Bayesian Optimization Workflow for Medslik-II (bayes_opt_mdk2)

Bayesian Optimization Workflow for the Medslik-II Numerical Model, used for simulations of Oil Spill Events.

Description

This tool allows you to initiate a workflow based on Bayesian Optimization to perform Oil Spill Event simulations using the Medslik-II model. The goal is to optimize the model's initial parameters in an informed, physically consistent manner without requiring specific domain knowledge.

Features

The optimization process improves the Oil Spill simulation results by maximizing a metric of choice between:

• Fraction Skill Score (FSS)
• Overlay
• Centroid Skill Score (CSS)

To use the tool, you need to configure the model by modifying the /simulation_setup_file/workflow_config.toml file according to the chosen event. If you wish to compare the simulation result with a past event, you need to provide the path of the observation to be compared, along with the paths for the workflow and the output folder.

Configuration

In the /simulation_setup_file/workflow_config.toml file, besides specifying the characteristics of the simulation, it is important to define the number of "probe" and "exploration" iterations you wish to perform.

Example configuration for simulation_setup_file/workflow_config.toml

[bayesian_optimization]
b = [[0.0, 0.05], [0.0, 15.0], [0.0, 20.0]]

[bayesian_optimization.setup]
eval_metric = "<metric>"
init_points = 3
n_iter = 7
random_state = "None"
decimal_precision = 6
verbose = 2

[medslik2.sim_extent]
SIM_NAME = "syria"
sim_length = "0008"
duration = "0000"
spillrate = "00027.78"
age = "0"
grid_size = "150.0"
oil_api = 28.0
oil_volume = 2000.0
number_slick = 151

[medslik2.sim_date]
day = "24"
month = "08"
year = "21"
hour = "08"
minutes = "10"

[medslik2.sim_coords]
lat_degree = 35
lat_minutes = 15.6
lon_degree = 35
lon_minutes = 54.6
delta = 1

[medslik2.sim_params]
k = ["Wind correction (Drift Factor)", "Wind correction (Drift Angle at zero wind speed)", "Horizontal Diffusivity"]
v = [0.0, 0.0, 2.0]
kparticles = "No of parcels used to model diffusion and dispersion"
vparticles = 90000

[medslik2.data_format]
type = "analysis"
time_res = "day"
process_files = "True"

Instead of <metric>, insert FSS to use Fraction Skill Score, overlay to use Overlay or CSS to use Centroid Skill Score.

Execution Modes

The tool supports two different execution modes:

1. Optimization with Bayes Opt (Mode 0)

   • This mode uses Bayesian optimization to optimize the model's initial parameters.

2. Simulation without Bayes Opt (Mode 1)

   • This mode performs a simulation using the provided initial parameters without applying Bayesian optimization.

Example of Execution

To run the tool, use the following bash script:

#!/bin/bash

!conda run -p <path to your conda env>
ROOT="<path to the bay_opt_mdk2 dir>"
SIM="<path to Medslik-II dir>"
OBS="<path to single observation to compare dir>"

export PYTHONPATH=$PYTHONPATH:$ROOT && \
export SIMPATH=$SIMPATH:$SIM && \
export OBSPATH=$OBSPATH:$OBS && \

python $ROOT/src/main/main.py \
--mode <mode>

Replace <mode> with 0 for Bayesian optimization or 1 for simple simulation.

Instructions for Use

1. Configure the Model: Modify the /simulation_setup_file/workflow_config.toml file with the parameters of the event to be simulated.
2. Provide Necessary Paths: If you want to compare with a past event, make sure to provide the path to the observation.
3. Start the Optimization or Simulation Process: Run the appropriate command to start the desired workflow.