Fitting and validation summary of the final GAMs.

The selected set of predictors, corresponding F/Chi-sq. values and total deviance explained (DEV) for each model, where models I-III use fish abundance estimates as response variables while model IV use presence-absence data only. Model fit and predictive accuracy was assessed by correlations (r) and linear regression between observed and fitted values (A), as well as between observed and predicted values (B), when fitted on only half the data set and predicting the remaining part. Finally, the explained variance (r2) of the linear regressions, the root mean square error (RMSE) and average error (AVE) between predicted and observed values, as well as the number of observations (n) are shown.