Sampling strategies that effectively assess disease intensity in the field are important to underpin management decisions. To develop a sequential sampling plan for the incidence of Cercospora leaf spot (CLS), caused by Cercospora beticola, 31 table beet fields were assessed in New York. Assessments of CLS incidence were performed in six leaves arbitrarily selected in 51 sampling locations along each of the three to six linear transects per field. Spatial pattern analyses were performed, and results were used to develop sequential sampling estimation and classification models. CLS incidence (p) ranged from 0.13 to 0.92 with a median of 0.31, and beta-binomial distribution, which is reflective of aggregation, best described the spatial patterns observed. Aggregation was commonly detected (>95%) by methods using the point-process approach, runs analyses, and autocorrelation up to the fourth spatial lag. For SADIE, 45% of the datasets were classified as a random pattern. In the sequential sampling estimation and classification models, disease units are sampled until a prespecified target is achieved. For estimation, the goal was sampling CLS incidence with a preselected coefficient of variation (C). Achieving the C = 0.1 was challenging with less than 51 sampling units, and only observed on datasets with an incidence above 0.3. Reducing the level of precision, i.e. increasing C to 0.2, allowed the preselected C be achieved with a lower number of sampling units and with an estimated incidence (p̂) close to the true value of p. For classification, the goal was to classify the datasets above or below prespecified thresholds (pt) used for CLS management. The average sample number (ASN) was determined by Monte Carlo simulations, and was between 20 and 45 at disease incidence values close to pt, and approximately 11 when far from pt. Correct decisions occurred in over 76% of the validation datasets. Results indicated these sequential sampling plans can be used to effectively assess CLS incidence in table beet fields.
Empirical Bayes Prediction in a Sequential Sampling Plan Based on Loss Functions
