Abstract
Contours are essential data for delineating landforms on maps. A contour tree provides an object-oriented description of landforms and can be used to enrich topological information. Traditional contour trees store topological relationships in a hierarchical structure, identifying eminences and depressions as sets of nested contours. This research proposes an improved "spatial contour tree" that incorporates both topological and geometric information. Regarded as a 3D terrain skeleton, it is established based on spatial nodes of contours containing latitude, longitude, and elevation data. The spatial contour tree is built by connecting spatial nodes from low to high elevation for positive landforms, and from high to low elevation for negative landforms. The connections provide real distance and direction as 3D Euclidean vectors. This paper tests the construction method and demonstrates how the resulting hierarchical structure can distinguish landforms and assist in contour generalization while considering geographic characteristics.
TL;DR
This research introduces the "spatial contour tree," an evolution of traditional contour trees that integrates 3D geometric data (latitude, longitude, and elevation) with topological relationships. By treating the tree as a 3D terrain skeleton, the authors provide a robust framework for automated map generalization and quantitative terrain analysis.
The Challenge: Beyond Topology
In cartography and GIS, contour trees are standard tools for representing the hierarchy of landforms—showing how peaks and depressions are nested within one another. However, traditional contour trees are purely topological; they indicate that a peak exists within a certain contour, but lack information on where it is located or the steepness of the slope between levels.
This lack of spatial context hinders automated tasks like map generalization. Without geometric data, algorithms cannot easily distinguish between a significant mountain ridge and a minor feature that should be removed to reduce visual clutter. The authors bridge this gap by embedding 3D Euclidean vectors into the tree structure.
Methodology: Constructing the Spatial Tree
The authors propose a multi-step process to build a spatial contour tree:
- Node Determination: Traditionally, the "center" of a contour is its centroid. However, for irregular shapes, the centroid might fall outside the contour. To fix this, the authors use the geometric center of the largest triangle within a Delaunay Triangulated Irregular Network (DTIN) generated inside the contour, ensuring the node remains within the landform.
- Tree Construction: Contours are sorted by elevation. For positive landforms (mountains), nodes are connected from low to high elevation; for negative landforms (depressions), they go high to low.
- Data Enrichment: Each node is assigned a type (Common, Branch, or Ending) based on its "out-degree" (number of sub-nodes).
- 3D Edges: Unlike traditional trees, these edges are treated as 3D vectors storing length (real-world distance), vertical angle (slope), and horizontal direction (azimuth).
Key Insights: The Terrain Skeleton
The spatial contour tree acts as a "skeleton" inside the terrain, allowing researchers to visualize the internal structure of a mountain range much like a medical X-ray.
- Quantifying Importance: The vertical angle (AngleV) of a branch edge serves as a proxy for geographical importance. A steep angle indicates a prominent feature, while a shallow angle suggests a minor feature suitable for removal during map simplification.
- Branching as Complexity: Branch nodes represent points where a landform splits into multiple units. Analyzing the "out-degree" of these nodes allows for the quantification of landscape complexity.
Experimental Results
The method was tested using a 1:10,000 topographic dataset of Lushan Mountain, China, covering 138 contours across an elevation range of 1120m to 1150m. The system was implemented using ArcGIS Runtime SDK for .NET.
The researchers successfully generated a 3D representation where every node and edge contained precise geo-information. By identifying edges with low vertical angles, they demonstrated a preliminary automated generalization, "cleaning" the map by preserving main mountain features while removing minor geographical units.
Limitations and Future Work
As preliminary work, the authors identify several areas for future development:
- Refining Generalization: Moving beyond simple sub-tree removal to incorporate complex operators like merging adjacent contours.
- Node Logic: Further verification of the "largest triangle center" logic across diverse landform types.
- Broader Application: Extending the model into full terrain simulation and advanced landform element classification (e.g., automatically identifying ridges and valleys).
Summary updated