Skip to content

A computationally efficient and massively parallel marine litter transport model.

License

Notifications You must be signed in to change notification settings

TheOceanCleanupAlgorithms/ADVECT

Repository files navigation

ADVECT V1.0

ADVECT is a whole-ocean marine debris transport model which is built to handle millions of particles and terabytes of data. It models the transport of debris based on its size, shape, and density, and simulates basic physical processes including 3D ocean current-driven advection, wind-driven drift, wind-driven near-surface vertical mixing, buoyancy-driven vertical transport, and eddy-driven diffusion. It automatically processes forcing datasets arbitrarily larger than memory capacity, and supports fully parallelized computation on CPUs and GPUs via OpenCL.

Model Description

ADVECT contains solvers (kernels) for two domains: ocean surface, and whole-ocean.

  • 2D kernel: model domain is constrained to the surface of the ocean (assumption: floating debris), and debris particles are idealized, with no consideration of their size/shape/density.
  • 3D kernel: model domain is the whole oceans, from surface to bathymetry, and physical processes depend on the size/shape/density of debris.

2D Kernel

Each particle is released at some location in space and time. Upon release, each particle is transported according to the following physical processes:

Surface ocean current-driven advection

Particles are transported in a time-evolving 2D velocity field of surface ocean currents, which the user must provide. The particles are advected according to one of two schemes: forward-Euler, or a second-order Taylor-expansion scheme which corrects for the outward-drift error the Euler method experiences in a curved field.

Wind-driven drift

Optionally, the user may provide a time-evolving 2D velocity field of 10-meter wind, which will move particles according to a user-provided windage coefficient.

Eddy-driven diffusion

Finally, the user may specify a constant eddy diffusivity, which will add random noise to the particle's movements. This simulates the effect of eddies smaller than the spatial resolution of the ocean currents.

Boundary processes

The model domain only includes the surface waters of the ocean (as defined by the non-null region in the ocean current vectorfield); particles cannot leave this domain, and thus the model does not include beaching. Instead, when particles are pushed against a coastline, their onshore displacement component is cropped to keep them in the model domain, generally resulting in a lateral displacement, as if the boundary was frictionless.

3D Kernel

Each particle is initialized with a size, shape, density, release date, and release location. Upon release, particles are transported according to the following physical processes:

3D ocean current-driven advection

Particles are transported according to a time-evolving 3D velocity field of ocean currents, which the user provides. The particles are advected according to one of two schemes: forward-Euler, or a 3D adaptation of the second-order Taylor-expansion scheme from the 2D kernel.

Buoyancy-driven transport

The user must provide a time-evolving 3D dataset containing the density of seawater in the ocean domain. Particles are transported vertically according to their terminal sinking (or rising) velocity, which is calculated using their size, shape, and density, as well as the density of the surrounding seawater.

Wind-driven drift

Optionally, the user may provide a time-evolving 2D velocity field of 10-meter wind, which will move particles floating at the surface based on a parameterization which depends on their emerged surface area. The user may optionally provide a multiplier which scales this drift, for the sake of experimentation.

Wind-driven vertical mixing

If wind is provided, the user may optionally enable the simulation of wind-driven vertical mixing. Mixing transport is based on an equilibrium between a particle's rising velocity and the size of ocean waves (estimated from wind).

Eddy-driven diffusion

Finally, the user may specify a vertical profile of vertical and horizontal eddy diffusivities, which will add noise to the particle's movements according to its depth. This simulates eddies smaller than the spatial resolution of the ocean currents, and allows the user the flexibility to account for the depth-dependent nature of eddy diffusivity in the world's oceans.

Boundary processes

The model domain only includes the waters of the ocean above bathymetry (as defined by the non-null region in the ocean current vectorfield); particles cannot leave this domain, and thus the model does not include beaching or sedimentation. Instead, when particles are pushed against coastline/bathymetry, their out-of-domain displacement components are cropped to keep them in the model domain. This is the 3D analog of the frictionless coastlines used in the 2D kernel, and similarly allows particles to travel parallel to domain boundaries.

Installation Instructions

  1. Install miniconda (if you don't already have it), to manage dependencies. If you are not already familiar with conda, what it is, and what it's for, you should read up here.

  2. Install ADVECT as a package by opening a terminal and running

    conda create -n advect  # create a fresh conda environment
    conda activate advect   # enter the environment
    conda install pip         # install pip (python package manager) into this environment
    pip install ADVECTOR      # install ADVECTOR from a repository in ~the cloud~
    
  3. Acquire forcing data

    Run ADVECTOR_download_sample_data and follow the prompts.

  4. Run example advection

    Run ADVECTOR_examples_2D or ADVECTOR_examples_3D and follow the prompts to see it in action!

Using ADVECT in your own programs

The key entry-point scripts to ADVECT are ADVECTOR/run_advector_2D.py and ADVECTOR/run_advector_3D.py. Those files include documentation on all their respective arguments. There are also supplementary documentation files in the documentation folder; you'll want to read all of these carefully to understand what you can/can't feed into ADVECT, and what it'll give you back.

In short, your script will look something like:

from ADVECTOR.run_advector_2D import run_advector_2D  # sim for run_advector_3D
outputfile_paths = run_advector_2D(<many arguments here>)

That's it!

If you need information on the arguments and don't want to refer directly to the source code, just open an interactive python prompt, import the runner as above, then run help(run_advector_2D).

As a general strategy, you can pretty much copy the structure of ADVECTOR/examples/ECCO_advect_2D.py or ADVECTOR/examples/ECCO_advect_3D.py, providing your own data and generating your own source/configfiles. The examples exist for your reference!

Extra: the INTEGRATOR

3D ocean model output generally only includes the zonal/meridional current velocity; ADVECT comes bundled with a tool called the INTEGRATOR which can generate vertical velocity fields from zonal/meridional velocity fields, using the continuity equation. Check out INTEGRATOR/README.md for more information. Currently it doesn't install via pip, so you'll need to clone this repository and run the files directly.

Hardware compatability

At this time, ADVECT only has known support for CPUs/GPUs with opencl driver versions 1.1, 1.2, and 2.1. If you are getting OpenCL/GPU related errors, you can run this in a python prompt to directly check your driver version, as that could be the problem:

import pyopencl
print(pyopencl.create_some_context(interactive=True).devices[0].driver_version)

Follow the instructions to select a compute device, and its driver version will be displayed.