<em>Abstract</em>.—Systematic conservation planning tools offer powerful and flexible means for addressing the protection of biodiversity in freshwater systems. Tools such as the software Zonation can be used to prioritize streams for protection, restoration, and management of aquatic resources. The flexible nature of these tools allow analyses to be tailored to specific objectives but also introduces uncertainty regarding the effects of selected input options on the rankings of stream segments and the representation of fish species within prioritized streams. The objective of our research was to evaluate the effectiveness of several species distribution modeling techniques (generalized additive models, multivariate adaptive regression splines, boosted regression trees, and random forest models, including an ensemble based on these techniques) for characterizing distributions of fish communities and to identify the influence of different prioritization options of Zonation conservation planning software within five input classes (species occurrence data, removal rule, species weighting, connectivity, and protected area masking) on both the resulting stream segment rankings and the representation of species within priority streams. All combinations of input options were compared based on the correlation and congruence of stream rankings and the mean richness of species, minimum level of species representation, and representation of rare species within streams across priority levels. Of the distribution modeling types we evaluated, boosted regression trees performed the best, followed closely by random forest models. The use of an ensemble approach allowed for the largest number of species with robust predicted distributions. Our results also suggested that protected area masking had the largest effects on conservation priority results, followed by choice of removal rule, while species occurrence data type had limited impacts. The information contained in this chapter is meant to aid planners in understanding how their selection of conservation planning inputs is likely to impact results.
Comparing pseudo-absences generation techniques in Boosted Regression Trees models for conservation purposes: A case study on amphibians in a protected area
