Quantifying uncertainty in coastal salinity regime for biological application using quantile regression

Abstract

AbstractSalinity regimes in coastal ecosystems are highly dynamic and driven by complex geomorphic and hydrological processes. Estuarine biota are generally adapted to salinity fluctuation, but are vulnerable to salinity extremes. Characterizing coastal salinity regime for ecological studies therefore requires representing extremes of salinity ranges at time scales relevant to ecology (e.g., daily, monthly, and seasonally). Here, we propose a framework for modeling coastal salinity with these overall goals: (1) quantify uncertainty in salinity associated with important terrestrial and oceanographic drivers, (2) examine time scales of salinity response to river streamflow events, and (3) predict salinity continuously over space at key time scales. Salinity is modeled as quantile surfaces related to river discharge, tidal dynamics, wind, and spatial location, applied to Suwannee Sound estuary, FL, USA, where salinity has been monitored spatially since 1981. Each quantile level is regressed independently, and together they comprise a distribution of salinity uncertainty across space, with upper and lower quantiles describing salinity extremes. Effects of physical drivers on salinity are compared through four base models with various combinations of tide and wind variables, each including spatial coordinates and a single streamflow metric (in cubic meters per second). Multiple time scales of streamflow are considered by taking means across various periods, from 1 to 12 days, and at various lagged intervals prior to salinity sample, totaling 144 streamflow metrics. We found that the Suwannee coastal salinity regime is dynamic at multiple time scales and varies nonlinearly across space from the river effluence outward. Salinity increases nonlinearly with decreasing river flow rates below 200 m3/s, most prominently in the lower quantiles of salinity (τ = 0.05–0.25). Wind appears to have a stronger influence on salinity than astronomic tides for this estuary. The regression approach developed here can be applied to any coastal system that has sufficient spatial and temporal monitoring coverage to capture multiple flood and drought events. It is implemented with a simpleRroutine, and is less computationally‐intensive than finite difference hydrodynamic modeling. The characterizations of salinity uncertainty developed in these analyses can be directly applied to future studies of fish and wildlife responses to changes in watershed management.