Geospatial Solutions Architect | Computational Hydrologist Lyon, France
I engineer production-grade environmental systems. My work bridges the gap between Physical Simulation (PDE solvers) and Artificial Intelligence (Stochastic inference).
Most environmental workflows are static and fragmented. I build persistent, auto-calibrating digital twins. I design architectures where satellite telemetry forces hydrological models in real-time, scaled via HPC and Cloud infrastructure. I do not just run models; I architect the pipelines that make them operational, reproducible, and scalable.
My core architectural pattern integrates deterministic physics with data-driven ML. This topology handles the velocity of Earth Observation data without compromising physical consistency.
flowchart TD
subgraph L1 ["I. DATA INGESTION (STAC/ETL)"]
A1[("Sentinel-1/2 (SAR/MSI)")]
A2[("ERA5 / CMIP6 Reanalysis")]
A3[("In-Situ Telemetry")]
end
subgraph L2 ["II. THE HYBRID KERNEL"]
direction LR
B1["Latent Space Mapping (TorchGeo)"]
B2["Physics-Informed ML (PINNs)"]
B3["Numerical Solvers (TELEMAC/Wflow)"]
B1 --> B2
B2 <--> B3
end
subgraph L3 ["III. DISTRIBUTED COMPUTE"]
C1["Dask / xarray Orchestration"]
C2["HPC Kernels (SLURM/MPI)"]
end
subgraph L4 ["IV. OPERATIONAL DELIVERY"]
D1["Vector Tile Services"]
D2["Decision Support Systems"]
end
A1 & A2 & A3 -->|Normalized Stream| B1
B3 -->|State Vector| C1
C1 <--> C2
C1 -->|Zarr/COG| D1 & D2
style L2 fill:#0d1117,stroke:#00d4aa,stroke-width:2px,color:#fff
style C1 stroke:#d2a8ff,stroke-width:2px
I operate at the intersection of Physics, Code, and Infrastructure:
- Hybrid Modeling (Physics + AI): Moving beyond black-box ML. I embed physical constraints (mass conservation, momentum) into neural networks to create robust predictors for data-scarce environments.
- HPC & Cloud Scalability: Designing "compute-agnostic" pipelines that run seamlessly on on-premise SLURM clusters or AWS Fargate. I optimize for I/O bottlenecks using lazy loading (Dask) and cloud-native formats (Zarr/COG).
- Automated Calibration: Replacing manual parameter tuning with differentiable programming. Using gradient-based optimization to auto-calibrate hydrological parameters (Manning’s n, conductivity) against real-time observation.
The foundational layer for spatial manipulation.
Deterministic solvers for fluid dynamics and hydrology.
Stochastic modeling and distributed processing.
Reproducibility and deployment.
"Code is the modern notation for physical law."
I advocate for Open Science as a strict engineering requirement. Environmental models must be version-controlled, containerized, and documented to withstand scrutiny. If it cannot be re-run from scratch by a third party, it is not science—it is an anecdote.