Exploring spatial non-stationarity of fisheries survey data using geographically weighted regression (GWR): an example from the Northwest Atlantic

Abstract

Windle, M. J. S., Rose, G. A., Devillers, R., and Fortin, M-J. 2010. Exploring spatial non-stationarity of fisheries survey data using geographically weighted regression (GWR): an example from the Northwest Atlantic. – ICES Journal of Marine Science, 67: 145–154. Analyses of fisheries data have traditionally been performed under the implicit assumption that ecological relationships do not vary within management areas (i.e. assuming spatially stationary processes). We question this assumption using a local modelling technique, geographically weighted regression (GWR), not previously used in fisheries analyses. Outputs of GWR are compared with those of global logistic regression and generalized additive models (GAMs) in predicting the distribution of northern cod off Newfoundland, Canada, based on environmental (temperature and distance from shore) and biological factors (snow crab and northern shrimp) from 2001. Results from the GWR models explained significantly more variability than the global logistic and GAM regressions, as shown by goodness-of-fit tests and a reduction in the spatial autocorrelation of model residuals. GWR results revealed spatial regions in the relationships between cod and explanatory variables and that the significance and direction of these relationships varied locally. A k-means cluster analysis based on GWR t-values was used to delineate distinct zones of species–environment relationships. The advantages and limitations of GWR are discussed in terms of potential application to fisheries ecology.

Geographically weighted regression with a non-Euclidean distance metric: a case study using hedonic house price data

“…Jordan 2006, Cho and Gimpel 2009, Lee et al 2009); and (5) ecology and environmental science (e.g. Svenning et al 2009, Windle 2010). However, for GWR studies involving transportation, river/stream networks or some complex terrain conditions, ED metrics may fail to reflect true spatial proximity, and instead, non-ED metrics should be considered, such as road ND, TT, water distance or landscape distance.…”

Spatio-temporal variations in invertebrate−cod−environment relationships on the Newfoundland–Labrador Shelf, 1995−2009

“…spatial non-stationarity) in the relationships under study (Windle et al 2010). Spatial stationarity implies that the parameters of a process (e.g.…”

“…GWR can also be used to explore the scale dependency of ecological interactions by systematically varying the bandwidth size of the spatial kernel at each observation. This technique has been applied to determine the scale at which a species−environment relationship becomes stationary (Osborne et al 2007, Windle et al 2010, Miller & Hanham 2011, Gao et al 2012) and is potentially an important step in building multi-scale predictive models (Graf et al 2005).…”

The Integrated Use of DMSP-OLS Nighttime Light and MODIS Data for Monitoring Large-Scale Impervious Surface Dynamics: A Case Study in the Yangtze River Delta

“…The fundamental principle of GWR is kernel regression, through which model coefficients are determined by neighboring subsets of training samples, and therefore, the problem of spatial nonstationarity can be effectively addressed. However, careful attention should be given to the potential collinearity relationship due to the local nature of GWR [55,56]. As a result, it is usually recommended as a supplementary exploratory technique with global regression models for exploring the variability between variables within a geographical extent [52,56].…”

“…However, careful attention should be given to the potential collinearity relationship due to the local nature of GWR [55,56]. As a result, it is usually recommended as a supplementary exploratory technique with global regression models for exploring the variability between variables within a geographical extent [52,56]. In contrast, SAR is developed using the entire training dataset, though neighboring samples are assigned with greater weights.…”