Best Algorithms for Procedural Roads and Rivers

Procedural generation of natural branching road and river networks requires algorithms that can balance organic growth patterns, environmental constraints, and logical connectivity. Creating believable networks depends on simulating physical forces like gravity and water flow for rivers, or socio-geographic and terrain-aware navigation rules for roads. This guide examines the top algorithms for generating these networks in game development, including Space Colonization, Flow Accumulation, L-Systems, Graph-Based Pathfinding, and Tensor Fields, breaking down their specific use cases and implementations.

1. Space Colonization Algorithm

Originally designed to model leaf venation and tree branching, the Space Colonization Algorithm is one of the most effective methods for generating natural, organic branching networks.

• How It Works: A bounding area or terrain is seeded with "attractor points." The algorithm iteratively grows nodes from starting points (such as river mouths or city centers) toward the nearest attractors within a influence radius. Once an attractor is reached, it is removed, and branch segments naturally merge and split.
• Best Used For: Tributary-heavy river basins, delta networks, and organic rural path systems.

2. Flow Accumulation and Drainage Direction (D8 / D-Infinity)

For heightmap-based terrains, physics-driven drainage algorithms produce the most realistic river paths by directly utilizing gravity.

• How It Works: Using the D8 or D-Infinity model, each cell on a digital elevation model (DEM) determines its steepest downslope neighbor. Flow accumulation is calculated across the entire grid by tracking how much upstream water drains through each cell. Cells exceeding a specific accumulation threshold become river segments, naturally merging smaller trickles into wide main rivers.
• Best Used For: Topography-driven river systems, mountain streams, and natural water networks shaped by terrain erosion.

3. Graph-Based Networks: Delaunay Triangulation and Minimum Spanning Trees

When roads must connect specific points of interest (settlements, resource nodes, or dungeons), graph theory algorithms provide optimal structural skeletons.

• How It Works:
  1. Points of interest are connected using Delaunay Triangulation to create a non-overlapping graph of potential connections.
  2. A Minimum Spanning Tree (MST) or Relative Neighborhood Graph (RNG) filters these edges to ensure all nodes connect without redundant clutter, while preserving controlled cycles for alternate routes.
  3. Individual paths between connected nodes are carved out using A* Pathfinding modified with cost functions for slope, water bodies, and terrain difficulty.
• Best Used For: Major highway networks, trade routes, and inter-city road systems.

4. Extended L-Systems (Lindenmayer Systems)

L-Systems are string-rewriting grammars that excel at generating self-similar, hierarchical branching structures.

• How It Works: Production rules define how an initial axiom expands recursively. For game roads and rivers, these systems are extended with environmental sensors that dynamically alter generation rules based on local obstacles, terrain steepness, or boundary limits.
• Best Used For: Suburban street hierarchies (arterials, collectors, and local streets) and stylized branching streams.

5. Interactive Tensor Fields

Tensor fields guide generative agents along continuous vector fields, allowing fine control over the orientation and curvature of networks.

• How It Works: Developers define global and local constraints using mathematical tensor fields that follow terrain elevation contours, coastlines, or radial points. Tracing hyperstreamlines through the tensor field produces smooth, natural road or river layouts that respect both geometric grids and organic topography.
• Best Used For: Modern road networks that blend structured urban grids with organic, contour-hugging mountain passes.