A statistical model for monitoring shell disease in inshore lobster fisheries: A case study in Long Island Sound

Abstract

The expansion of shell disease is an emerging threat to the inshore lobster fisheries in the northeastern United States. The development of models to improve the efficiency and precision of existing monitoring programs is advocated as an important step in mitigating its harmful effects. The objective of this study is to construct a statistical model that could enhance the existing monitoring effort through (1) identification of potential disease-associated abiotic and biotic factors, and (2) estimation of spatial variation in disease prevalence in the lobster fishery. A delta-generalized additive modeling (GAM) approach was applied using bottom trawl survey data collected from 2001–2013 in Long Island Sound, a tidal estuary between New York and Connecticut states. Spatial distribution of shell disease prevalence was found to be strongly influenced by the interactive effects of latitude and longitude, possibly indicative of a geographic origin of shell disease. Bottom temperature, bottom salinity, and depth were also important factors affecting the spatial variability in shell disease prevalence. The delta-GAM projected high disease prevalence in non-surveyed locations. Additionally, a potential spatial discrepancy was found between modeled disease hotspots and survey-based gravity centers of disease prevalence. This study provides a modeling framework to enhance research, monitoring and management of emerging and continuing marine disease threats.

Mesoscale climatic impacts on the distribution ofHomarus americanusin the US inshore Gulf of Maine

“…depth and temperature) are often correlated. Variance inflation factors (VIF) were therefore calculated and variables with VIF value > 3 were removed to minimize collinearity and improve model performance (Table 1; Zuur et al 2007;Tanaka et al 2017). Following Sagarese et al 2014, boosted regression tree (BRT) analysis was used to identify potentially significant bivariate interaction terms, which were incorporated in the GAM fitting process.…”

“…Stepwise backward selection using chi-square statistical tests and Akaike's information criteria (AIC) was used to reduce a full model (with univariate and bivariate terms identified through VIF and BRT analyses) to a parsimonious final model with lowest AIC and only significant variables (Tanaka et al 2017). The stepwise model selection procedure was repeated as long as the removal of the variable with the lowest significant p-value reduced AIC.…”

“…In this procedure, 80 % of the original data was randomly partitioned for model calibration (training set), while the remainder was used for model validation (testing set). The final model was fit to each training set and used to predict the corresponding testing set (Miller and Franklin 2002;Brotons et al 2004;Tanaka and Chen 2016;Tanaka et al 2017). The cross-validation procedure was repeated with random partition of testing and training data in each iteration.…”

“…Final GAMs were also used to project how lobster distribution may change under two thermally contrasting climatologies (Sagarese et al 2014). Predictive fields were interpolated using ordinary kriging to describe spatial variability in lobster catch density (Froeschke and Froeschke 2016;Tanaka et al 2017). Ordinary kriging procedures were conducted via the automap package (Hiemstra et al 2008) and kriged maps were prepared at 0.03 × 0.03 latitude/longitude grid.…”

Quasi‐Stationary Atlantic Cod Bycatch Estimation in the Maine American Lobster Homarus americanus Trap Fishery

“…All smoothed terms were given five knots (maximum degrees of freedom) to balance flexibility with predictive capacity (Tanaka et al. ). The second‐stage GAM (GAM2) modeled abundance ( M ) conditional on presence using an identity‐link function and log‐transformed catch data (Ohshimo et al.…”