PISCES Modeling

A unique aspect of PISCES is that it purposefully uses robust data architecture to capture the presence of a species, or even the probability of a species being present. By relying on either expert opinion, which can have subjective bias, or solely empirical observation, which suffers from both sample bias and incompleteness, the outcomes of our biogeographic mapping effort would be limited. Therefore, we intentionally included a ‘modeled’ presence, with corresponding probabilities if appropriate. The reason for this approach is two-fold: one, conservation efforts require that we reduce false negatives in our mapping of sensitive species, and two, it provides a check or validation on other data.

Effective species and habitat conservation requires that managers have access to spatially explicit information in order to prioritize conservation action in areas with the greatest potential for successful outcomes (e.g., recolonization, increases in biodiversity, long-term population stability). At present, there are few resources available to watershed managers that provide a taxonomic checklist or map of fish species – particularly sensitive species that require special management consideration – that should be included in management plans. Thus, in addition to providing expert opinion or actual observation from the field, our biogeographic modeling outcomes reflect environmental and anthropogenic variables that “predict” where a given species may occur. Predictive modeling can incorporate both current and future conditions to assign probability surfaces and thereby focus management decisions in areas where data are unavailable or uncertain. PISCES outputs combined with discriminant analysis (a classification technique used to maximize differences between groups and assign categories based on a given set of multivariates) can produce predictive fine-scale distribution maps. The following discriminant analyses illustrate the potential for predictive mapping and provide a framework for future conservation efforts.

To showcase the possibilities of using predictive mapping, 17 environmental and 5 anthropogenic variables were used to model 4 fish species for the Central Valley and west slope of the Sierra Nevada (a mapping domain of 1,504 HUC12s). The species selected for this pilot covered a range of environmental tolerances and life history strategies. Environmental variables were modeled in conjunction with the historical expert opinion dataset from PISCES, and environmental plus anthropogenic variables were modeled with the current expert opinion dataset. Models were then validated with observed data to see how accurately they predicted occupancy in HUCs with observed data.