Nonlinearity and Spatial Autocorrelation in the Species Distribution and Modeling: An Example Based on Atlantic Weakfish Cynoscion regalis

Spatial autocorrelation (SAC) is frequently observed in fishery’s data in which samples collected are not independent from each other at nearby locations. I examined catch rates of Atlantic weakfish (Cynoscion regalis) along west coast of Atlantic Ocean offshore area, which varies substantially over space and time. Data used in this study were obtained from the Northeast Area Monitoring and Assessment Program (NEAMAP). Because of high percentage of zero catches in the samples, a delta approach was used. Five models were used to test spatial autocorrelation: 1) delta model comprising two generalized linear model; 2) delta model comprising two generalized additive model; 3) simultaneous autoregressive error model combined with auto covariate model; 4) SAR lag model combined with auto covariate model; 5) SAR mixed model combined with auto covariate model. The result from 3-fold cross-validation indicated that the Delta-GAM yielded smallest training error and testing error followed by SAR error model combined with auto covariate model. The residual maps also indicate that the Delta-GAM and SAR error model managed to decrease the spatial autocorrelation in the data. We suggest Delta-GAM and SAR error model with auto covariate regression to be alternative to deal with data with nonlinearity and spatial autocorrelation.