Yi Hong

Hydrologic models have been increasingly used as a numerical tool to support urban stormwater management. Evaluation of modeling approaches helps identify the strength and weakness of a model to meet end-user requirements. However, traditional model evaluations only focus on the technical performance of a model, whereas very few studies have been conducted to quantitatively evaluate practical constraints for model applications. Therefore, this study proposed a quantitative model evaluation framework, to analyze tradeoffs between scientific reliability and practical feasibility of four process-based urban stormwater quality models. These models were based on different levels of spatial discretization, including lumped, sub-catchment, UHE and grid based approaches; test simulations were applied to an urban catchment near Paris. Six criteria were introduced to quantitatively assess the characteristics of modeling approaches, including (1) match to observation, (2) forecast accuracy, (3) forecast variability, (4) data accessibility, (5) computational costs, and (6) model reusability. The results showed that the lumped model was the best tradeoff between scientific reliability and practical feasibility for the study case. Moreover, the greater spatially distributed exponential build-up/wash-off processes from the lumped to sub-catchment based model could only improve the numerical approximation of simulations to observations at the outlet, but performed much less well in other scientific reliability aspects. Which implies that these processes may not properly represent mechanisms for stormwater quality dynamics at the catchment scale. In addition, it was suggested that complex grid based models should work together with advanced parameter calibration approaches, in order to achieve good scientific reliability for research purposes. In perspective, quantitative evaluation of the stakeholder participation throughout the modeling processes could help to improve model-based outcomes with more adaptive stakeholder engagement.

MAELIA (Multi-Agent for EnvironmentaL norms Impact Assesment) is an agent-based simulation platform designed to assess the impact of alternative water management policies at the watershed level. It simulates interactions between human activities (agricultural, domestic and industrial withdrawals, regulations of water uses) and ecological processes (crop growth, plant evapotranspiration and water flow). To simulate the water cycles at the sub-watershed level, the representations of hydrological processes in SWAT are analysed and re-implemented in MAELIA. In this paper, we discussed some of the design choices, simplifications (e.g. overland flow, soil temperature, perched water table and bypass flow are not simulated) and modifications (e.g. different methods of generating HRUs, elevation bands, snow water are calculated in the sub-basin loops etc.) we made in order to integrate SWAT hydrological formalisms: surface-runoff, soil water, snow water, groundwater, evapotranspiration and water-routing. Otherwise, MAELIA use some specific processes for simulating crops growth (Jeud’O model), weather generator (SAFRAN model) and water management (Multi-agents). The MAELIA platform was tested on a 6000 km2 watershed at the upstream part of Adour-Garonne-Basin in south-west of France for a 10-years period (2000-2009). The results show that the MAELIA platform reasonably reproduces stream flow hydrograph for a long-term simulation, the effects of water management restrictions are well represented, especially during the low-water periods. Sensibility analysis shows that some parameters in MAELIA are less influential comparing with SWAT model (base-flow and soil properties parameters). The MAELIA platform is a promising agent-based, real-time model for simulating the impacts of water management policies including human behaviour and economical aspects.

Hydrologic forecasting in large, transboundary basins such as the Laurentian Great Lakes remains a significant challenge due to complex hydrological processes, extensive surface water coverage, and widespread ungauged regions. These factors limit the predictive skill of existing models and constrain effective operational water resource management. This study evaluates and compares two modeling approaches for seasonal runoff prediction: a data-driven Long Short-Term Memory (LSTM) neural network model and the process-based Large Basin Runoff Model (LBRM). Using the Climate Forecast System Reanalysis (CFSR) as climate forcing from 1979 to 2023, both LSTM and LBRM models were developed and tested to simulate runoff across the entire Great Lakes basin. Their seasonal forecasting performance was then assessed using archived 9-month forecasts from the Climate Forecast System (CFS) as input. Additionally, total regional runoff was evaluated from a water balance perspective to assess model effectiveness for seasonal water supply forecasting. This research aims to refine hydrologic modeling frameworks using the Great Lakes as a testbed for seasonal-to-annual water supply forecasting. The results offer insights into the comparative strengths of data-driven and process-based models and highlight their potential integration for improving forecast accuracy. Ultimately, this work supports the development of more reliable forecasting tools for operational water management in the Great Lakes and other large transboundary basins worldwide.

Accurate hydrological simulation of the Laurentian Great Lakes-the world’s largest surface freshwater system-is critical for effective water resource management, yet remains challenging due to the basin’s vast size, international boundaries, and data heterogeneity. While deep learning models such as long-short-term memory (LSTM) networks have shown promise in hydrological forecasting, a unified, data-driven model for daily streamflow prediction across the entire Great Lakes basin has not been realized. This study addresses this gap by developing and evaluating an Entity-Aware LSTM (EA-LSTM) model, trained on a newly compiled dataset spanning 975 sub-basins across the U.S. and Canada from 1980 to 2023. The dataset integrates daily meteorological forcings, static catchment attributes, and observed streamflow. The EA-LSTM, using a temporal training split, achieved a median Nash-Sutcliffe Efficiency (NSE) of 0.685, significantly outperforming both a standard LSTM (median NSE 0.567) and the operational NOAA National Water Model (median NSE 0.208). Performance analysis reveals that the model’s predictive skill varies by catchment characteristics, with reduced performance observed in very small and highly urbanized basins. Our findings demonstrate that a single, data-driven model can effectively simulate streamflow across a vast, international watershed without basin-specific calibration, offering a robust and unified tool for transboundary water resource management. This work highlights the potential of advanced deep learning models to enhance hydrological prediction and support transboundary water resources management.

Hydrological simulation of large, transboundary water systems like the Laurentian Great Lakes remains challenging. Although deep learning has advanced hydrologic forecasting, prior efforts are fragmented, lacking a unified basin-wide model for daily streamflow. We address this gap by developing a single Entity-Aware Long Short-Term Memory (EA-LSTM) model, an architecture that distinctly processes static catchment attributes and dynamic meteorological forcings, trained without basin-specific calibration. We compile a cross-border dataset integrating daily meteorological forcings, static catchment attributes, and observed streamflow for 975 sub-basins across the United States and Canada (1980–2023). With a temporal training/testing split, the unified EA-LSTM attains a median Nash–Sutcliffe Efficiency (NSE) of 0.685 and a median Kling–Gupta Efficiency (KGE) of 0.678 in validation, substantially exceeding a standard LSTM (median NSE 0.567, KGE 0.555) and the operational NOAA National Water Model (median NSE 0.209, KGE 0.440). Although skill is reduced in the smallest basins (median NSE 0.554) and during high-flow events (median PBIAS −29.6%), the performance is robust across diverse hydroclimatic settings. These results demonstrate that a single, calibration-free deep learning model can provide accurate, scalable streamflow prediction across an international basin, offering a practical path toward unified forecasting for the Great Lakes and a transferable framework for other large, data-sparse watersheds.

Floods are posing increasing threats to lives, infrastructure, and ecosystems across global, continental, and regional scales, driven by intensifying hydrological extremes, aging infrastructure, and socioeconomic disparities. Despite advances in hydrologic modeling, challenges remain in developing high-resolution flood forecasting systems that integrate physical processes and human dimensions. Here, we advance flood modeling and forecasting by developing a high-resolution, coupled hydrological-hydrodynamic modeling framework, using the Laurentian Great Lakes Basin (GLB) as a testbed. This framework couples the National Water Model (NWM) with the CaMa-Flood hydrodynamic model that accurately simulates streamflow, water levels, and flood inundation by utilizing runoff from NWM to drive CaMa-Flood model that includes reservoir operation, at 1-arcmin (~1.6 km) spatial resolution, with flood inundation downscaled to a finer resolution of 3 sec (~90m). Incorporating dynamic reservoir operation substantially improves streamflow simulation in the regulated sub-basins. The framework exhibits high streamflow forecast skill at lead times up to 48 hours and accurately captures the timing and extent of historical extreme flood events. Long-term simulations reveal increasing flood occurrence across most of the sub-basins over recent decades. Overlaying flood hazard with the Social Vulnerability Index (SVI) highlights high-risk areas in Southwestern Wisconsin, Southern Michigan, and parts of New York, where flood exposure intersects with social vulnerability. This modeling framework underscores the importance of coupling physical and social data for equitable flood risk management, and the proposed framework offers a scalable tool for broader regional and global applications.

Optimizing empirical baseflow filters using environmental tracers (e.g., specific electrical conductance (SEC), turbidity) is an effective and efficient way to quantify the contribution of baseflow to total flow. To execute this baseflow separation, three key components are needed: The tracer, the method to estimate tracer concentration in different flow components, and the empirical baseflow filter. However, a comprehensive evaluation of the various combinations of these components, especially with a large sample of catchments, is currently lacking in the literature. Therefore, our study assembles 16 hybrid baseflow filters from two tracers, two concentration estimation methods, and four empirical baseflow filters, and evaluated their performance in baseflow separation and producing two long‐term baseflow signatures for 1,100 catchments in the Contiguous United States. Our results suggest that SEC is a superior tracer to turbidity for baseflow separation. Additionally, using monthly maximum and minimum values to represent tracer concentration in flow components produces better separation than using a power function relationship between flow rate and concentration. The four empirical baseflow filters offer a similar level of performance, regardless of the other options used. Yet, some of these filters produce inconsistent results in calculating the baseflow signatures for the catchments. Our analysis shed light on the optimization of hybrid baseflow filters for the accurate quantification of baseflow contribution.