Effects of prevailing wind direction on spatial statistics of plant disease epidemics

A stochastic simulation model was used to study the effects of the strength of prevailing wind (W), the size/shape (Q) of sampling quadrats and their orientation in relation to the prevailing wind direction (D) on spatial statistics describing plant diseases. Spore dispersal followed a half-Cauchy distribution with median distance ?, which depended on simulated wind speed. The relationship of spatial autocorrelation at distance k (?k) to disease incidence (p) and distance was well described by a four-parameter (?, ?1, ?2, ?3) power-law model; at a given p, ?k declined exponentially with distance. A total of 35 different quadrat sizes, ranging from 4 to 432 plants, were used to sample the simulated epidemics for estimating intraclass correlation (?). The ?-values decreased exponentially with increasing quadrat size; a binary power law model with three parameters (?1, ?4, ?5) successfully related ? to p. In general, the effect of W and D was greatest on the parameters ?, ?1, ?2 and ?3. The effect of W on ?, ?1, ?2 and ?3 depended critically on the spatial pattern of initial infected plants (Y); W had greatest effect for the random pattern. In contrast, the main effect of D and its interaction with W on the parameters ?, ?1, ?2 and ?3 were large and consistent over different initial conditions. Variations in ?1, ?4 and ?5 were predominantly due to Y and Q. Only for ?5 under the clumped pattern was the effect of W very large. For the parameters ?1, ?4 and ?5 there was a large interaction among W, Q and D for the clumped and regular patterns. As expected, in general, the effect of D increased with increasing prevailing wind strength, quadrat size and quadrat length : width ratio. Using square quadrats reduced significantly the effect of W on the parameters ?1, ?4 and ?5; however, the effect of W on ?5 was still very large for the clumped pattern. Sampling perpendicular to the prevailing wind direction generally resulted in larger differences in the nine estimated parameters.