Comparison of Methods for Forecasting Yellow Rust in Winter Wheat at Regional Scale

Abstract

Yellow rust (YR) is one of the most destructive diseases of wheat. To prevent the prevalence of the disease more effectively, it is important to forecast it at an early stage. To date, most disease forecasting models were developed based on meteorological data at a specific site with a long-term record. Such models allow only local disease prediction, yet have a problem to be extended to a broader region. However, given the YR usually occurs in a vast area, it is necessary to develop a large-scale disease forecasting model for prevention. To answer this call, in this study, based on several disease sensitive meteorological factors, we attempted to use Bayesian network (BNT), BP neural network (BP), support vector machine (SVM), and fisher liner discriminant analysis (FLDA) to develop YR forecasting models. Within Gansu Province, an important disease epidemic region in China, a time series field survey data that collected on multiple years (2010-2012) were used to conduct effective calibration and validation for the model. The results showed that most methods are able to produce reasonable estimations except FLDA. In addition, the temporal dispersal process of YR can be successfully delineated by BNT, BP and SVM. The three methods of BNT, BP and SVM are of great potential in development of disease forecasting model at a regional scale. In future, to further improve the model performance in disease forecasting, it is important to include additional biological and geographical information that are important for disease spread in the model development.