Senior Research Engineer | Climate & Geophysical Models · HPC · Predictions · AI Adoption

I am a Senior Research Engineer at the Barcelona Supercomputing Center (BSC), where I work on Earth system models (EC-Earth4, OpenIFS, NEMO), co-lead Auto-EC-Earth4 ensemble workflows on MareNostrum 5, and lead AI adoption across the research group.

For years, advancing the fidelity of climate models has been hindered by escalating physical and software complexity—multi-million line legacy Fortran codes, fragile multi-component coupling layers, and intractable debugging across supercomputing tiers. The promise of modern AI is making high-fidelity Earth system prediction affordable to science again: coupling fast neural emulators with physical dynamical cores, and replacing brittle manual scripts with schema-validated, deterministic workflows like Climate Commons (cvc-commons).

HPC Architecture AI Adoption OpenIFS CY48R1.1 EC-Earth4 MareNostrum 5 cvc-commons Consortium Coordination

What I Focus On

  • HPC Systems & Ensemble Workflows: Co-leading Auto-EC-Earth4 ensemble workflows on MareNostrum 5. I focus on optimizing multi-component coupling (OASIS3-MCT), eliminating parallel filesystem bottlenecks (Lustre OST striping and metadata contention), and orchestrating CMIP7 and WMO decadal production cycles under strict capacity constraints.
  • AI Adoption & Model Emulation: Leading group-wide AI adoption to transform scientific workflow development, evaluate neural emulators for coupled Earth system components, and implement automated validation harnesses.
  • Atmospheric Physics & Numerics: Making sure atmospheric models stay physically realistic when pushed to higher resolutions or lower precision. Right now, that means investigating tracer mass conservation in ECMWF OpenIFS CY48R1.1, diagnosing why single-precision runs drift, and implementing SIMD-friendly stochastic rounding to eliminate numerical truncation errors.
  • Declarative Scientific Pipelines: Author of Climate Commons (cvc-commons) and declarative catalog pipelines for ocean initial conditions (pisces-inidata / NEMO). Transforming scientific software into schema-validated, testable components.
  • Physical Modeling & Fluid Dynamics: Grounded in classical geophysical fluid dynamics—from high-resolution ICON sea-ice rheology solvers (mEVP) to global barotropic and internal ocean tide spectral solvers (baro_tides_fd, internal_tides_fd).

Tech Stack & Tools

  • Languages: Python (Xarray, Dask, PyTorch, NumPy, Pandas), Modern & Legacy Fortran (MPI, OpenMP), C++, Bash.
  • HPC & Platforms: MareNostrum 5, Slurm, Autosubmit, LMOD modules, uv, CDO, NCO, NetCDF-C, Git/CI/CD.
  • Models: EC-Earth4, OpenIFS (CY48R1.1), NEMO/PISCES, OASIS3-MCT.