GitHub

TravellingSailorProblem

Documentation for TravellingSailorProblem, a repository to fly particles (like balloons) inside a SpeedyWeather.jl simulation.

The Problem

The actual problem is:

Reach N predefined destinations with particles flying with the wind, launched anywhere on the globe. There will be more points for longer distances particles have flown before reaching their destination but and negative points if they miss it.

So it's only vaguely related to the Travelling Salesman problem which is about minimizing a path you can choose. Here it's about maximising a path determined by the (turbulent) wind field but you can choose the starting point.

Contents

And specifically for submitting to the TravellingSailorProblem:

Older versions of the documentation correspond to previous challenges used for teaching and can be looked up as an archive of submissions to the TravellingSailorProblem

Quick start

After the Installation copy and paste this into your Julia REPL, notebook etc

using SpeedyWeather, TravellingSailorProblem

# create a simulation for 26 children
spectral_grid = SpectralGrid(trunc=31, nlayers=8, nparticles=26)
particle_advection = ParticleAdvection2D(spectral_grid, layer=8)
model = PrimitiveWetModel(spectral_grid; particle_advection)
simulation = initialize!(model, time=DateTime(2026, 11, 14))

# add 10 children as destinations
children = TravellingSailorProblem.children(26)
add!(model, children)

# add particle tracker
particle_tracker = ParticleTracker(spectral_grid)
add!(model, :particle_tracker => particle_tracker)

# adjust initial locations of particles
(; particles) = simulation.prognostic_variables
particles .= rand(Particle, 26)     # all 26 random (the default)
particles[1] = Particle(25, 35)     # or individually 25˚E, 35˚N
particles[2] = Particle(-120, 55)   # 120˚W, 55˚N
particles[3:10] .=                  # or several ones along the equator
  [Particle(lon, 0) for lon in 100:20:240]

# then run! simulation until Christmas
run!(simulation, period=Day(40))

# evaluate
evaluate(particle_tracker, children)

# and visualise
using GLMakie
globe(children, particle_tracker)

And you have the basic setup to create a simulation, place particles in it, run, evaluate and visualise. Change the particle departures to score higher, see more information in the TravellingSailorProblem instructions.

Installation

First install Julia via juliaup (the recommended default way). TravellingSailorProblem.jl is a registered Julia packge so from the Julia REPL you can then do

julia> ] add TravellingSailorProblem, SpeedyWeather, GLMakie

where ] opens Julia's package manager interactively. SpeedyWeather is a dependency anyway and you will need its exported function and types to run a simulation. GLMakie is Makie's backend we use for visualisation: static and 3D+interactive. Don't use CairoMakie, as it will not render the 3D properties correctly. Equivalently, you can do

using Pkg
Pkg.add(["TravellingSailorProblem", "SpeedyWeather", "GLMakie"])

and after all dependencies are installed with julia> using TravellingSailorProblem, SpeedyWeather, GLMakie you load the packages, ready to solve the TravellingSailorProblem!

You can use this software in different ways,

Coming from Python? Have a look at a concise list of noteworthy differences.

Julia Jupyter kernel

To install a Julia kernel for Jupyter notebooks (assuming you already have jupyter notebooks installed) you need a package called IJulia

julia> ] add IJulia

After IJulia is installed do once

using IJulia
installkernel("Julia")

this will add a Julia kernel you can choose among other Python kernels in jupyter lab or jupyter notebooks.