Salmonid or nonsalmonid lakes: predicting the fate of northern boreal fish communities with hierarchical filters relating to a keystone piscivore Johan Spens and John P. Ball Abstract: We determine if lacustrine salmonids show large-scale patterns of coexistence with the keystone predator north- ern pike (Esox lucius) and test an approach to predict fish communities using coexistence rules set in the context of three hierarchical filters that a species must pass to be present. The mutually exclusive species distribution patterns that we de- tected among 1029 lakes were repeatedly verified from results of whole-lake interventions with rotenone and introductions. Essentially, pike did not coexist with self-sustaining salmonid populations in lakes. High connectivity to pike (derived from maps) largely predicts the absence of lacustrine salmonids. Our analysis strongly suggests that pike prevented self- sustaining populations of brown trout (Salmo trutta), Arctic char (Salvelinus alpinus), and brook trout (Salvelinus fontina- lis) in lakes. High connectivity to pike resulted in nonsalmonid lake fish communities, most often including both European perch (Perca fluviatilis) and roach (Rutilus rutilus). Our analysis suggests that if pike were not present in many boreal lakes where they now dwell, salmonid fish assemblages would prevail, a sharp contrast from the present pike-driven ho- mogenized state with mainly nonsalmonid fish communities. Re ´sume ´: Nous de ´terminons si les salmonide ´s lacustres posse `dent des patrons de coexistence a ` grande e ´chelle avec le bro- chet du nord (Esox lucius), leur pre ´dateur le plus important, et nous testons une me ´thode pour pre ´dire les communaute ´s de poissons a ` l’aide de re `gles de coexistence qui comprennent trois filtres hie ´rarchiques qu’une espe `ce doit traverser pour e ˆtre pre ´sente. Les patrons de re ´partition des espe `ces mutuellement exclusives que nous avons de ´tecte ´s dans les 1029 lacs se ve ´rifient a ` re ´pe ´tition lors d’interventions a ` la rote ´none ou d’introductions a ` l’e ´chelle des lacs entiers. En somme, les bro- chets ne se retrouvent pas avec des populations autosuffisantes de salmonide ´s dans les lacs. Une forte connectivite ´ avec le brochet (de ´duite a ` partir de cartes ge ´ographiques) permet la plupart du temps de pre ´dire l’absence de salmonide ´s lacustres. Notre analyse indique clairement que le brochet empe ˆche l’e ´tablissement de populations autosuffisantes de truites brunes (Salmo trutta), d’ombles chevaliers (Salvelinus alpinus) et d’ombles de fontaine (Salvelinus fontinalis) dans les lacs. Une connectivite ´ au brochet produit des communaute ´s de poissons lacustres sans salmonide ´s, mais contenant souvent a ` la fois la perche (Perca fluviatilis) et le gardon (Rutilus rutilus). Notre analyse laisse croire que si le brochet e ´tait absent de plu- sieurs lacs bore ´aux ou ` il habite pre ´sentement, il y aurait pre ´dominance de peuplements de poissons salmonide ´s — en con- traste marque ´ avec l’e ´tat d’homoge ´ne ´ite ´ actuel des communaute ´s de poissons non salmonide ´s, cause ´ par le brochet. [Traduit par la Re ´daction] Introduction Understanding the main factors that determine the distri- bution of species is critical for the effective conservation and management of many ecosystems. For freshwaters, a promising tool for predicting species’ distributions is the use of series of hierarchical filters (Poff 1997; Quist et al. 2005) that a species must pass to be present in a given loca- tion (Tonn et al. 1990). A promising avenue to construct this filtering framework may be to focus on taxa that are strong interactors (i.e., ‘‘keystone species’’; Poff 1997) like some predators or competitors. We propose that barriers to the dispersal of keystone species, i.e., (i) keystone-focused con- nectivity, used in sequence with (ii) abiotic habitat charac- teristics followed by (iii) coexistence rules with keystone species, can advantageously be used as three hierarchical fil- ters for predicting fish communities in lakes. A previous single-species study predicted the distribution of the keystone species northern pike (Esox lucius) (here- after pike) with very high accuracy (86.6% correct classifi- cation among 1365 lakes), mainly based on a map-derived connectivity variable (Spens et al. 2007). Specifically, in this earlier paper, a geographic information system (GIS) de- rived variable (S V5max ) that quantified dispersal barriers to pike provided a measure of steepness in relation to 5 m (ver- tical) contour intervals rather than the traditional measure of rise over a fixed, and typically long, length of stream. For example, the conventional measure would quantify a rise of so many metres per 100 m, or 200 m, of river or whole stream and is less suited to differentiate a gradual slope, which would not impede dispersal, from a waterfall, which would. In contrast, the S V5max variable will indicate a disper- sal barrier whenever there was a rise of 5 m (i.e., in £76 m or ‡6.6% for pike; this figure is empirically derived in Spens et al. 2007). Thus, this previous study shows that con- Received 30 September 2007. Accepted 18 April 2008. Published on the NRC Research Press Web site at cjfas.nrc.ca on 26 August 2008. J20202 J. Spens 1 and J.P. Ball. Department of Wildlife, Fish, and Environmental Studies, Swedish University of Agricultural Sciences, SE-901 83 Umea ˚, Sweden. 1 Corresponding author (e-mail: johan.spens@vfm.slu.se). 1945 Can. J. Fish. Aquat. Sci. 65: 1945–1955 (2008) doi:10.1139/F08-103 # 2008 NRC Canada