Hourly precipitation (sum) and temperature (average) data were plotted starting 1 March and ending 15 June (Fig. 1). This period corresponds roughly with plant growth stages 29–69 in winter wheat in the area of observation. We decided to start before growth stage 31 is reached, because of the time lag between infection and symptom expression/reaching of the control threshold. Then, the time shift between each wet hour and the point of time at which the control threshold was observed over the period 2015–2016 was estimated using the empirical relationship published in the first paper of the present series (Beyer et al. 2022). A histogram was then generated from the time points at which the reaching of the control threshold was forecasted (Fig. 1A). A graph of the type in Fig. 1A will subsequently be referred to as a risk profile. The more wet hours are emphasized by the respective temperature conditions on the same time slice, the higher the bars of the histogram become. The bar width of the histogram was adjusted to 125 h, corresponding to the effect of one rank in the winter wheat cultivar susceptibility ranks regularly published by BSA (2017 ff.). For instance, in the period 2005–2016, a cultivar with susceptibility rank 4 reached the control threshold on average 125 h later than a cultivar with susceptibility rank 5 (Beyer et al. 2022). The default setting of the model estimates the risk profile (Fig. 1A) for cultivars with susceptibility rank 5 (the most frequent rank of the cultivars grown in the region of observation). For cultivars with other susceptibility ranks, the risk profile is shifted, with the size of the shift corresponding to the susceptibility rank of the cultivar grown. For instance, for a cultivar with susceptibility rank 4 (lower than average), the default histogram is shifted to later dates by one bar width (= 125 h).
The period in which fungicide use is allowed (between the growth stages 31 and 65–69, depending on fungicide) is estimated from the sowing date according to Beyer et al. (2012). Therefore, users of the forecast model need to give the sowing date of the winter wheat. The precision of the estimates can be enhanced by manually entering the current growth stage of the crop, thus requiring it to be determined in (each) field concerned. Users can decide whether they want to enter the sowing date or the growth stage of the winter wheat, although the growth stage is preferred due to better precision. During the period when fungicide use is permitted, spraying is recommended, before a red bar is reached (Fig. 1).
A typical risk profile (Fig. 1A) contains several peaks. Each peak represents a period of time, during which it is highly likely that the control threshold for Z. tritici will be reached. No warning is displayed for peaks that occur before or after fungicide spraying is allowed (for an example, see Fig. 1A). A histogram was generated, showing the peak heights from the period in which fungicide use was allowed (Fig. 2). A clear maximum was observed at 15 (Fig. 2). Hence, we assume that a peak height of 15 or higher indicates the reaching of the control threshold and thus the need to apply a fungicide spray. Among the six peaks that were smaller than 15 but associated with a breaking of the control threshold (Beer 2005), four were observed in the year 2014 (Fig. 2).
The pattern used to construct the model was detected in data acquired in field experiments during the period 2005–2016 (Beyer et al. 2022). Data from 2017–2019 were used for external validation. If a predicted point of time was within ± 11 days (being equivalent to 50% of the efficacy duration of a systemic fungicide (Greiner et al. 2019)) of the observed point of time, the prediction was accepted as correct, splitting the risk of applying either too early or too late. If predicted times were outside of the period specified above, the predictions were categorized as false. Of 13 cases, 2 were identified as false with external data (Table 1). This corresponds to 84.6% correct predictions. Predicted and observed dates were closely related (r = 0.92, P < 0.0001, Fig. 3).
Critical evaluation of model performance
In contrast to previous models, ShIFT forecasts the time for which a fungicide spray is needed directly. Users do not need to interpret epidemiological outputs to identify a suitable time frame for a fungicide spray based on the control threshold concept of integrated pest management.
The winter wheat crop is particularly sensitive toward septoria leaf blotch between plant growth stages 31 and 65. Therefore, peaks occurring before GS31 or after GS65 indicate favorable weather conditions for epidemics but do not indicate a need for fungicide use. Note that in many countries, including EU countries, fungicide spray application outside of the mentioned time frame is illegal due to the registration conditions of the products.
In the year 2014 (that was part of the data set used to detect the pattern used for the model), a massive yellow rust epidemic was observed in Luxembourg for the first time (Dam et al. 2020), probably due to the spread of more aggressive strains throughout Europe at that time (Aslanov et al. 2019). The poor performance of the model in that year, as indicated in Fig. 2, suggests that other diseases can seriously interfere with the percentage of correct model outputs at high disease levels. Thus, the model described here should not be used in other regions without locally validating model outputs with field observations, particularly if fungal plant pathogens other than Z. tritici are dominant.
The model allowed correct predictions in 84.6% of cases, while 15.4% of the cases were predicted falsely. The average deviation between the observed and predicted dates of relevant epidemic outbreaks was 0.62 ± 2.4 days with a maximum deviation of 19 days. Observed and predicted dates were closely correlated (r = 0.92, P < 0.0001). The model demonstrated considerable prognostic power when being tested with independent new data, and however, the possibility of outliers being falsely classified cannot be denied.
Considerations on fungicide efficacy
Greiner et al. (2019) determined the period of fungicide efficacy for Bravo 500 (a representative of contact fungicides containing chlorothalonil as active ingredient), Epoxion (a representative of systemic azole fungicides containing epoxiconazole as active ingredient) and Imbrex (a representative of systemic succinate dehydrogenase inhibitors containing fluxypyroxad as active ingredient) at full dose rates. The effective period of the fungicides ranged from 16 days for the contact fungicide Bravo to 22 days for the systemic fungicide Imbrex. Due to the ban of chlorothalonil and epoxiconazole, the situation in Luxembourg was greatly simplified, such that only products from the group with an effective period of approximately 22 days remained on the market and were therefore considered in the ShIFT model.
Considerations on the number of sprays recommended–potential for pesticide savings
Under weather conditions that continuously allow for infections and disease development, three sprays may be applied for the best protection of the upper three-leaf layers that are largely responsible for grain filling, namely one spray after the formation of each of the leaf layer. Under weather conditions that do not allow for infections and disease development continuously, the number of sprays and thereby costs may be reduced without giving rise to an epidemic. ShIFT recommended on average 1.4 sprays on winter wheat per season against Zymoseptoria tritici (Fig. 4). In a survey with 108 participants, Luxembourgish farmers responded that they believed that on average, 1.6 fungicide applications are needed per season in winter wheat (Beyer et al. 2019). In neighboring Germany, 2.3 sprays are applied on average per season (https://papa.julius-kuehn.de/index.php?menuid=46). The difference between the 1.4 sprays per season recommended by ShIFT, the 1.6 sprays mentioned by Luxembourgish farmers and the 2.3 sprays applied on German farms can probably be attributed to the need for controlling other diseases besides leaf blotch. However, most commercial fungicides show efficacy against several fungal pathogens, and therefore, an additional spray is needed only if the temporal distance between the occurrences of the diseases is larger than the period of efficacy of the fungicide. Besides leaf blotch, yellow rust has also often been observed since 2014 in Luxembourg (Aslanov et al. 2019). In the 2005–2017 period, leaf blotch and yellow rust reached their respective control thresholds in 17 of 62 cases. The breaking of the control threshold for both diseases was within the period of a systemic fungicide in 14 cases. In 3 of 62 cases (= 5%), the temporal distance between the occurrence of the two diseases was too large to control both diseases with the same spray. In approximately one out of five years, weather conditions for Fusarium head blight were favorable enough to allow for at least local mycotoxin contamination (Pallez et al. 2021). The 1.4 sprays per season recommended by ShiFT against leaf blotch in winter wheat seem to be roughly realistic with regard to farmers’ opinions and factual use, leaving little room to further reduce fungicide use in winter wheat without accepting avoidable losses.
Availability of the model
ShIFT is freely available in the year 2022 for a test period in English, French, German and Luxembourgish at https://shift.list.lu/ (Identifier: JPDP, Password: DPG_2022).