Interactions between species are often assumed to occur whenever species pairs co-occur, allowing us to reconstruct local ecological networks from a deterministic \'metaweb\'. Thus, a species\' set of metaweb interactions defines its Eltonian niche - the biotic conditions under which it can exist. However, both species and interactions may also have Grinnellian niches, reflecting relationships between probability of occurrence and environmental conditions. Over the past few years, we have explored how stochasticity in the occurrence of species and links, and environmental imprints on these, shape ecological interaction networks. In this perspective, we use a set of our empirically characterised networks to show that both species and link occurrence are stochastic, with probabilities conditional on the environment. This insight improves our ability to model observed variation in network structures. We argue that future research should incorporate environmental impacts on both species and links, thus merging the Grinnellian and the Eltonian niche concept.