reported from Australia (Murray et al., 1994), Argentina (Lavilla, 2022), the UK (Gladders et al., 2007), and the USA (Sharma-Poudyal & Chen, 2011). Thus, most of the previous reports on predicting wheat yield reduced by yellow rust epidemics have been conducted based on measuring non-specific disease descriptors as mentioned above. Furthermore, interactions of yellow rust progress predictors with effective agronomic and environmental characteristics (Naseri & Marefat, 2019; Naseri & Sharifi, 2019; Naseri & Kazemi 2020) in estimating wheat yield remained unclear. Therefore, a more accurate and up-to-date simulation of wheat yield according to the multiple disease measurements within and over growing seasons based on specific yellow rust progress curve elements is still required. To meet this requirement, assessing the magnitude of associations of specific yellow rust progress curve elements with agroecological variables appears to be a necessary prerequisite to predict wheat yield more concisely. This better understanding is also required to breed more resistant wheat cultivars with longer-lasting genetic resistance.

According to Luo’s (2008) recommendations, the principal component analysis (PCA) leads to a minimum problem with selecting the best-fitted variables for a multiple regression model. Indeed, it is also crucial to maximize the data variance following the treatment of experimental plots with influential agronomic traits such as cultivar and planting date across several growing seasons (Naseri & Kazemi 2020). It is believed that this high variability in the dataset could increase the descriptive value of variables (Kranz, 2003). Hence, the following objectives were considered in this research to: (i) determine predictive values of cultivar resistance index, crop maturity time, air temperature, planting date, relative humidity, rainy days, and yellow rust progress curve elements; (ii) examine associations between the selected agro-ecological and disease progress curves; (iii) model wheat production according to this combination of agro-ecological and yellow rust predictors.

MATERIALS AND METHODS

Experiments

From the autumn of 2013 to the summer of 2017, the progression of yellow rust in commercial winter wheat (Triticum aestivum L.) cultivars was characterized across 282 experimental field plots prepared as presented in Table 1. Experiments were conducted according to a split-plot design with three replicates per treatment. To maximize variations in the development of yellow rust epidemics across the experimental plots, different dates of planting and cultivars of wheat were considered in the current four-season investigation. Eight bread wheat cultivars registered by the Seed and Plant Improvement Institute, Karaj, Iran were involved in experiments (Naseri & Kazemi, 2020). According to the maximum yellow rust severity rating during the current study (2013-2017), the resistance levels of these eight wheat cultivars were classified as moderately resistant for Pishgam and susceptible for the other cultivars examined (Naseri & Marefat, 2019).

Table 1. Experimental plots properties, factor categories, and measurements conducted at the research site, Kermanshah, Iran

Datasets

The yellow rust incidence was measured as the percentage of diseased plants at three spots observed per plot (Naseri & Marefat, 2019). The severity of wheat yellow rust was rated weekly as the percentage of leaf area pustulating for the three newest leaves of 3-5 randomly tested plants per plot (Naseri & Marefat, 2019). The progression of yellow rust was evaluated across 282 plots (2013-2017) according to the disease severity recorded at 4-6 assessment times depending on the length of epidemics developed each growing season. This resulted in 1419 disease measurements at highly diversified levels which were needed to improve the accuracy of the yield-predicting model.

Analysis

To characterize the progression of yellow rust in wheat cultivars, the disease severity dataset was subjected to modeling (Naseri & Kazemi, 2020). The progression of wheat yellow rust over the growing seasons was characterized based on the following disease progress curve elements: (1) the disease onset time; (2) the AUDPC calculated for the disease severity ratings over time; (3) maximum disease incidence over the study years; (4) maximum disease severity; (5) Gaussian curve parameters (Naseri & Kazemi, 2020). The GENSTAT (Payne et al., 2009), which commonly fits standard curves by maximum likelihood, was used for all the statistical procedures used in this investigation. The Gaussian model indicated the best fitness for the progression of yellow rust (Naseri & Kazemi, 2020). In this model, a describes the initial value or the constant term, b describes the height of the curve’s peak, m describes the position of the peak’s center, s (the standard deviation) describes the width of the Gaussian bell, and x describes the time intervals (Table 2). 

Table 2. Variables studied in commercial wheat cultivars differing in planting date, maturity time, and resistance index under yellow rust epidemics

^a Ranges of maturity time differing between cultivars, planting dates, and years.

^b Resistance index = 100 – Highest disease severity rating over the four-year study.

A parametric analysis of variance was performed on yellow rust severity and yield datasets collected from experimental plots over four growing seasons as reported earlier (Naseri & Marefat, 2019; Naseri, 2022). Therefore, mean values for the disease and yield for different levels of cultivar, planting date, assessment time, and growing season were provided by previous publications. In the present research, attempts were made to classify factors (cultivar, disease onset, growing season, planting date, and resistance index) based on a non-parametric analysis. Thus, this analysis provided further information on rankings of factor classifications needed for one-way interpretations of factor effects. To compare wheat yield between factor levels, a Kruscal-Wallis one-way ANOVA by H-test was used to rank the effects of disease onset time, maturity time, planting date, resistance index, and study year on yield. Using a factor analysis (FA), interrelationships among 13 agronomic, weather, wheat yield, and yellow rust progress predictors were determined according to the principal factors receiving eigenvalues (data variance proportion) ≥ 1.0. To predict productivity in wheat cultivars attacked by yellow rust, it was essential to reduce the number of predictors for fewer collinear yellow rust predictors for regression modeling. Hence, this reduction of collinear variables was performed with the help of this FA. For regression modeling, the adjusted coefficient of determination (R²) and Mallows Cp were used for stepwise selecting procedures (Brusco & Stahl, 2005; Naseri & Sasani, 2020). Furthermore, the graphical appraisal of normally distributed residuals, F-test, and R² were used for evaluating the model fitness (Naseri & Sabeti, 2021).

RESULTS AND DISCUSSION

Kruscal-Wallis one-way ANOVA

To compare wheat yield between factor levels, a Kruscal-Wallis one-way ANOVA according to the H-test was used to rank the effects of cultivar, disease onset time, planting date, resistance index, and year on yield. The productivity in the eight cultivars of wheat was not affected (Chi P > 0.05) by cultivar, disease onset time, and resistance index (Table 3). The highest yield ranking was determined for the early planting class of the planting date factor, which affected (Chi P < 0.001; mean H = 32.93) wheat yield. Then, the rankings of wheat yield decreased as the planting was delayed from optimum to late and very late plantings. This suggested that the ranking of wheat yield in very late planted plots improved by 184.55% when the cultivars were planted early in October over the four growing seasons. Planting wheat cultivars on an optimum date (November) improved yield ranking by 153.09% in comparison with the very late planting. For the year factor, which affected (Chi P < 0.001; mean H = 36.25) the productivity of wheat cultivars, the highest and lowest yield rankings were detected for the first (2013-2014) and the fourth (2016-2017) years of the study, respectively. Therefore, the Kruscal-Wallis one-way ANOVA results based on the H-tests demonstrated that early plantings of the eight wheat cultivars in the 2013-2014 experiments provided the highest yield levels compared to the other classes of planting date and year factors.  

Table 3. Kruskal-Wallis one-way ANOVA ranked yield levels for wheat cultivar, yellow rust onset, planting date, maturity time, resistance index, and year factors

Factor analysis

According to the results of FA, the three principal factors explained 74.14% of the variability in wheat maturity and planting date, resistance index, weather, yellow rust progress, and yield datasets (Table 4). These datasets were collected from the eight commercial wheat cultivars planted at different dates over the four growing seasons. The first principal factor, justifying 44.27% of the total data variance, evidenced the highest loading value for the contributions of the disease onset date (0.37) and Gaussian parameter m (-0.37), and the predictor of rainy days in spring (-0.37). Furthermore, the AUDPC and maximum disease severity showed negatively moderate contributions in the first principal factor. Among the seven yellow rust progress predictors, the highest significant contributions in the first principal factor were determined for the disease onset date and Gaussian parameter m predictors. Hence, the disease onset was considered an easier-to-assess disease predictor for the following regression analysis. Furthermore, this factor of the FA was defined as the factor of yellow rust progress in interaction with the weather predictor. The maturity time, planting date, and wheat yield predictors contributed significantly to the second principal factor accounting for 18.13% of total data variance. The maturity and planting date predictors provided negatively moderate contributions in the second principal factor, while wheat yield contributed with a highly positive loading. Thus, the second principal factor was regarded as the factor of wheat characteristics. The third principal factor, explaining 11.74% of the data variance, indicated a moderately negative loading for the Gaussian parameter b of yellow rust and a highly positive loading for the resistance index. This factor was defined as the factor of yellow rust progress in association with the resistance level of wheat cultivar (Table 4). This FA simplified the development of the regression model by determining the predictive values of disease, weather, and wheat variables studied at an experimental plot scale over four growing seasons.

Table 4. Factor analysis of yellow rust progression in wheat cultivars differing in planting date, resistance index, and maturity time

^a Number of days (autumn-winter-spring) with minimum temperature within 5-12°C & maximum relative humidity ≥ 60%.

*Bold numbers indicate significant loadings ≥ 0.35.

Regression analysis

The predictive values of cultivar resistance, disease, maturity, planting dates, and weather variables were determined according to the FA results. Independency of the principal factors provided by this FA characterized the best predictors of wheat yield in the datasets examined. The multivariate regression analysis justified 90% of variations in yield (F probability = 0.001; R² = 0.90) of eight commercial wheat cultivars planted at different dates across 282 plots over the four growing seasons (Table 5). Therefore, this multivariate regression model was developed to predict wheat yield according to the cultivar resistance index, maturity and planting date, weather, and yellow rust progress predictors. The yellow rust onset time was identified as the most predictive variable describing yellow rust fluctuations across cultivars, planting dates, and growing seasons studied based on the FA results. The selection of predictors to fit the best model was performed using the stepwise procedure. The linear combination of the disease onset time, maturity and planting date, resistance index, and weather predictors corresponded significantly with the dependent variable, wheat yield (kg/ha). Planting date and weather predictor were negatively linked to wheat yield variables. The two-way interaction of disease onset and resistance index predictors, and maturity time predictor corresponded positively with the yield variable. The simple regression analysis demonstrated that the observed data for wheat yield were significantly linked to the data fitted by the yield model (Figure 1). Thus, this model suggested that higher yield levels corresponded with an earlier planting date, further rainy days in spring months, a later maturity of wheat cultivar, a delayed onset of yellow rust, and a greater wheat resistance index.    

Table 5. Multiple regression analysis of wheat yield (kg/ha) based on factor analysis of maturity time, planting date, resistance index, weather, and yellow rust progress predictors (R² = 90%; P < 0.001)