Seasonal and interannual variability of the Columbia River plume: A perspective enabled by multiyear simulation databases

Abstract

As integral capability within an end-to-end observatory for the Columbia River estuary-plume-shelf system, we routinely create simulation databases of 3-D baroclinic circulation with unstructured grid models SELFE (Semi-implicit Eulerian-Lagrangian Finite Element) and ELCIRC (Eulerian-Lagrangian Circulation). Here, 1999–2006 SELFE simulations are used to study plume variability at multiple temporal scales: interannual, seasonal, and event scale. Time series of plume metrics, together with climatology and anomalies of surface salinity, suggest that simulations usefully capture key features of plume dynamics. In particular, simulations capture seasonal variability around two known trends: a coastally attached northward winter plume and a detached southward summer plume. Results show significant interannual variability of the plume orientation and extent, with potential implications on the variability of productivity in the system. An empirical orthogonal function analysis confirms that a bidirectional plume is prevalent in summer, showing that the result holds true regardless of interannual variability. Short-term bidirectional plumes, previously observed or modeled only in summer, can also occasionally develop in winter as a result of episodically strong upwelling-favorable winds. Across years, the predominantly coastal attached northward plume in late fall and winter is found to separate frequently from the coast, during wind relaxation events or weak wind reversals. Multiple skill scores are used to evaluate the quality of the simulations against earlier circulation databases and data. Analysis of root-mean-square error and bias suggests overall superiority of SELFE-generated over ELCIRC-generated simulation databases, but the generality of the conclusions is limited by (1) models not being the only difference between simulation databases and (2) no model prevailing across all error metrics.