FLARE

FLARE forecasts reservoir hydrodynamics and water temperature using real-time sensor data and partitions uncertainty among driver data, initial conditions, model processes, and parameters to inform water quality management.

Key Features:

• Real-time iterative forecasting system: Forecasts water temperature in lakes and reservoirs using continuously streaming water quality and meteorology sensor data.
• Data assimilation algorithm: Assimilates sensor observations into a hydrodynamic model and updates predictions while calibrating model parameters.
• Ensemble-based forecasting algorithm: Produces ensemble forecasts that represent probabilistic uncertainty rather than single deterministic outcomes.
• Uncertainty quantification: Quantifies contributions of driver data, initial conditions, model processes, and parameters to each daily forecast at multiple depths.
• Performance evaluation: Employs root mean square error (RMSE) metrics to quantify forecast accuracy at specified lead times and depths.

Scientific Applications:

• Falling Creek Reservoir case study: Applied to Falling Creek Reservoir (Vinton, Virginia) over a 475-day period encompassing stratified and mixed thermal conditions and predicted onset of fall turnover 4–14 days in advance across two sequential years.
• Forecast performance: For 7-day ahead forecasts RMSE was 1.13°C at 1.0 m and 0.87°C at 8.0 m; for 16-day ahead forecasts RMSE was 1.62°C at 1.0 m and 1.20°C at 8.0 m.
• Uncertainty analysis: Partitioning identified meteorology driver data as the primary source of forecast uncertainty at most depths and thermal conditions, while model process uncertainty dominated near-sediments during summer months.

Methodology:

Computational methods explicitly include assimilation of real-time sensor observations into a hydrodynamic model using a data assimilation algorithm, ensemble-based forecasting, calibration of model parameters, uncertainty partitioning, and evaluation with RMSE metrics.