Further Development of a Bayesian Belief Network Model for Estimating Fresh Market Yield In ‘Beauregard' Sweetpotato

A uniform set of management variables was used in the development of a prototype Bayesian Belief Network (BBN) model for predicting fresh market yield in ‘Beauregard’ sweetpotato. Further field validation is necessary to determine the structural soundness as well as extent and limitation of model validity. In addition, further calibration is required in order to adjust system parameters so that simulated results approach that of actual observations. Field experiments were conducted in 2010 to verify model validity using different planting densities (variable in-row spacing, fixed row width). Due to drought conditions during most of the growing season (July-September), comparison of irrigated vs. non-irrigated plots was possible for certain planting dates. Experimental results provided evidence for the mitigating influence of soil moisture regime on planting density and harvest date effects on yield. Under a uniform supplemental irrigation regime, plots with simulated high plant density (i.e., 20 cm in-row spacing=46,876 plants/ha) either had lower yields or were harvested later relative to plots with standard plant spacing (30 cm in-row spacing=31,250 plants/ha). When soil moisture was not limiting, plots with higher planting densities had similar or relatively higher yields relative to plots with standard planting densities. In general, the results from 2010 model validation trials suggested the expansion of the BBN model to include soil moisture and planting density. A prototype model that incorporates soil moisture and planting density variables will be shown. Preliminary calibration and validation using extant data will be shown to demonstrate the applicability of the prototype model under experimental conditions. Future field calibration and validation activities will be outlined.