Sequential sampling for incidence of phomopsis leaf blight of strawberry

ABSTRACT Sequential sampling models for estimation and classification were developed for the incidence of strawberry leaflets infected by Phomopsis obscurans. Sampling protocols were based on a binary power law analysis of the spatial heterogeneity of Phomopsis leaf blight in commercial fields in Ohio. For sequential estimation, samples were collected until mean disease incidence could be estimated with a preselected coefficient of variation of the mean (C). For sequential classification, samples were collected until there was sufficient evidence to classify mean incidence as being below or above a threshold (p(t)) based on the sequential probability ratio test. Monte-Carlo simulations were used to determine the theoretical average sample number (ASN) and probability of classifying mean incidence as less than p(t) (operating characteristic) for any true value of incidence. Estimation and classification sampling models were both tested with bootstrap simulations of randomly selected data sets and validated by data sets from another year that were not utilized in developing the models. In general, achieved (or calculated) C after sequentially sampling for estimation was close to the preselected C of 0.2, and mean incidence was estimated with little bias. Achieving a C of 0.1 with less than 75 sampling units (the nominal value for many original data sets) was more problematical, especially with true incidence less than 0.2. ASN for classification was only 9 to 18 at disease incidence values near p(t), and approximately five or less at incidence values far from p(t). Correct classification decisions were made in over 88% of the validation data sets. Results indicated that it is possible to estimate Phomopsis leaf blight with high precision and with high correct classification probabilities.