Weather fluctuation can override the effects of integrated nutrient management on fungal disease incidence in the rice fields in Taiwan

Plant material

Rice (Oryza sativa L.) plants used for the experiment were from the collection of Taiwan Agricultural Research Institute. The rice variety (Tainan No. 11) used in this study has enhanced resistance to rice blast. The use of plant materials complies with international, national, and/or institutional guidelines and legislation.

Field area

This study was carried out in experimental rice fields under low-external-input and conventional farming in central Taiwan (23.5859 N, 120.4083 E; 8.0 ha). The annual average temperature ranged from 23 to 25 °C, the annual average relative humidity ranged from 75 to 92%, and the annual rainfall ranged from 1020 to 2873 mm year⁻¹ (average data between 2006 and 2016 measured at a nearby weather station; Fig. 1). The experimental paddy plots were defined by considering the typical dimensions of the agricultural fields in Taiwan (0.5 to 1.0 ha). A long-term experiment was conducted from 2006 to 2016 to study the effects of different agronomic management on biodiversity, productivity, and environment, including traceability system, soil fertility, nitrogen leaching, production costs, disease incidence and severity, the abundance of pest and beneficial insects, and weed dynamics.

The treatments consisted of conventional farming with high chemical fertilizer input (CF) and low-external-input farming with low fertilizer input (LF). In the CF farming, we followed the fertilizer recommendations that are constructed to meet the nutrient requirements of the crop. In the LF farming, the chemical fertilizers were largely reduced compared to the recommendation (see next paragraph for the details). The experiment was conducted as a randomized complete block design (RCBD) with four replicates. In agricultural experiments, the RCBD is a standard procedure by grouping experimental units into blocks. For example, the design can control variation in the experiments by considering spatial influences and adjusting the effects of target factors in fields. Each experimental unit consisted of a 0.58 ± 0.16 ha of the area of the field. Additional nutrient management in the LF system includes (1) nitrogen-fixing and cover crops, (2) manure and compost applications, (3) plant and soil nutrient analyses for adjusting fertilization, and (4) reduced tillage. Soil-available potassium gradually decreased during the 10-year study period in the area of the LF system. Over the study period, the LF system achieved the similar level of crop production as that of the CF system (Fig. S1).

In our study area, there were two growing seasons within a year: one in the first half of the year (from February to June) and one in the second half (from August to December). The ground fertilizers were applied before rice seedlings were transplanted, followed by additional fertilizations during the tillering and boosting stages. The total amount of fertilizers used for the CF system included 140–180 and 120–140 kg N ha⁻¹, 70–72 and 60 kg P₂O₅ ha⁻¹, and 85 and 60 kg K₂O ha⁻¹ for the first and second seasons, respectively. For the LF system, 100 and 80 kg N ha⁻¹, 30 and 30 kg P₂O₅ ha⁻¹, and 30 and 30 kg K₂O ha⁻¹ were applied in the first and second seasons, respectively. The larger amount of fertilizers for the first season was due to its longer duration. For each rice growing season, fungicides were applied to both farming systems once during the boosting stage. During the fungicide application, a 10% mixture of Cartap plus Probenazole or 6% probenazole for rice blast (both 30 kg ha⁻¹) and 1.5% Furametpyr for sheath blight (20 kg ha⁻¹) were used.

Rice disease monitoring

The major rice disease (rice blast; Fig. S2) was monitored biweekly in the CF and LF systems over the two growing seasons per year, with each growing season including (in chronological order) the tillering, flowering, and maturing stages. There was a total of 123 occasions during our study. The plants were disease free when planted out. When the lesion of the rice blast began to appear in the fields from the tillering stage to the maturing stage, the effects of the two treatments (CF and LF systems) in the paddy fields on the disease incidence of rice blast (caused by Pyricularia oryzae) were investigated. For each plot (or experimental unit), the incidence of rice disease was randomly examined at 5 points and for 25 plexuses (i.e., each derived from one primary tiller) per point. The disease incidence was quantified as the percentage of infected plexuses that were determined based on the presence of infected leaves.

The area under the disease progress curve (AUDPC) was used to quantify disease incidences over time, and the relative AUDPC ((RAUDPC)) was used because of unequal sampling duration in the growing seasons during our study period. For each plot (or experimental unit), we used the (RAUDPC) to summarize the incidences of disease during each growing season as follows:

where ({y}_{i}) and ({t}_{i}) are the disease incidence (%) and time (day) at the (i)th observation, respectively, and (n) is the total number of observations.

Bayesian modeling

We built a mechanistic model that was applied to assess the interplay within a network of relationships among weather fluctuation, farming system, and disease incidence in the paddy fields. The model describes how (1) temperature and relative humidity together influence the development of primary inoculum, (2) rainfall detaches the fungal spores on the host tissues, and (3) rainfall and wind catch the airborne spores onto the leaf area. These environmental processes determine the disease incidence in the model. In addition, this model considers that farming systems can suppress or accelerate disease incidence. By fitting our model to the observed incidence, Bayesian inference was used as the parameter estimation technique for the models. In addition, we tested the alternative mechanistic hypotheses based on a model-selection criterion and cross vaidation (see subsequent paragraphs).

With a linearity assumption, the incidences of disease (RAUDPC) were modeled as an inverse-logit function of the progress rate of the development of primary inoculum ((IP) with values between 0 and 1) and the net catchment of the airborne spores by rainfall and wind ((CT) with values between 0 and 1; when subtracting the detachment of spores by rainfall from the host tissue) as follows:

where ({a}_{f}), ({b}_{1}), and ({b}_{2}) describe the constant baseline for different farming systems ((f) = the CF or LF system), the direct effect size of (avg_IP), and the mediating effect size of (avg_CT) through (IP), respectively. The two parameters ((avg_IP) and (avg_CT)) are averaged (IP) and (CT) during the growing season, respectively (see below for details). The effect sizes ({b}_{1}) and ({b}_{2}) have values more than zero. The constant baseline allows the management-specific acting in the model when they can influence the disease incidence.

The process rate (IP) was simulated as a function of the temperature response ((fleft(Tright)) with values between 0 and 1) and hourly air relative humidity ((RH,) %) as follows²⁰:

where ({RH}_{t}) the threshold of air relative humidity and (LW) is the leaf wetness. Here, we used the RH condition because of the data availability. The relative humidity if lower than the threshold inhibits inoculum development⁴⁷. Otherwise, the development of the primary inoculum is promoted, relying on the change in temperature. Herein, (fleft(Tright)) describes the development of the primary inoculum under temperature change¹⁹:

where (T) is the hourly air temperature (°C). We denoted the ({T}_{min}), ({T}_{max}), and ({T}_{opt}) pathogen-specific cardinal temperatures (minimum, maximum, and optimal temperatures, respectively) for inoculum development. To meet valid values of (fleft(Tright)), the minimum and maximum temperatures are considered as out of the two ends of the measured range. The temperature response function (fleft(Tright)) is also associated with the sporulation efficiency²⁰.

The net catchment of the airborne spores ((CT)) was simulated as a nonlinear function of rainfall and wind velocity as follows^20,48:

where ({D}_{R}) (between 0 and 1) describes the direct detachment of spores by rainfall from the host tissue, in which (R) is the daily rainfall (mm) and (R50) is the rainfall able to provoke detachment of 50% of the spores. ({C}_{R}) and ({C}_{W}) (both between 0 and 1) describe how rainfall and wind allow airborne spores caught on the leaf area, respectively. ({C}_{act}) (between 0 and 1) is the actual portion of spores caught at a rainfall of 1 mm. (W) is the daily wind velocity (m s⁻¹), and ({C}_{s}) and ({C}_{m}) (both > 0) are the steepness and midpoint parameters to control the portion of spores caught by the wind, respectively.

The Bayesian framework ‘Stan’⁴⁹ and its R interface ‘RStan’⁵⁰ were used to construct and fit the models. There were two competing models: either considering the difference between the CF and LF systems by not fixed to the same values of the constant baseline ({a}_{f}) in Formula (2) or not. For each model, four Markov Chain Monte Carlo (MCMC) chains (for numerical approximations of Bayesian inference) ran, each with 5,000 iterations, and the first half of the iterations of each chain were discarded as burn-in. The R-hat statistic of each parameter approaches a value of 1, indicating model convergence. With a total of 2,000 samples, collected as one sample for every 5 iterations for each chain, the model parameters and their posterior distribution were estimated. To compare the two competing models, we calculated the widely applicable information criterion (WAIC) using the R package ‘loo’⁵¹. The best model was determined based on the lowest WAIC. By using the same R package, we also performed the approximate leave-one-out cross-validation (LOO-CV) to estimate the predictive ability of the two Bayesian models. Here, we used the expected log predictive density (ELPD) to be the predictive performance.

Compliance with ethical standards

The authors declare that they have no conflict of interest. This article does not contain any studies involving animals performed by any of the authors. This article does not contain any studies involving human participants performed by any of the authors.

Source: Ecology - nature.com