Prototype water quality forecast for Ormstrup Sø

Automated near-term water quality forecasting system for Ormstrup Sø (Lake Ormstrup), Denmark (56.327°N, 9.645°E). The system couples the General Lake Model (GLM) with the Aquatic EcoDynamics (AED) library inside the FLARE (Forecasting Lake And Reservoir Ecosystems) framework to produce daily, 14-day ensemble forecasts of water temperature and water quality variables.

Forecasts are published to an S3-compatible object store and visualized through a Quarto dashboard deployed via GitHub Actions.

The dashboard is located at flare-forecast.org/ORMS-forecast

FLARE References

The forecasting system is built on the FLARE framework. Key papers describing the methodology:

• Thomas, R.Q., et al. (2020). A near-term iterative forecasting system successfully predicts reservoir hydrodynamics and partitions uncertainty in real time. Water Resources Research, 56, e2019WR026138. https://doi.org/10.1029/2019WR026138
• Carey, C.C., et al. (2022). Advancing lake and reservoir water quality management with near-term, iterative ecological forecasting. Inland Waters, 12(1), 107–120. https://doi.org/10.1080/20442041.2020.1816421
• Thomas, R.Q., et al. (2023). Near-term forecasts of NEON lakes reveal gradients of environmental predictability across the U.S. Frontiers in Ecology and the Environment, 21(5), 220–226. https://doi.org/10.1002/fee.2623
• Wander, H.L., et al. (2024). Data assimilation experiments inform monitoring needs for near-term ecological forecasts in a eutrophic reservoir. Ecosphere, 15, e4752. https://doi.org/10.1002/ecs2.4752

Repository Structure

ORMS-forecast-code/
├── configuration/
│   └── glm_flare_v4/               # Model and FLARE configuration files
│       ├── configure_run.yml        # Run-level settings: dates, horizon, sim name
│       ├── configure_flare_glm.yml  # FLARE settings: DA method, met/S3 sources, uncertainty
│       ├── configure_flare_glm_ecmwf_ifs025.yml  # Alternate config using ECMWF IFS 0.25° driver
│       ├── configure_run_ecmwf_ifs025.yml
│       ├── glm3.nml                 # GLM hydrodynamic model namelist
│       ├── aed2.nml                 # AED water quality model namelist
│       ├── observations_config_aed.csv       # Observation variable mapping
│       ├── states_config_aed.csv             # State variable configuration
│       └── parameter_calibration_config_aed.csv  # Parameters estimated by the EnKF
│
├── workflows/
│   └── glm_flare_v4/
│       ├── forecast_workflow.R      # Main forecast workflow (GEFS/OpenMeteo driver)
│       ├── forecast_workflow_ecmwf_ifs025.R  # Workflow using ECMWF IFS driver
│       └── add_metrics.R            # Post-processing: computes additional mixing metrics
│
├── dashboard/
│   ├── index.qmd                   # Quarto dashboard source (plots, maps, forecast panels)
│   ├── _quarto.yaml                # Quarto project configuration
│   ├── _brand.yaml                 # Brand colors and typography (VT/CEF theme)
│   ├── style.css                   # Additional CSS overrides
│   ├── style.scss                  # SCSS theme extensions
│   ├── sites.json                  # GeoJSON with monitoring site location
│   ├── install.R                   # R package installation script for dashboard deps
│   ├── vt_cef.jpg                  # Dashboard logo image
│
└── .github/
    └── workflows/
        ├── run_flare_v4.yml            # GitHub Actions: manually run daily GEFS forecast
        ├── forecasts_dashboard.yml     # GitHub Actions: run forecasts. Rebuild and deploy dashboard
        └── dashboard.yml               # GitHub Actions: manually rebuild and deploy dashboard

How It Works

1. Data assimilation — The system uses an Ensemble Kalman Filter (EnKF, 100 members) to assimilate in-situ water temperature observations from Ormstrup Sø, updating both model states and selected parameters.

2. Meteorological drivers — Forecast periods use NCEP Global Ensemble Forecasting System (via OpenMeteo) for uncertainty propagation. An alternate configuration uses the ECMWF IFS 0.25° driver.

3. Forecast horizon — 14-day probabilistic forecasts generated daily. Modeled depths span 0–6 m in 0.5 m increments.

4. Output — Forecasts are written as Parquet files to an OSN (Open Storage Network) S3 bucket and visualized on the Ormstrup Sø Water Quality Forecast Dashboard.

5. Automation — GitHub Actions workflows trigger the forecast run and dashboard rebuild on a daily schedule (or manually via workflow_dispatch).

Running Locally

1. Install R dependencies including FLAREr, GLMAEDr, rLakeAnalyzer, and ropenmeteo.
2. Set AWS credentials for the OSN S3 endpoint (amnh1.osn.mghpcc.org).
3. Source the appropriate workflow script:

source("workflows/glm_flare_v4/forecast_workflow.R")

To run the ECMWF IFS–driven configuration:

source("workflows/glm_flare_v4/forecast_workflow_ecmwf_ifs025.R")