How sensitive is a simulated river plume to uncertainties in wind forcing? A case study for the Red River plume (Vietnam)

Abstract

This paper aims at characterizing the sensitivity of a simulated plume's properties to uncertainties in the wind fields which force the ocean model using an ensemble method. The case study is the Red River plume in the Gulf of Tonkin. The variability of the Red River plume in the mid and far field is described in a previous paper using a clustering analysis (Nguyen-Duy et al., 2021) and is shown to be mostly driven by monsoon winds and tides. In the present study, we also aim at assessing the robustness of the classification with respect to the wind forcing uncertainties. The variability of the wind uncertainty is estimated as 60% of the wind variability with a higher variability near the coast. Based on that estimation, two ensembles of 50 simulations each with perturbed wind forcing are run over the summer 2015 period. Then, we examine the ensemble spread (defined as the standard deviation across the members) of the wind stress and of the ocean variables. The coastal current shows similar spread for both meridional and zonal flows, with the highest spread related to the highest wind stress spread. The sensitivity is the largest at the surface for salinity and at the base of the mixed-layer for temperature. The properties of the river plume are analyzed. The spread of the plume area is maximum in August, which is the same time as when the plume is the most spread out. The clustering analysis applied to the ensemble members shows some cluster attribution shifts between different members, but the cluster that is most likely to occur is still the one from the reference simulation (with unperturbed wind). These limited changes suggest that the cluster analysis of the reference simulation in Nguyen-Duy et al. (2021) is indeed robust to the wind forcing errors. The uncertainty on the plume thickness is typically less than 2m, sometimes reaching 4m (for a total thickness of 10m). The freshwater transport mainly follows the variations of the current due to the changes of wind. Possible implications of this study for the assimilation of high-frequency radar data are discussed at the end of the paper. Firstly, the relevance of the ensemble in simulating the model errors is assessed: the comparison with the data suggests that the model suffers from systematic errors that are not represented by the ensemble (by construction). Secondly, the ensemble is used to provide examples of model correction in a hypothetical data assimilation, highlighting its potential to constrain the plume by correcting directly the surface salinity, but also correcting the surface coastal current and the wind stress.