|
Abstract:
|
From climate change to policy driven shifts in pollution, coastal ecosystems are rapidly changing and in dire need for scientific monitoring. These monitoring programs face challenges from missing data, low signal-to-noise, and the difficulty of studying systems which are structurally changing with time. Here, we describe how a multistage variation of dynamic linear model (DLM) structures can be used to simultaneously characterize long-term patterns, infer missing data, and test predictive relationships. We implement this modeling approach with long-term data from Narragansett Bay (NB), Rhode Island, USA: an ecosystem which has under-gone major ecological changes including policy-driven reductions in anthropogenic nutrient inputs. With this example, in a two-stage structure, the first DLM stage provided missing data inference and aided to describe changes in seasonal and long-term trends for chemical, physical, and biological attributes including phytoplankton stocks. Conditional on the posterior inference of predictors modeled in Stage 1, in Stage 2, a dynamic regression was performed to reveal a new seasonal pattern of dependence between phytoplankton and nutrients in the bay.
|
