Climatic drivers of Verticillium dahliae occurrence in Mediterranean olive-growing areas of southern Spain

Verticillium wilt, caused by the soil-borne fungus Verticillium dahliae, is one of the most harmful diseases in Mediterranean olive-growing areas. Although, the effects of both soil temperature and moisture on V. dahliae are well known, there is scant knowledge about what climatic drivers affect the occurrence of the pathogen on a landscape scale. Here, we investigate what climatic drivers determine V. dahliae occurrence in olive-growing areas in southern Spain. In order to bridge this gap in knowledge, a landscape-scale field survey was carried out to collect data on the occurrence of V. dahliae in 779 olive groves in Granada province. Forty models based on competing combinations of climatic variables were fitted and evaluated using information-theoretic methods. A model that included a multiplicative combination of seasonal and extreme climatic variables was found to be the most viable one. Isothermality and the seasonal distribution of precipitation were the most important variables influencing the occurrence of the pathogen. The isothermal effect was in turn modulated by the seasonality of rainfall, and this became less negative as seasonality increases. Thus, V. dahliae occurs more frequently in olive-growing areas where the day-night temperature oscillation is lower than the summer-winter one. We also found that irrigation reduced the influence of isothermality on occurrence. Our results demonstrate that long-term “sophisticated” climatic factors rather than “primary” variables, such as annual trends, can better explain the spatial patterns of V. dahliae occurrence in Mediterranean, southern Spain. One important implication of our study is that appropriate irrigation management, when temperature oscillation approaches optimal conditions for V. dahliae to thrive, may reduce the appearance of symptoms in olive trees.


Introduction
technique was designed in order to select the olive groves [14]. This technique is commonly 107 applied to heterogeneous populations and collecting data based on sample sizes that are 108 proportional to the relative size of the strata [40]. Proportions in the first sampling level were based on the surface area (ha) of olive groves in each olive-growing zone. The second level 110 consisted of olive groves whose probability of infection was proportional to their size (PPS). PPS 111 gives a probability of selecting a sampling unit (e.g., olive grove) in proportion to the size of its 112 population [40]. It was assumed a priori that the probability of an olive grove being infected with 113 V. dahliae was 50%. Finally, the sampling survey for evaluating occurrence of V. dahliae in the 114 study area was performed in 779 olive groves (Fig 1), over 2833 ha with 139 olive trees/ha and a 115 standard error of ≤ 2.48 [14].   Likewise, the origin of the propagation material also had a bearing on this [33]. Finally, since 153 plantation size was expected to positively affect pathogen occurrence, the olive grove area (in 154 ha) was also included as a covariate.

155
Modelling approach 156 Although pathogens are mainly dependent on the occurrence of their hosts, climate is also 157 an important factor that influences their distribution on a wider scale [45]. Indeed, complex where π is the probability of V. dahliae occurrence. In addition, Akaike weights (w i ) were calculated to rank models with ΔAIC < 2, in turn. In order 182 to verify whether the interactions between climatic and abiotic factors were significant, we 183 included interaction terms between climatic variables, watering and plant material origin in the 184 top-ranked model (see S1 Table for details). 185 We assessed how accurate the predictions of the model were by comparing the Log-Loss

186
[49] achieved by the null-model (i.e., intercept-only) as a baseline with that of the top-ranked 187 model using a 10-fold cross-validation of the sampled data. Log-Loss measures the uncertainty 188 of fitted probabilities by comparing them to actual observations. Each data partition was made by 189 randomly selecting 75% of the locations as training data and the remaining 25% as test data.

190
Finally, a key assumption for regression is that model residuals were not correlated, we checked   (Table 1). These models consisted of a combination of seasonal (rainfall seasonality) and  dahliae occurrence according to the most plausible model (see Table 1). Effect sizes are shown 225 on a cloglog scale. Irrigation and annual pruning were treated as reference levels for watering 226 and plant material origin, respectively.

227
As expected, the predictive capacity of the top-ranked model was better than that of the pathogen on a landscape scale. We found that long-term "sophisticated" climatic factors rather 237 than 'primary' variables, such as annual trends, provided a more comprehensive explanation of 238 the spatial patterns of V. dahliae occurrence in olive-growing areas in southern Spain.

239
In the Western Mediterranean basin, seasonal rainfall regimes are equinoctial, with a 240 rainy season in autumn and another secondary rainy period in spring [ We expected annual precipitation to be a significant factor affecting the occurrence of the 271 pathogen in olive groves. However, we found that models that included rainfall seasonality had a 272 better fit. This finding may be due to the fact that optimal soil moisture and temperature that daily -as opposed to both weekly and biweekly-irrigation helps Verticillium wilt thrives.
They also advised that in order to reduce symptoms of the disease, soil water content should be 288 lower than 24% and irrigation should be done on a biweekly basis. However, it should be 289 stressed that although less frequent irrigation could reduce the risk of Verticillium wilt, this will 290 need to be balanced with the water needs of the olive trees in order to achieve good and high- periods for crop productivity, during which time soil moisture must be kept optimal.

298
Unfortunately, these periods also coincide with Verticilllium wilt development. Therefore, 299 studies need to be made to determine disease development, soil moisture levels and rainfall 300 during those critical periods to assess influences on the pathogen and olive tree yields.

301
Evidentially the response of any given soil-borne disease to the range of ways of managing 302 irrigation varies widely and must be addressed for each individual plant-pathogen system [70].  Table 1 in the main manuscript). We ordered the models by 519 their AIC values and checked whether the interaction term was significant or not.  Table 1 in the main manuscript) including an 522 interaction term between isothermality and watering. (b) Spline correlogram of the residuals 523 using the function spline.correlog in the "ncf" R package [1]. The spatial dependence is tested as 524 a continuous function of distance. The gray shadows represent the 95 % confidence interval.