A regional model to predict the occurrence of natural events: Application to phytoplankton blooms in continental waterbodies

There are significant social, economic and environmental costs related to phytoplankton blooms, particularly when they are dominated by cyanobacteria. However, the climatic, physiographic and morphological drivers involved in bloom phenology create complex spatiotemporal patterns in biomass and taxonomic composition, making monitoring and forecasting challenging. This complexity also limits the ability of process-based algorithms to model bloom phenology. Stochastic approaches and predictive modelling are valuable risk management tools for natural events. This paper presents a novel model to estimate the conditional density of phytoplankton blooms based on environmental covariates, namely the physiography and climate descriptors of lake watersheds. The goal was to provide a tool to project future scenarios of bloom phenology in response to climate change and anthropogenic developments, and to test the efficiency of mitigating approaches. Specifically, we developed a non-stationary regional model to estimate the probability of bloom occurrence in 580 lakes in Quebec, Canada, and to study the impact of physiographic or climate changes over our reference period and future horizons. Results show that the regional model is more precise and accurate than local models, which are usually used in frequency analyses of extreme values. Simulations conducted for different climatic and physiographic scenarios on the watershed of Lake Brome indicate a 65% increase in bloom frequency on the horizon of years 2080–2100. The model also evaluates the effects of different environmental disturbances on bloom phenology, such as agriculture or settlement increase in the watershed. Disturbance case studies were carried out on lakes with different trophic levels, physiographic and climatic characteristics. Simulations indicate that bloom frequency in a highly urbanized meso-eutrophic lake (Lake Brome) and a eutrophic lake with a watershed dominated by agriculture (Missisquoi Bay) would be more responsive to changes in climate rather than land cover, although physiographic disturbances would still generate considerable effects. A meso-eutrophic lake with a watershed dominated by forest (Lake Aylmer) would neither respond to increases in agriculture nor settlement but would be most affected by increases in degree days. Another scenario focused on a highly forested oligo-mesotrophic lake (Lake St. Jean). Bloom phenology in this lake was most affected by disturbances in agriculture or settlement land use. Overall, these results suggest that lakes with a high level of eutrophication are the least responsive to the pressures resulting from land cover changes.