Modeling spatial heterogeneities in sporadically surveyed Atlantic cod occurrence data on Georges Bank

Following the substantial decline in its population, Atlantic cod on Georges Bank has exhibited increasingly heterogeneous and spatially clustered distributions that pose challenges for traditional parametric and spatial modeling approaches. In this paper, using Northeast Fisheries Science Center (NEFSC) spring bottom trawl survey catch data (1978–2019), we developed a hybrid modeling framework that combines locally weighted regression (LOESS) with a beta-binary mixed model to examine the relationships between cod occurrence and environmental and population-related covariates, as well as the significance of nonparametric trends. Results revealed pronounced nonlinear spatial structure, a significant positive association with population density, and a nonlinear relationship with bottom temperature, with an optimal temperature of around 4–6 ∘C. Location-specific random effects identified persistent hotspots in the northeastern region of Georges Bank. The model also revealed substantial within-stratum heterogeneity, highlighting complex spatial variability in cod occurrence. This framework provides a flexible alternative for modeling irregular spatial patterns in fish occurrence data and offers new insights into habitat structure and survey design under changing environmental conditions.