Biscayne Aquifer POMDP: ERA5-calibrated simulation data and Monte Carlo results

Description

# Biscayne Aquifer POMDP Experiment Companion code for: > Jakaite, L. & Schetinin, V. (2026). *A Unified POMDP Framework with Robust Satisficing, Directed Learning, and Real Options for Environmental Decisions under Deep Uncertainty.* Environmental Modelling & Software (submitted). --- ## Repository structure ``` biscayne/ biscayne_main.py # Main experiment — runs all 6 strategies make_graphical_abstract.py # Generates graphical abstract figure modules/ data_loader.py # USGS/NOAA groundwater data loader enso_loader.py # NOAA CPC ONI index loader regime_calibrator.py # HMM climate regime calibration (ERA5) obs_calibrator.py # Observation noise calibration salinity_proxy.py # Physics-based salinity proxy data/ raw/ # ERA5, NOAA sea-level, ONI, salinity proxy CSVs calibration_params.json # Pre-computed HMM parameters transition_matrix.csv # Calibrated regime transition matrix results/ dapp_final/ # Canonical results cited in the paper (n=1,000) mc_results.json # 6-strategy failure rates and costs fig_mc_comparison.png fig_barrier_timing.png fig_belief_trajectory.png fig_sensitivity_heatmap.png concept_colab.py # Conceptual illustration of Algorithm 1 requirements.txt ``` --- ## Reproducing the results ```bash pip install -r requirements.txt cd biscayne python biscayne_main.py --mc-trials 1000 --seed 42 ``` Expected output (`results/mc_results.json`): | Strategy | Failure rate | Cost | |---|---|---| | Unified Framework | 0.0 % | 34.2 | | RDM-Conservative (Lempert 2006) | 0.0 % | 35.7 | | DAPP (Haasnoot 2013) | 9.5 % | 118.9 | | Myopic POMDP | 41.6 % | — | | Robust Satisficing | 43.4 % | — | | BDT Baseline | 59.1 % | 393.9 | Runtime: ~3–5 minutes on a standard laptop (Python 3.10+). --- ## Data sources | Dataset | Source | Period | |---|---|---| | ERA5 precipitation | ECMWF open data | 2000–2023 | | NOAA Key West sea level | NOAA Tides & Currents | 2000–2023 | | ENSO ONI index | NOAA CPC | 2000–2023 | | Physics-based salinity proxy | Computed (see `modules/salinity_proxy.py`) | 2000–2023 | --- ## Conceptual illustration (optional) `concept_colab.py` is an interactive illustration of Algorithm 1 on a synthetic 6-step trajectory. It does not reproduce the paper's quantitative results but demonstrates the belief-update and decision logic in isolation. ```bash python concept_colab.py # Outputs: output/f4_reproduced.pdf, output/f5_reproduced.pdf ``` --- ## License MIT — see `LICENSE`.

Institutions

• University of Bedfordshire

  England, Luton

Categories

Related Links

Licence