It is now recognized that predictive models of HABs are necessary to supplement limited time-series data and identify characteristic physical-biological interactions that may influence HAB formation over regional scales [McGillicuddy, 2010]. Future efforts will combine mechanistic and statistical approaches for making robust, reliable forecasts in response to a diverse range of environmental processes. Toxic blooms that are not observed by standard monitoring methods nor noted by a rise in animal strandings still threaten wildlife populations, for example, via vertical export of toxins to benthic biota [Sekula-Wood et al., 2009]. Spatially-explicit models of HAB probabilities provide a complementary tool for assessing public health risk, predicting ecosystem disturbance, and warning resource managers of offshore blooms.
 The conditions that lead to Pseudo-nitzschia blooms and the subsequent production of DA may differ as evidenced in laboratory and field experiments linking toxin production to varying forms of macro and micro- nutrient limitation [Bates et al., 1998; Fehling et al., 2004; Kudela et al., 2004, 2010; Maldonado et al., 2002; Pan et al., 1996; Wells et al., 2005]. As a result, the accuracy of current statistical HAB models that are not based on unique optical signatures [Cannizzaro et al., 2008; Tomlinson et al., 2009] is generally constrained by estimation of the key environmental fields necessary to predict blooms and toxin production, both fundamentally a response to nutrient type and availability [McGillicuddy et al., 2003, 2005]. Application of the better-performing full HAB models that combine relevant physical and chemical fields ultimately requires a validated ROMS and biogeochemical model nested within ROMS to produce daily, high resolution simulations of chlorophyll, silicic acid, nitrate, and phosphate [Moore et al., 2001] as well as the increasingly-recognized organic nitrogen sources that trigger toxin production in Pseudo-nitzschia [Cochlan et al., 2008; Howard et al., 2007; Thessen et al., 2009]. The remote-sensing models in this study miss important dynamics linked to the nutrient environment, and this likely contributes to an inflated rate of false positive predictions. Separating the Pseudo-nitzschia model, for instance, into spring/summer and fall/winter models (data not shown) suggests it may be possible to distinguish Pseudo-nitzschia blooms in fall/winter using a chlorophyll anomaly method from MODIS data [e.g., Tomlinson et al., 2009], while in spring and summer, using simulations of nutrient levels to distinguish Pseudo-nitzschia from other blooms in the SBC. The additional information provided by nutrient ratios and concentrations from biogeochemical model-to-ROMS coupling, particularly at subsurface depths within the euphotic zone, will greatly aid in forecasting surface to subsurface blooms. Ideally, the more inclusive approach using ecosystem models that parameterize nutrients and differentiate carbon species will not only facilitate better predictions for resource managers but allow for experimental exploration of eutrophication and ocean acidification effects on HAB variability in the California Current System.