Towards the development of integrated modelling systems in aquatic biogeochemistry: a Bayesian approach

Abstract

Modelling constructs are designed to shed light on different facets of biogeochemical cycles, but their application involves substantial uncertainty contributed by model structure, parameters, and other inputs. The Bayesian paradigm is uniquely suitable for developing integrated environmental modelling systems, overcoming the conceptual or scale misalignment between processes of interest and supporting information, and exploiting disparate sources of information that differ with regards to the measurement error and resolution. A network of models is developed to connect the watershed processes with the dynamics of the receiving waterbody in the Hamilton Harbour (Ontario, Canada). The SPAtially Referenced Regressions On Watershed attributes (SPARROW) along with an intermediate complexity eutrophication model were used to reproduce the phosphorus cycling in the system, including the exchange between sediment and water column as well as the interplay between the ambient and phytoplankton intracellular pools. The novel features of the framework include (i) the development of a downscaling algorithm that transforms the SPARROW annual phosphorus loading estimates to daily inputs for the eutrophication model; and (ii) a neural network that emulates the posterior linkages between model parameters/phosphorus loading inputs and the predicted total phosphorus, chlorophyll a concentrations, and zooplankton abundance. Our integrated watershed-receiving waterbody model is independently tested against a 22-year period (1988–2009) and is subsequently used to gain insights into the ecological factors that shape the current water quality conditions in the system and may modulate its future response to the nutrient loading reductions proposed by the Hamilton Harbour Remedial Action Plan.