A model for assessing the effect of distance on disease spread in crop fields

Abstract

Standard logistic regression is applied to model spatio-temporal plant disease spread in crop fields. The binary response variable observed at a given point in time is the disease status Y, taking on the value Y=1 for plants newly infected and Y=0 for healthy plants. The predictor variable used and tested here is the distance of a newly infected or healthy plant to the closest plant that was previously diseased, hypothesizing that disease spread originates largely from foci within the crop field. The method is illustrated with three sets of artificial data, and with four real data sets from published studies and our own studies. In the artificial data sets a random disease spread, a distance-dependent, and a both distance- and orientation-dependent disease spread were simulated, and then clearly identified as such by the corresponding logistic regression models. For one of the real data sets (citrus variegated chlorosis) the results correspond directly to the published results where clustering of disease incidence was found, which is one form of distance dependence. For the other published data set (papaya ringspot) no statistically significant effect of distance was identified. The non-randomness of disease spread as identified in the original source is supposedly due to reasons other than the distance effect investigated here. In one of the tomato fields analyzed, a clear distance dependence of the tomato yellow mottle disease (ToYMoV) spread was found. In the other field, where disease progress was very fast, no effect of distance could be observed. The proposed method is a flexible analysis tool. It does not require the plants in a regular lattice; rather the plant's location is defined by its metric coordinates. Like in any spatio-temporal analysis, the definition of the time interval between measurements plays an important role. Too short or too long intervals will make it impossible to identify a systematic pattern in disease spread. The method tests not only for non-randomness of disease spread in general, but it also helps to identify the effect of specific predictor variables. Thus, a specific hypothesis can be researched, helping to obtain insight into the mechanisms of disease spread and helping to develop specific disease management approaches. Previous article in issue