Modeling algal toxin concentrations in a changing world: the importance of cross-scale and space-time interactions

English Abstract:

Several factors are believed to determine the concentration of cyanobacteria in lakes. In particular, eutrophication and climate warming are suggested to be driving a global expansion of cyanobacteria in freshwater ecosystems. This is cause for concern not only because of the substantial economic loss it entails, but also because cyanobacteria are known to produce harmful neuro- and hepatotoxins. Importantly, although strong local-scale cyanotoxin models have been described, modeling the abundance of microcystin (a group of cyanotoxins), across broader spatial scales has been difficult and production remains highly dynamic through space and time. For instance, at the local scale (southern Québec lakes), 46% of the microcystin variance was explained by explanatory variables, whereas a meta-analysis developed across Canadian lakes explained far less variance (10% explained). Such scale dependencies may be due, in part, to 1) interactions between processes across different spatial scales (defined as cross-scale interactions), where overarching regional drivers cause local-scale relationships to differ, or 2) different response models over time (significant space-time interactions), which may result in poor relationships if syntheses are based on data collected at different time points. The goal of this project was thus to develop robust microcystin response models that account for these cross-scale and space-time interactions. To address our first objective, we analyzed a dataset of >1000 randomly selected lakes, ponds, and reservoirs sampled by the U.S. Environmental Protection Agency (USEPA) and applied generalized zero-inflated mixed-effect models, which effectively modeled the large number of sites with microcystin concentrations below the detection limit, allowed for local relationships to vary by region, and tested the cross-scale interactions among environmental drivers. To address our second objective, we will examine whether space-time interactions have developped over time using the NLA samplings from two separate years (2007 and 2012) and apply a two-way Analysis of Variance (ANOVA) for crossed designs with limited replication.