Using species distribution models to predict stream fish restoration potential

In human-dominated landscapes, sensitive species may occupy considerably fewer places than they would have under natural conditions. Because agricultural activities often accelerate sediment delivery to streams, fish species that are sensitive to sediment have suffered declines in agricultural areas. Soil conservation practices can reduce sediment inputs to streams, and therefore improve conditions for these species. However, this improvement in conditions will only be meaningful where fixed habitat characteristics, such as stream size and gradient, are appropriate for these species. Furthermore, even when it makes habitat suitable, restoration will not induce colonization by a species if potential source populations are located outside the range of its dispersal capability. Using a large fish survey database and GIS-derived habitat descriptors, we built logistic regression models that predict the presence of 14 sediment-sensitive fish species in Wisconsin streams. Each model included an autocovariate term that accounts for the influence of habitat conditions in neighboring stream segments. Models also included a measure of human influence (sum of agricultural and urban land in watershed), and were structured to allow simulation of the removal of this influence. We used these models to predict where the implementation of agricultural conservation practices would most likely result in re-colonization by each species. Predicted potential (human influence removed) species distributions were 21-74% larger than current distributions. However, when dispersal ability was accounted for, populations of individual species are predicted to be restorable in only 1-12% of Wisconsin streams. Our results indicate that habitat restoration for stream fishes will be most effective when it is conducted near existing suitable habitat, because of both dispersal limitation and spatial autocorrelation of habitat characteristics.