Aim To incorporate dispersal through stream networks into models predicting the future distribution of a native, freshwater fish given climate change scenarios.
Methods We used logistic regression to fit climate and habitat data to observed pike (Esox lucius Linnaeus) distributions in 13,476 lakes. We used GIS to map dispersal pathways through streams. Lakes either (1) contained pike or were downstream from pike lakes, (2) were upstream from pike lakes, but downstream from natural dispersal barriers, or (3) were isolated from streams or were upstream from natural dispersal barriers. We then used climate projections to model future distributions of pike and compared our results with and without including dispersal.
Results Given climate and habitat, pike were predicted present in all of 99,249 Swedish lakes by 2100. After accounting for dispersal barriers, we only predicted pike presence in 31,538 lakes. Dispersal barriers most strongly limited pike invasion in mountainous regions, but low connectivity also characterized some relatively flat regions.
Main conclusions The dendritic network structure of streams and interconnected lakes makes a two-dimensional representation of the landscape unsuitable for predicting range shifts of many freshwater organisms. If dispersal through stream networks is not accounted for, predictions of future fish distributions in a warmer climate might grossly overestimate range expansions of warm and cool-water fishes and underestimate range contractions of cold-water fishes. Dispersal through stream networks can be modelled in any region for which a digital elevation model and species occurrence data are available.