Terrain Representation and Distinguishing Ability of Roughness Algorithms Based on DEM with Different Resolutions

Jiang Wu, Junjie Fang, Jiangbo TianISPRS Int. J. Geo Inf 2019 · DOI

TL;DR

This paper investigates how different digital elevation model (DEM) resolutions affect the ability of four common roughness algorithms (SLOPE, RMSH, VD, and 2D CWT) to represent and distinguish terrain features. The findings reveal that SLOPE and RMSH are more robust to resolution changes, while VD and 2D CWT are more sensitive, offering insights into selecting appropriate algorithms and DEM resolutions for terrain analysis.

What Problem Does This Paper Address?

Understanding the roughness of terrain is crucial for various earth science applications, from geomorphology to fluid dynamics. Roughness quantifies the surface's irregularity and can reveal information about erosion, surface processes, and landscape characteristics. However, the accuracy and effectiveness of terrain roughness calculations are heavily influenced by the resolution of the Digital Elevation Model (DEM) used as input. Coarser DEMs can smooth out fine details, potentially leading to misinterpretations of terrain features. This paper addresses the critical question of how different DEM resolutions impact the performance of various roughness algorithms, aiming to provide guidance on selecting the most suitable methods for terrain representation and distinguishing ability across different landscape types.

Methods

The study employed four distinct roughness extraction algorithms:

  • SLOPE: Measures the angle between the tangential plane and the horizontal plane at any given point.
  • Root Mean Squared Height (RMSH): Quantifies surface roughness by calculating the standard deviation of elevation within a local neighborhood.
  • Vector Deviation (VD): Assesses roughness based on the variation in the unit direction vector of the surface, reflecting changes in surface orientation.
  • Two-Dimensional Continuous Wavelet Transform (2D CWT): A spectral analysis method that analyzes the energy distribution in the spatial-frequency domain to capture terrain characteristics within specific frequency ranges.

These algorithms were applied to DEMs of varying resolutions (5m, 10m, 25m, and 50m). The effectiveness of each algorithm was evaluated using two key metrics:

  • Effect Size (ES): A statistical indicator measuring the magnitude of difference between roughness values in different sample areas (plain, hill, mountain). Larger ES values indicate better distinguishing ability.
  • Entropy: A measure adapted from information theory to quantify the completeness and complexity of the roughness results. Higher entropy suggests richer surface information.

Datasets

The research utilized DEM data generated from 1:10,000 topographic map data for the Fuyang District in Hangzhou, China. This data was processed using Anudem software to create hydrologically-correct DEMs at four different resolutions: 5m, 10m, 25m, and 50m. Three specific sample areas representing diverse geomorphological types were selected: a hilly area (A), a plain area (B), and a mountainous area (C).

Key Insights and Ideas

  • Resolution Matters, But Not Equally: While DEM resolution generally impacts terrain expression, some roughness algorithms are far more sensitive to these changes than others. This highlights the importance of algorithm selection in conjunction with DEM resolution.
  • SLOPE and RMSH are Robust: These two algorithms consistently performed well across different resolutions and terrain types, suggesting they are reliable choices when DEM resolution might be a concern or when a general understanding of terrain variation is needed.
  • VD and 2D CWT are Sensitive: The performance of Vector Deviation and 2D CWT varied significantly with resolution. This suggests they are better suited for analyses where high-resolution DEMs are available and fine-scale terrain details are critical.
  • Complex Terrain Loses Detail Faster: The rate at which terrain detail is lost due to decreasing resolution is faster in more complex geomorphological areas (hills and mountains) compared to flatter regions.
  • 2D CWT Excels at High Frequencies: In complex terrain, 2D CWT showed higher entropy, indicating its strength in capturing high-frequency surface information, which is crucial for detailed terrain analysis.

Key Results

  • Terrain Expression: All four algorithms showed a decrease in terrain expression ability as DEM resolution decreased. SLOPE and RMSH were found to be better than VD and 2D CWT in this regard.
  • Terrain Distinguishing Ability (Effect Size):
    • SLOPE and RMSH showed little change in their ability to distinguish between different terrain types (plain, hill, mountain) as resolution decreased.
  • VD and 2D CWT were more sensitive to resolution changes. For VD, the distinguishing ability between plain and hilly/mountainous areas generally increased with coarser resolutions, while the difference between hilly and mountainous areas decreased. The trend for 2D CWT was often the opposite.
  • Entropy:
    • Entropy generally decreased with decreasing DEM resolution for all algorithms.
  • The loss of entropy was faster in complex terrain (hills and mountains) than in plain areas.
  • In complex terrain, 2D CWT had higher entropy than VD, indicating it better captured high-frequency components. In flat areas, VD had higher entropy.

Limitations and Future Work

  • Window Size: The study used a fixed 5x5 pixel moving window for most analyses. The authors acknowledge that the window size is an important factor that influences results and suggest further investigation into its impact on roughness measures and its relationship with DEM resolution.
  • Advancements in Data Acquisition: The paper notes the increasing availability of high-resolution point-cloud data (from LiDAR, InSAR, etc.) which can generate centimeter-level DEMs. Future research could leverage these datasets for even more detailed roughness analysis.
  • Scale Considerations: The study emphasizes the importance of considering roughness at large, medium, and small scales. Future work could further explore how different algorithms perform across a wider range of scales and resolutions.

Summary updated

BibTeX
@article{wu2019terrain,
  author = {Jiang Wu and Junjie Fang and Jiangbo Tian},
  title = {{Terrain Representation and Distinguishing Ability of Roughness Algorithms Based on DEM with Different Resolutions}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2019},
  doi = {10.3390/ijgi8040180},
}
Figure