Clustering coupled biochemical-physical model results formalizes regional provinces in a coastal region

Coastal regions by their very nature are dynamically diverse.  Within one geographical region there are often multiple areas dominated by substantially different dynamics that shape not only the physical characteristics but also the ecosystem.  The Salish Sea, in the northeast Pacific, is an excellent example with strongly tidally mixed regions, freshwater-dominated regions, and regions directly influenced by the open ocean.  These regions are generally well known and multiple disciplines refer to them with various boundaries and under various names.  Here we use unsupervised clustering on numerical model results to formalize these regional provinces.  The model is SalishSeaCast,  a three-dimensional real-time coupled bio-chem-physical model based on the NEMO framework.  We find that the regions clustered on ecosystem variables (phytoplankton biomass) spatially coincide with those clustered on physical variables, particularly the stratification as diagnosed by the halocline depth.  The clusters are robust across years with interannual variability manifesting mostly in changes in the size of the clusters.  As the clusters are dynamically distinct, they provide a natural framework on which to evaluate the model against observations.  We find that the model accurately simulates each of the major clusters.  The spatial and temporal resolution of the model can then characterize these different clusters more systematically than the observations, revealing biases associated with sparse sampling in the observations. Two examples will be given, one addressing a long-standing issue of the productivity gradient in the stratified main basin, the Strait of Georgia, and another concerning the seasonal cycle of productivity in the ocean-influenced Juan de Fuca Strait.