Use of geographically weighted regression to investigate spatial non-stationary environmental effects on the distributions of black sea bass (Centropristis striata) and scup (Stenotomus chrysops) in the Mid-Atlantic Bight, USA

Regression analyses used to describe a species' distribution in relation to environmental conditions often employ global models based on the assumption that the influence of environmental variables is constant or uniform across geographic space. We tested the assumption of spatial stationarity in relationships between environmental variables and the distributions (encounters/non-encounters) of black sea bass (Centropristis striata) and scup (Stenotomus chrysops) caught during fishery-independent bottom trawl surveys conducted in US shelf waters of the Mid-Atlantic Bight (MAB). Our tests employed global logistic generalized linear models (GLMs), global logistic generalized additive models (GAMs), and local logistic geographically weighted regression models (GWR). GWR models had a higher goodness-of-fit and predictive accuracy than GLMs and GAMs for both species, and the residuals of the GWR models also demonstrated less spatial autocorrelation. Results of GWR analyses indicated that local relationships between species' distributions and environmental conditions were significant and varied spatially in strength and direction between sampling locations. Local GWR parameter coefficients and t-values were mapped for further evaluation and a k-means cluster analysis using t-values was performed to identify any areas within the survey region with distinct species and environmental relationships. Overall, our results provided new insight about local spatial variability in habitat associations for black sea bass and scup and how environmental conditions shape their distributions in the MAB. Further, our results suggest that GWR should be considered for exploring spatial varying habitat relationships for other species in the MAB as well.

Automated summary

The study investigated the distribution of black sea bass and scup in the US shelf waters of the Mid-Atlantic Bight, finding that local relationships between environmental conditions and species distributions varied spatially. Local logistic geographically weighted regression models showed higher accuracy in predicting species distributions, indicating significant spatial variability in habitat associations for the two species. GWR should be considered for exploring spatial varying habitat relationships for other species in the MAB.