Physicist and Climate Systems Engineer specializing in Numerical Weather Prediction (NWP), Earth System Modeling, and High-Performance Scientific Computing (HPC).
My background spans research & operational environments at Met Éireann and Nanjing University, focusing on:
- 🌎 Earth System Science & Climate Dynamics: Global & regional model porting, dynamical cores, and physical parameterizations.
- ⚡ High-Performance Architecture: Low-level dynamics (C++/Fortran), high-productivity ecosystems (Julia/Python), and HPC containerized infrastructure (Docker).
- 🔬 Scientific Machine Learning (SciML): Hybrid physical–AI modeling, system identification (SINDy), and probabilistic forecasting for climate risk and energy grid analytics.
✍️ Writing & Open-Source: I publish technical deep-dives on climate modeling on my Substack (with companion code in: GitHub - A Journey Through Climate Modelling.
- Operational Weather & Climate Dynamics (Met Éireann R&D experience)
- Climate Variability, Extremes & Physical Risk Analytics
- Hybrid Physical–AI Emulators & Parameterizations
- Energy Infrastructure & Meteorological Coupling
- Sparse Identification & Data-Driven Dynamical Discovery (SINDy)
- Time-Series Forecasting & Uncertainty Quantification (UQ)
- High-Performance I/O & Parallel Pipelines for Terabyte-Scale Grids (NetCDF/HDF5/Zarr)
- Systems & Core Computing: C++, Fortran (NWP/Legacy integration), Julia (High-performance dynamics & SciML), Bash
- Data-Driven & Orchestration: Python, PyTorch
- Scientific Data & High-Performance I/O: NetCDF4, HDF5, Zarr, Xarray, Dask, Pandas
- Infrastructure & Reproducibility: Docker, Git, Linux/HPC Clusters, Vim, Tmux
Three-stage framework for day-ahead electricity price forecasting and automated battery storage arbitrage, developed at Nanjing University for the 4th World AI4S Competition.
- Key Architecture:
- Stage 1: Bias correction of raw meteorological and market forecasts.
- Stage 2: Day-ahead electricity price forecasting (HGBR vs. Task-based End-to-End DL).
- Stage 3: Automated battery storage dispatch optimization for profit maximization.
- Key Insight: Demonstrated that profit-aware models outperform traditional RMSE-aware models—proving that lower forecast error does not strictly imply higher market arbitrage value.
- 🔗 GRIPS2026-project
Equation discovery and reduced-order modeling of non-linear paleoclimate transitions using sparse regression and system identification.
- Key Method & Physics:
- Applied Sparse Identification of Nonlinear Dynamics (SINDy) and symbolic regression to discover governing non-linear differential equations directly from paleoclimate data.
- Modeled non-linear phase transitions, orbital forcing responses, and low-dimensional attractor dynamics governing Quaternary glacial–interglacial cycles.
- 🔗 GitHub Repository | 📄 Read Full Thesis (Institutional Repository)