dRoute: Differentiable River Routing Library

A differentiable river routing library for hydrological modeling. dRoute implements multiple routing methods with automatic differentiation support, enabling gradient-based parameter optimization and integration with machine learning workflows.

Note dRoute is in active development - expect experimental code

Features

• 6 Routing Methods: From simple lag routing to Saint-Venant equations
• Dual AD Backends: CoDiPack (tape-based) and Enzyme (source-to-source)
• Network Topology: Full support for river networks with tributaries and confluences
• PyTorch Integration: Use dRoute gradients in ML training loops

Routing Methods

Method	Class	Physics	Speed	Use Case
Muskingum-Cunge	MuskingumCungeRouter	Kinematic + diffusion approx	~4,500/s	Production routing
Lag	LagRouter	Time delay buffer	~20,000/s	Baseline comparison
IRF	IRFRouter	Gamma unit hydrograph	~1,000/s	Fast calibration
KWT-Soft	SoftGatedKWT	Kinematic wave tracking	~4,000/s	Differentiable Lagrangian
Diffusive Wave	DiffusiveWaveIFT	Diffusion wave PDE	~3,000/s	Flood wave attenuation
Saint-Venant	SaintVenantRouter	Full shallow water eqs	~100/s	High-fidelity benchmark

Quick Start

Build requirements: CMake 3.15+, a C++17 compiler, and Python development headers. Optional features (NetCDF, Enzyme, SUNDIALS) require those libraries installed.

Build Options

CMake Options

Option	Default	Description
DMC_BUILD_PYTHON	OFF	Build Python bindings
DMC_USE_CODIPACK	ON	Enable CoDiPack AD
DMC_ENABLE_ENZYME	OFF	Enable Enzyme AD backend
DMC_ENABLE_SUNDIALS	OFF	Enable SUNDIALS for SVE solver
DMC_ENABLE_NETCDF	OFF	Enable NetCDF topology I/O
DMC_ENABLE_OPENMP	OFF	Enable OpenMP parallelization
DMC_BUILD_TESTS	ON	Build C++ test suite

Build with All Features

cmake -S . -B build \
    -DDMC_BUILD_PYTHON=ON \
    -DDMC_ENABLE_ENZYME=ON \
    -DDMC_ENABLE_SUNDIALS=ON \
    -DSUNDIALS_ROOT=/path/to/sundials/install \
    -DDMC_ENABLE_NETCDF=ON
    -DMC_ENABLE_OPENMP=ON

cmake --build build -j4

macOS (Apple Silicon)

brew install cmake netcdf sundials

cmake -S . -B build -DDMC_BUILD_PYTHON=ON -DDMC_ENABLE_SUNDIALS=ON
cmake --build build -j$(sysctl -n hw.ncpu)

Python installation

git clone https://github.com/DarriEy/dRoute.git
cd dRoute
pip install -e .

# Or use PyPi

pip install droute

Data

Download Sample Data

The example dataset is hosted as a GitHub release asset (v0.5.0).

python scripts/download_data.py

Run with Sample Data

# Forward pass comparison (all methods)
python python/test_routing_with_data.py --data-dir data

# Fast optimization with Enzyme kernels (30 epochs in ~30s)
python python/test_routing_with_data.py --data-dir data --optimize --fast

# Include Saint-Venant high-fidelity benchmark
python python/test_routing_with_data.py --data-dir data --sve

Architecture

dRoute/
├── include/dmc/
│   ├── router.hpp              # MuskingumCungeRouter with CoDiPack AD
│   ├── advanced_routing.hpp    # IRF, KWT, Diffusive routers  
│   ├── kernels_enzyme.hpp      # Enzyme-compatible kernels (all 5 methods)
│   ├── unified_router.hpp      # EnzymeRouter wrapper
│   ├── saint_venant_router.hpp # Full SVE with SUNDIALS CVODES
│   ├── network.hpp             # Network topology
│   └── types.hpp               # AD type definitions
├── python/
│   ├── bindings.cpp            # pybind11 bindings
│   └── test_routing_with_data.py
├── tests/                      # C++ test suite
└── CMakeLists.txt

Requirements

• C++17 compiler (GCC 7+, Clang 5+, MSVC 2019+)
• CMake 3.15+
• pybind11 (auto-downloaded if not found)
• CoDiPack (auto-downloaded)
• Optional: Enzyme, SUNDIALS, NetCDF-C, OpenMP

Citation

@software{dRoute2025,
  title={dRoute: Differentiable River Routing Library},
  author={Eythorsson, Darri},
  year={2024},
  url={https://github.com/DarriEy/dRoute}
}

License

MIT License - see LICENSE for details.

Acknowledgments

• CoDiPack - Tape-based automatic differentiation
• Enzyme - Source-to-source AD compiler plugin
• SUNDIALS - Implicit ODE solvers
• SUMMA & mizuRoute - Hydrological modeling inspiration