Generalised Additive Model Improves Estimates of Vibrio Species Abundance in Penaeus vannamei Boone, 1931 Biofloc Production System

Abstract

Environmental factors influence the abundance of Vibrio species in shrimp culture systems. Prediction of the abundance of presumptive Vibrio species can help prevent the occurrence of bacterial diseases as this will provide insights about when and which environmental factors to manage. In this study, the parametric linear regression model (LRM) and negative binomial model (NBM), and semiparametric generalised additive model (GAM) were used to identify correlations and predict changes of Vibrio abundance with physicochemical and biological water parameters. Water parameters were recorded from three 300 m2 biofloc ponds stocked with Penaeus vannamei Boone, 1931, at 500 individuals.m-3 over four culture run periods. Each culture run lasted for 16 weeks. Imputed data were initially subjected to univariate analysis and Pearson’s correlation analysis. The abundance of presumptive Vibrio species was found to be highly correlated with alkalinity, pH, and phytoplankton density. GAM performed best among the three models based on Akaike’s information criterion (AIC), having the smallest value of 5,743.222 compared to 6,572.014 and 5,857.997 values for ordinary LRM and NBM, respectively. It also had the largest deviance explained statistic with 41.2 % of the deviance reduced by including the predictors compared with ordinary LRM and NBM with only 16.04 % and 14.5 % deviance reduced, respectively. GAM introduced flexibility that predicts the dependent variable better in terms of statistical significance than LRM and NBM. It is important to consider using a semiparametric modelling approach as a tool for aquaculture management.