Robustness to link removal: data and computer codes

We implemented three interaction extinction sequences which intended to produce best-, worst- and intermediate case scenarios, as has been done in node extinction based scenarios. The first sequence always removed the interaction between the most connected species (hereafter, most connected first) and its most connected mutualistic partner. When more than one species had the same degree of connectivity, we randomly chose one of them. This process was repeated as many times as the number of interactions in the network until the last interaction (the least connected species and its least connected mutualistic partner) was removed. . Note that, contrary to NBRA, interaction removal did not follow a predefined sequence because the degree of species was updated after each interaction extinction. The second sequence always removed the interaction between the least connected species (hereafter, least connected first) and its least connected mutualistic partner, and then continued by removing the remaining interactions belonging to the least connected species and its least connected mutualistic partner at each iteration step. The third sequence, named random, removed interactions at random, to simulate an intermediate extinction scenario.