Using ecological niche - based modeling techniques to predict the establishment , abundance 1 and relative impacts of introduced species : application to the common carp ( Cyprinus carpio ) 2 3

25 In order to efficiently manage non-indigenous species (NIS) predictive tools are needed to 26 prioritize locations where they are likely to become established and where their impacts will be 27 most severe. While predicting the impact of a NIS has generally proved challenging, forecasting 28 its abundance patterns across potential recipient locations should serve as a useful surrogate 29 method of estimating the relative severity of the impacts to be expected. Yet such approaches have 30 rarely been applied in invasion biology. We used long-term monitoring data for lakes within the 31 state of Minnesota and artificial neural networks to model both the occurrence as well as the 32 abundance of a widespread aquatic NIS, common carp (Cyprinus carpio). We then tested the 33 ability of the resulting models to (i) interpolate to new sites within our main study region, (ii) 34 extrapolate to lakes in the neighboring state of South Dakota and (iii) assessed the relative 35 contribution of each variable to model predictions. Our models correctly identified over 83% of 36 sites where carp are either present or absent and explained 73% of the variation in carp abundance 37 for validation lakes in Minnesota (i.e. lakes not used to build the model). When extrapolated to 38 South Dakota, our models correctly classified carp occurrence in 79% of lakes and explained 32% 39 of the variation in carp abundance. Variables related to climate and water quality were found to be 40 the most important predictors of carp distribution. These results demonstrate that ecological niche-41 based modeling techniques can be used to forecast both the occurrence and abundance patterns of 42 invasive species at a regional scale. Models also yielded sensible predictions when extrapolated to 43 neighboring regions. Such predictions, when combined, should provide more useful estimates of 44 the overall risk posed by NIS on potential recipient systems. 45 46