SPATIOTEMPORAL SALINITY DYNAMICS IN THE PATUXENT RIVER: GOVERNING FACTORS AND RECONSTRUCTION USING BOUNDARY CONDITIONS

Abstract

Estuarine salinity integrates the effects of freshwater forcing, marine boundary conditions, mixing, and channel geometry, but these controls are often expressed unevenly in space and time. The Patuxent River, a tributary of the Chesapeake Bay, provides a setting in which these controls can be traced along a single channel axis. This thesis examines the processes driving salinity variability in this system and tests whether a machine-learning extension can improve a reduced-complexity physical framework. Chapter 2 shows that Patuxent salinity is organized by station-specific seasonal climatology rather than by a single estuary-wide signal. A reduced-complexity physical model quantifies the contributions of the two boundary conditions: freshwater discharge at the head of tide and salinity at the mouth of the Patuxent River. Counterfactual and cross-sectional diagnostics indicate that salinity in the uppermost reach and at the river mouth is each largely governed by a single driver, whereas the interaction of the two drivers controls spatiotemporal salinity variation in the mid-estuary. This intermediate region, here termed the transition zone, is where freshwater influence is filtered by channel geometry and boundary effects. Chapter 3 reconstructs salinity variability using both the reduced-complexity physical model and a multilayer perceptron (MLP). The physical model performs well at stations where the annual salinity cycle is stable from year to year. The MLP, in contrast, improves aggregate prediction and adds the most value in the upper transition reach, particularly under high- and low-flow conditions and in representative stress-test years. Overall, the thesis shows that a reduced-complexity model can quantify the contributions of the two dominant drivers of spatiotemporal salinity change along the Patuxent River, and that an MLP offers an effective means of combining those drivers to reconstruct salinity variability. More broadly, this pairing offers a transferable strategy for estuaries where monitoring records are sparse or discontinuous and could be extended to other coastal systems and to variables beyond salinity.