PLN Models
Given potential convergence issues with the PLN models, we restricted our analysis to species that appeared in the dataset at least four times in each season. This restriction resulted in 36 species in the Fall with four unique species (i.e. species found only in the Fall), and 35 species in the Spring with three unique species, and a total of 32 species present in both seasons (top panel of Figure 3).
The three PLN models with highest BIC scores in both Fall and Spring were Site name , Site name + Water temperature , andSite name + Dissolved oxygen (Table 1). For model validation in the Fall, the model with the lowest RMSE and RMSE obs>0 wasSite name + Water temperature . Although Site name had the highest BIC score in the Fall, Site name + Water temperature had only a marginally lower score and it performed the best during model validation. Therefore, Site name + Water temperature was selected to construct the Fall species interaction network. In the Spring,Site name had both the highest BIC score as well as the best predictive performance, and was hence chosen as the best model for the Spring network.