2D particle tracking in R, part 2: kernel density

In my previous post, I showed how particle movement can be simulated in the marine environment. Now that we have tracked the particle movement, how can we use this information to translate this to something more meaningful? This can be done by determining the particle density on the map.

The most simple way would be to lay a grid over the particles and simply count the number of particles in each grid cells. However, I would like to go for something a little bit complicated that gives some smoother results. I will use kernel density estimates. This means that I will consider the particles as a density function, were the position of the particle gives the highest density for that particle. You can get a density map of all particles by simply adding up the probabilities of the density functions of each individual particle. Fortunately, the KernSmooth package will take care of all this.

Remember that in the original simulation, I released 5,000 particles within a specific time-frame. With the kernel density estimated, I can estimate the number of particles per square kilometer at any time and location. This will become even more interesting if these particles represent something in the real world, let's say oil. If we know how much cubic meter of oil is represented by each particle, the estimates can be transformed in to the thickness of the oil slick. Keep in mind that this is just an initial test simulation that does not include all processes in order to properly simulate oil spills (e.g., oil evaporation, degradation and dispersion is not simulated).

Here's a video created with ffmpeg of the particle density movement over time.

And here's the fully commented script used to create the frames for the video.

2D particle tracking in R

How hard is it nowadays to simulate particle movement in the sea and track them in R? Well I discovered that it is not all that difficult. Within a couple of hours I had a simple simulation model running in the North Sea. All you need are time dependent water currents and wind data, both of which can be obtained from marine.copernicus.eu.

The main steps that you will have to follow:

• load the water currents and wind data
• specify particle release details
• specify simulation details
• loop all simulation time steps and for each step do the following with the particles that are released:
• for each particle determine the local water current and wind vector at the current time index
• transform the particle coordinates to a projection in meters
• move the particles with the water current (m/s) multiplied by the time step duration (s)
• move the particles with the wind speed (m/s) multiplied by the time step duration (s) and a drift factor (between 0 and 1)
• move the particles in random directions (to simulate diffusion)
• transform the particle coordinates back to the original projection
• save particle information every n^th simulation step, where n is any positive integer value

Below you will find the script that does all this (each of the steps above are marked in the script). In this simulation I simulate the release of 5,000 particles with a constant release rate during 5 days and followed them for a couple of days. I found the trickiest bit not the simulation itself, but reading the the NetCDF files such that it could be used in a convenient manner. I create a video of this simulation with ffmpeg:

So what's the practical use of all this? Well obviously you can simulate the trajectory of any floating materials, such as oil. However, you have to keep in mind that the simulation is only as good as the underpinning data and parameterisation. For instance, the water current data used here is not that accurate near estuaries. Of course you can always do multiple runs of your simulation with different parameter settings (or make it even a Monte Carlo simulation). Also, here I only simulate in 2D. But there is nothing that would stop you from modifying it to include the third dimension. What also should be added is some detection that particles have beached. But that should not be very difficult when you have coastline data.

Why in R? Well, commercial particle tracking software is generally expensive. Whereas free software are often rigid and don't allow the user to include custom modifications. In R the simulation is free, and you can take the exact implementation into our own hands... Hopefully, this post will help you get started with your own simulations. Good luck!

Here's the script:

Edit: if your interested, you might want to check out the follow-up post on particle density estimates of this simulation.