A framework for Improving Hydrologic and Water Quality Prediction in Urbanized Watersheds through Stakeholder Co-Design and Multi-Model Integration

Abstract

Urban watersheds present a unique modeling challenge due to the complex interplay between natural and hydrologic processes, engineered infrastructure, and the diverse decision-making by multiple stakeholder groups. These interactions span multiple spatial and temporal scales, making it difficult for any single modelling approach to represent the system's full complexity or adequately address diverse stakeholder needs. Many existing modeling frameworks fail to align with stakeholder decision processes, reducing their relevance for applied watershed management. As the first objective, this dissertation introduces a stakeholder driven collaborative design process for developing the Occoquan Watershed Modeling Framework (OWMF), a multi-model, co-designed watershed modeling framework for simulating water quantity and quality applied within the Occoquan Watershed in Northern Virginia, USA. The framework represents a novel advancements in watershed modeling by addressing persistent design limitations in existing approaches by: (1) supporting multi-functional design objectives across hydrologic and water-quality domains, (2) embedding stakeholder priorities from the outset through an iterative, user-centered co-design process, (3) integrating scientifically rigorous, high-fidelity models, and (4) applying competency-based evaluation criteria to quantify performance, feasibility, and decision relevance. These design principles were operationalized through a structured, iterative co-design process that translated stakeholder priorities and model competencies into an implementable framework incorporating models such as GR4J-CemaNeige, SWAT, WAMRF and StormWise. The model selection was further validated by site-specific prototyping and performance evaluation. Following this, the finalized framework development plan was obtained through the co-design process. The analysis presented here provides a structured methodology to build stakeholder-driven, multi-model frameworks that can predict the short- and long-term impacts of natural and anthropogenic drivers that influence watershed resilience. The conclusions aim to bridge the gap between hydrologic modeling and watershed management, enabling a transparent, adaptive and transferable approach for enhancing watershed resilience. Understanding how future land use – land cover (LULC) and climate change (CC) can alter watershed hydrology and water quality is critical for effective long-term watershed management and planning. With the second objective, this dissertation incorporated a multi-model approach for improving the watershed-scale impact assessment under rapid urbanization and climate change. First, high-resolution LULC and CC projections were developed for the year 2040. Second, the watershed-scale dynamics under baseline (present) and future (2040) scenarios were simulated using three models: SWAT, HSPF, and WARMF. Third, an inter-model comparison was conducted that related the differences in model architecture, spatial discretization, process characterization and calibration strategy to watershed responses under future scenarios. The differences in simulated streamflow, pollutant loads (e.g., nitrogen, phosphorus), and sediment loads were quantified across the three models and three future scenarios. Despite using the same forcing, the three models produced different magnitudes of change in streamflow, sediments and nutrient loading, reflecting the impact of model structure in affecting processes such as simulated runoff generation, sediment detachment, subsurface flow partitioning, phosphorous transport and nitrogen cycling. Moreover, the simulation timestep (hourly vs daily), calibration timestep (hourly vs daily vs monthly) and input data resolution directly impacted the sensitivity of these models to LULC change and climate variability. The inter-model comparison concluded that in addition to the model structure, the calibration methodology impacted how the models projected into the future. Whether the calibration was biased or unbiased towards extremes and which objective functions (streamflow, ET, nutrients etc.) were chosen for calibrating the baseline models had a profound impact on predicting the future watershed responses across the three models. The study showed that multi-model assessments should be the standard methodology for improving confidence in future watershed-scale hydrologic and water quality predictions under the influence of future variability in LULC and climate. Seasonally shifting hydro-meteorological conditions can introduce substantial variability in watershed response, yet majority of rainfall–runoff models often rely on fixed parameter sets that do not adjust to these changes. The third objective incorporated a multi-pronged approach for improving seasonality representation by improving model parameterization and coupling it with data-driven modeling of hydrologic systems. This study leveraged both these approaches for representing seasonality in hydrologic models for improved streamflow prediction. Using the GR4J-CemaNeige model for the Occoquan Watershed in Northern Virginia, this study tested the application of a predefined four-season parameterization, followed by univariate and multivariate clustering to identify data-driven hydro-climatic patterns. Insights from these analyses informed the development of a hybrid dynamic parameterization in which model parameters varied continuously with time and varied with respect to local-scale potential evapotranspiration observations. Results showed that traditional four-season parameterization improved hydrologic performance only when seasonal boundaries coincided with actual hydrometeorological behavior. The univariate clustering analysis showed that temperature and evapotranspiration followed repeatable annual cycles, whereas precipitation and streamflow displayed irregular and highly variable seasonal behavior, including transitional months without consistent cluster identity. The multivariate clustering further demonstrated that the combined hydro-climatic variables did not align reliably with fixed patterns, reflecting the irregular timing of hydrologic conditions in the study basins. The dynamic formulation generated continuously evolving parameter trajectories and produced more consistent performance across evaluation periods. Collectively, the stepwise progression, from seasonal calibration to clustering-based diagnostics and dynamic parameterization provided a systematic framework for diagnosing seasonal hydrologic behavior and enhancing the temporal adaptability of conceptual hydrologic models.