🌊 Meteo-Scenarios

Reducing weather scenario sets for efficient and robust performance optimisation of wind-assisted ships

---

🧭 Overview

Meteo-Scenarios is an open-source Python package developed for the study and optimisation of Wind-Assisted Ships (WASP) under realistic meteorological variability.
It provides a complete and reproducible framework for scenario reduction, time-series clustering, and synoptic-scale weather analysis.

The objective is to reduce the computational burden associated with large-scale weather scenario sets, which often involve thousands of simulated voyages, while retaining the essential statistical and dynamical variability of the original data.

By identifying representative synoptic sequences using clustering techniques such as MiniBatch K-Means and Dynamic Time Warping (DTW) + PAM medoids, the method enables efficient yet robust performance prediction, design optimisation, and mission-level analysis of wind-assisted vessels.

---

🔬 Scientific Context

Performance prediction and optimisation of wind-assisted ships typically rely on multi-year weather scenario ensembles.
Brute-force evaluation ensures statistical robustness but becomes computationally prohibitive when coupled with ship design optimisation, routing simulation, or fleet-level planning.

This package introduces a methodology for synoptic scenario reduction based on clustering of meteorological time-series windows.
Each window represents a sequence of spatial weather states (e.g. 10 m wind, significant wave height), clustered according to similarity metrics tailored to time series:

• Euclidean distance for rapid analysis via MiniBatch K-Means
• Dynamic Time Warping (DTW) for alignment-invariant temporal comparison, combined with Partitioning Around Medoids (PAM) for robust representative selection

The reduced scenario sets preserve both the statistical diversity and the seasonality of the original data, while reducing routing computations by orders of magnitude.

Such reduced sets can reproduce voyage-time and fuel-consumption distributions with high fidelity, making joint design–routing optimisation of WASP tractable in realistic industrial studies.

---

⚙️ Key Capabilities

• Aggregation of multiple GRIB/NetCDF datasets onto a unified grid and timeline
• Sliding-window sequence generation and feature embedding (with optional PCA)
• Clustering with K-Means or DTW+PAM medoids
• Export of representative medoid scenarios as GRIB2 and CSV summaries
• Validation tools for spatial maps and local time-series diagnostics
• Mission-level analysis enabling integration of design, routing, and fleet-planning decisions under uncertainty

---

📚 Related Research

This package builds on methods widely used in other domains for time-series aggregation and uncertainty reduction:

• Paparrizos, J., et al. (2024). Bridging Time Series Data Mining and Deep Learning via Representation Learning: A Survey.
• Teichgräber, H., et al. (2022). Time-Series Aggregation for Energy System Design: Review and Application.
• Yerbury, E., et al. (2025). Comparing Clustering Techniques for Renewable Energy Forecasting.
• Wei, L., et al. (2022). Joint Design and Routing Optimization for Wind-Assisted Ships.
• Meng, Q., et al. (2014). Containership Routing and Scheduling under Weather Uncertainty.
• Cao, J., et al. (2025). Risk-Aware Fleet-Level Optimisation under Weather Uncertainty.

---

🧩 Repository Structure

meteo-scenarios/
├─ src/meteo_scenarios/
│  ├─ io.py              # I/O utils (open, export GRIB, normalize dims)
│  ├─ gridtime.py        # Temporal & spatial alignment
│  ├─ clustering.py      # KMeans & DTW-PAM
│  ├─ windows.py         # Sliding window generation
│  ├─ export.py          # GRIB exports + label handling
│  ├─ plotting.py        # Maps & probe diagnostics
│  └─ cli/
│     ├─ aggregate.py    # meteo-aggregate
│     ├─ reduce.py       # meteo-reduce
│     ├─ plot_map.py     # meteo-plot-map
│     └─ probe.py        # meteo-probe
└─ examples/
   ├─ 01_merge_wind_wave.sh
   ├─ 02_reduce_sequences.sh
   └─ 03_probe_plots.sh

🚀 Typical Workflow

1 - Merge heterogeneous meteorological datasets

meteo-aggregate \
  --in "gribs/wind_2020.grib" \
  --in "gribs/wave_2020.grib" \
  --target-grid finer \
  --time-mode freq --time-freq 12H \
  --out merged_12H.grib2

2 - Reduce to representative synoptic scenarios

meteo-reduce merged_12H.grib2 \
  --vars u10,v10 \
  --window-hours 72 \
  --stride-hours 24 \
  --clusters 6 \
  --seq-metric euclid \
  --out reduced/

3 - Visualise and validate

meteo-plot-map reduced/original_with_cluster_id.grib2
meteo-probe reduced/original_with_cluster_id.grib2 \
  --probe-lat 45.0 --probe-lon -20.0 \
  --plot-out probe.png

🔧 Installation

git clone https://github.com/cdhainaut/meteo-clustering.git
cd meteo-clustering
pip install -e .

🧠 Research Scope

This tool supports studies on:

• Scenario reduction for performance prediction of WASP
• Design–routing coupling under uncertain meteorological forcing
• Fleet-level and mission-level optimisation
• Uncertainty quantification and robust decision-making for maritime decarbonisation

The resulting datasets and models can be used as reproducible inputs for optimisation frameworks, routing solvers, or Monte-Carlo performance assessment.

🧾 Citation

If you use this package in your research, please cite:

@inproceedings{dhainaut2025meteo,
  title   = {Reducing Weather Scenario Sets for Efficient and Robust Performance Optimisation of Wind-Assisted Ships},
  author  = {Dhainaut, Charles and Sacher, Mathieu},
  booktitle = {7th Innov'Sail Symposium},
  year    = {2026},
  address = {Göteborg, Sweden},
  organization = {ENSTA Bretagne, IRDL}
}