| Resumo: |
In the 2001 growing season, the Asian soybean rust, caused by Phakopsora pachyrhizi, was reported in the south american. Later on, it was reported that disease ocurrence was associated with plant growth stages. Therefore, the growth stage was used as a criterion to indicate the right time for fungicides spray. Warning systems for plant diseases are used to indicate the time of fungicide spray and also to help rationalize the number of sprays and to reduce the production costs. This study aimed to compare the performance of four warning systems to identify the onset of soubean rust in the middle plateau region of "Rio Grande do Sul" as well as to verify whether disease occurrence depends on the plant growth stage. The experiments were conducted at the Experimental School of Agronomy and Veterinary Medicine, University of Passo Fundo, in the 2007/2008 growing season. The performanceof the following warning systems were compared: a) Sum of Daily Values of Severity (SDVS) considering the average temperature during leaf wetness, b) modified SDVS considering the latent period of the fungus nd the average temperature during leaf wetness; c) Daily Value for Infection (DVI) based on the study of Alves (2007), and d) Neural Networks (NN) using the backward propagation of error. In order to simulate the different growth stages, the cultivars Nidera 4910 RR, Relmó Andrea 66 RR, and Munasca RR from the early, medium, and late seasons, respectively, were used. Seeds from these cultivars were sown on the dates of November 11, December 3 and December 21 in the year of 2007. The experimental design used was the randomized blocks in split plots with three replications. The plots were the cultivars and the subplots were the sowing times in a total of 27 plots. The occurrence of soybean rust was observed in the first and seconde sowing time for the three cultivars, demonstrating that disease occurrence does not depend on the plant growth stage. For comparison of early warning systems, neural networks and values for DVI had higher accuracy to detect the occurrence predicted over the observed |
|---|
