[GSoC Project Proposal]: Scalable OCSMesh: Parallelization and Spatial Partitioning

[felicio93]

Project Description

OCSMesh V.2.0+ is a Python package designed for generating unstructured meshes tailored to ocean modeling. A parallel version of OCSMesh size function was developed in GSoC 2025. This project expands on last years’ success to make OCSMesh a truly HPC-native tool capable of handling global-scale domains. There are multiple problems that can be addressed as a part of this year’s GSoC. The parallelization of size functions in GSoC 2025 needs to be expanded to ensure all mesh size functions and refinement algorithms are fully parallelized. Furthermore for OCSMesh to be relevant on HPC platforms the parallelization must be refactored such that it utilizes distributed architectures, using MPI (i.e. mpi4py). Lastly, the current parallelization logic relies on distinct input files, which creates significant bottlenecks when processing massive, single-source datasets or complex overlapping DEMs. This may be addressed by designing and implementing a spatial partitioning strategy to subdivide the domain prior to the mesh gen process, allowing distinct geographic regions to be processed independently regardless of the input file structure. This approach effectively decouples parallel performance from the input DEM data extent, enabling the efficient creation of high-resolution large domain meshes on supercomputing clusters.

Expected Outcomes

• Fully implementation of parallel processing for all size function generation and mesh refinement steps
• A fully functional implementation of mpi4py-based parallelization.
• Development of a robust spatial partitioning engine capable of “windowing” massive datasets into manageable chunks for parallel processing.
• Address current memory bottlenecks and the reliance on pre-tiled inputs.
• Evaluate the scalability of the approach and identify optimal number (and size) of partitions (or windows).

Skills Required

Python, Parallel Computing (multiprocessing and mpi4py), Unit Testing, Geospatial Technologies (GIS and geospatial python libraries)

Expected Project Size

350 hours

Project Difficulty

Expert

Mentor Name(s)

Felicio Cassalho, Soroosh Mani

Mentor GitHub Handle(s)

@felicio93, @SorooshMani-NOAA

Mentor Contact Email(s)

felicio.cassalho@noaa.gov, soroosh.mani@noaa.gov

Contributor Evaluation/GSoC GitHub Issue Links

No response

Additional Project Background/Links

No response

Previous Project Idea Link(s)

No response

[felicio93]

@eduardomatielo

[felicio93]

@gcoelho2

[felicio93]

@jbabatinca

[gcoelho2]

@kainguyen104

[felicio93]

@novaisc

[felicio93]

Dear potential collaborators,

I am excited to share that our organization (IOOS) has been officially approved for GSoC 2026! This puts us one step closer to getting our specific project approved as well.

In the meantime, I encourage anyone interested in collaborating to start getting familiar with OCSMesh. The package has undergone a major refactoring over the last couple of months, and we are looking forward to continuing its development with your help!

To get started, please review our OCSMesh report and tutorials. Please note that this report is currently a bit outdated. As a great first exercise, try going over the tutorial and adapting it to work with the newest version of OCSMesh.

For your reference, here are the major recent updates you will need to account for:

• Optional Dependencies: Mesh Engines (Jigsaw and Triangle) are now optional. For versions 2.0.0 and later, you will need to install them manually.
• New Default Engine: Gmsh has been adopted as the new default mesh engine, replacing Jigsaw.
• Class Replacement: jigsaw_msh_t was replaced by a new custom Python class called MeshData.
• Function Updates: Functions that previously operated on msh_t objects have been updated to support the new MeshData class.

Finally, please keep all communication centralized here on this ticket. Feel free to ask questions as we go along, we are happy to help!

Best regards,

[novaisc]

Hi @felicio93 , I'm interested in collaborating on this project.
I'll work on this first exercise and I'll get back to you soon. Thank you!

[parthshah02]

Hi @felicio93 and @SorooshMani-NOAA ,

Thank you for sharing this exciting GSoC project idea: Scalable OCSMesh: Parallelization and Spatial Partitioning. I’ve reviewed the description on GitHub and am very interested in contributing to this effort.

I’m particularly enthusiastic about working on improving the parallelization and spatial partitioning components of OCSMesh, and I believe my experience with Python and geospatial data aligns well with the project requirements.

I would love to learn more about your expectations for contributors and discuss any initial steps you recommend I take — such as setting up the development environment or reviewing related OCSMesh modules.

Regards !

[novaisc]

Hi @felicio93 ,

I worked on the first part of the tutorial and successfully created a mesh using the new Gmsh engine. I used a DEM file downloaded from the GEBCO website, focusing on a small region southwest of Greenland.

What would you suggest as the next steps? Is there any specific feature or region you would like me to test further?
Also, I have a question regarding the program's duration: I will likely only be able to commit to 175 hours. Given this constraint, is it still possible for me to apply?

Thank you.

[anas-ibrahem]

Hi @felicio93 @SorooshMani-NOAA,

I’ve been going through the OCSMesh report and tutorials and decided to try updating the QuickStart section. Since I could only find the PDF version, I put the updates into a Jupyter Notebook instead. I figured it’s actually pretty handy to have a Notebook version of the QuickStart for people to run directly.

If there’s a .docx or source file for that PDF, let me know! I’m happy to move my changes over to the document as well.
I’ve highlighted all my edits in the Notebook with "##Change" so they’re easy for you to spot. Here is the link: https://colab.research.google.com/drive/1opMRoGSdSvC8-RH18k6NaR4JyI15rVJX?usp=sharing

Waiting for your review and feedback to finalise it.

Also, I had a quick question about this part: "After interpolation the mesh object can be written to a file. Note that currently OCSMesh only supports writing mesh to disk in EPSG 4326." Is this still the case, or has it been updated recently?

[felicio93]

Hi @anas-ibrahem, thank you for putting this together. I don't' think we can share the editable version of the report, just because that is an official document. But please feel free to create tutorials or any kind of "unofficial" materials, we really appreciate that!

That is a good question and observation. We do have ways to convert the mesh from/to different projections. But it would be interesting to have something in the write function (like something that calls the reproject function from the utils).

[felicio93]

Hi everyone,

We have a very exciting project this year, and we are thrilled to see so much interest from such strong candidates!

For those of you who have already gone through the old tutorials and reports and successfully run OCSMesh, I suggest reviewing the recent evolution of our codebase to understand the latest changes.

We had a highly successful GSoC 2025 where our former collaborator parallelized the hfun class. Since parallelization is at the core of our 2026 GSoC project, we highly recommend spending some time analyzing this work and brainstorming potential improvements that align with this year's project. To see how this was implemented, please compare these two versions:

OCSMesh v1.6.4

OCSMesh v1.6.5

More recently, OCSMesh underwent a major refactoring and redesign effort. To get up to speed on our current architecture, please compare the post-parallelization version with our latest release:

OCSMesh v1.6.5 vs. OCSMesh v2.1.0

We look forward to hearing your thoughts and ideas!