Projecting stream water quality using Weighted Regression on Time, Discharge, and Season (WRTDS): An example with drought conditions in the Delaware River Basin

Abstract

Future water availability depends on understanding the responses of constituent concentrations to hydrologic change. Projecting future water quality remains a methodological challenge, particularly when using discrete observations with limited temporal resolution. This study introduces Weighted Regression on Time, Discharge, and Season for Projection (WRTDS-P), a novel, computationally efficient method that enables the projection of daily stream water quality under varying hydrologic conditions using commonly available discrete monitoring data. WRTDS-P model performance was validated using 39 sites in the Delaware River Basin (DRB) and four key constituents: specific conductance (SC), nitrate (NO 3 − ), magnesium (Mg 2+ ) and calcium (Ca 2+ ). Projections were tested against holdout data from the final 1 to 5 years of each time series, demonstrating robust predictive capability, with median Nash-Sutcliffe efficiencies of 0.67 for SC, 0.56 for NO 3 − , 0.65 for Ca 2+ , and 0.79 for Mg 2+ . Model uncertainty was correlated with indicators of hydrologic or geochemical mass-sinks, such as groundwater storage and adsorption in wetland soils. Drought scenario analyses for SC used ranges of reduced discharge including flows from the 1965 drought of record. Scenarios predicted widespread increases of SC, especially in southern DRB streams where baseline SC levels are already elevated. Fractional increases of SC were more uniformly distributed, indicating potential risk to sensitive ecosystems. Notably, drought-induced SC increases were positively correlated with interannual SC trends, indicating that hydrologic extremes could exacerbate ongoing salinization. This work provides a transferable and interpretable framework for projecting future water quality and assessing hydrologic risk to water resources and aquatic ecosystems.