Nonlinearity and Spatial Autocorrelation in Species Distribution Modeling: An Example Based on Weakfish (Cynoscion regalis) in the Mid-Atlantic Bight

Abstract

Nonlinearity and spatial autocorrelation are common features observed in marine fish datasets but are often ignored or not considered simultaneously in modeling. Both features are often present within ecological data obtained across extensive spatial and temporal domains. A case study and a simulation were conducted to evaluate the necessity of considering both characteristics in marine species distribution modeling. We examined seven years of weakfish (Cynoscion regalis) survey catch rates along the Atlantic coast, and five types of statistical models were formulated using a delta model approach because of the high percentage of zero catches in the dataset. The delta spatial generalized additive model (GAM) confirmed the presence of nonlinear relationships with explanatory variables, and results from 3-fold cross-validation indicated that the delta spatial GAM yielded the smallest training and testing errors. Spatial maps of residuals also showed that the delta spatial GAM decreased the spatial autocorrelation in the data. The simulation study found that the spatial GAM over competes other models based on the mean squared error in all scenarios. That indicates that the recommended model not just works well for the NEAMAP survey but also for other cases as in the simulated scenarios.