High-Resolution Terrain Ruggedness Mapping Through Remote Sensing and Machine Learning
High-Resolution Terrain Ruggedness Mapping is the focus of a scientific report developed by the Research Institute for Earth Sciences in cooperation with the UNESCO Chair on Coastal Geo-Hazard Analysis.
Published in 2026, the report presents the development and validation of a hybrid remote sensing and machine learning framework for estimating the K-coefficient as a continuous and transferable measure of terrain ruggedness and geomorphological complexity.
The research integrates Digital Elevation Model derivatives with Sentinel-1 Synthetic Aperture Radar data, Sentinel-2 vegetation information and machine learning techniques to produce standardized terrain ruggedness maps at a spatial resolution of 10 metres.
A Hybrid Framework for K-Coefficient Mapping
Conventional terrain assessment methods often rely on individual morphometric parameters such as slope, curvature or the Terrain Ruggedness Index. Although these indicators provide important information about surface geometry, they may not adequately represent physical surface characteristics such as vegetation coverage, radar scattering roughness, moisture conditions and human-induced landscape modifications.
To address these limitations, the proposed framework combines several sources of geospatial information, including:
- Digital Elevation Model data and morphometric derivatives
- Terrain slope, aspect, curvature and Terrain Ruggedness Index
- Sentinel-1 VV radar backscatter
- Sentinel-2 Normalized Difference Vegetation Index
- Water-related and agricultural human-layer indicators
- Machine learning and Ridge Regression models
All spatial datasets were projected into appropriate Universal Transverse Mercator coordinate systems and aligned to a standardized 10-metre grid to ensure accurate pixel-level correspondence.
Study Areas in Iran
The framework was evaluated across three contrasting geomorphological regions in Iran: Zagros, Ravar and Moghan.
The Zagros study area represents a complex mountainous and lithology-controlled environment. Its carbonate rocks, fracture networks and karst-related landforms create conditions in which surface morphology alone cannot fully explain terrain ruggedness.
Ravar represents an arid and rugged landscape in which terrain characteristics are predominantly controlled by elevation, slope and local topographic variation.
Moghan represents a low-relief agricultural plain influenced by vegetation, irrigation infrastructure and other human modifications.
The comparison of these regions allowed the researchers to evaluate whether a single model could maintain its reliability across landscapes with fundamentally different geological, climatic and topographic conditions.
Machine Learning and the ReliabilityGate Mechanism
The research progressed from a morphometric baseline model to a hybrid machine learning architecture incorporating topographic, radar, optical and anthropogenic variables.
A major innovation introduced in the report is the ReliabilityGate mechanism. This mechanism evaluates the level of disagreement between a stable topographic Ridge Regression model and a non-parametric baseline model.
When disagreement exceeds a defined reliability threshold, the framework limits the influence of the unstable component and relies more strongly on the topographic model. This approach was designed to prevent “mixture collapse,” in which combining inconsistent models reduces the reliability of cross-regional predictions.
The report indicates that a reliability threshold of 0.80 provided the most effective balance between model stability and predictive performance.
Main Research Findings
The results demonstrate that terrain ruggedness behaves differently across distinct geomorphological regimes.
In Ravar and Moghan, the K-coefficient showed strong relationships with topographic parameters, with correlations of approximately 0.88 to 0.90. These regions were therefore identified as predominantly topo-driven environments.
In contrast, the Zagros region showed a substantially weaker relationship between terrain ruggedness and standard morphometric variables. The report identifies this issue as the Zagros Generalization Problem and suggests that missing geological variables—particularly lithology, carbonate distribution, fracture density and karstification—may be responsible for the reduced model transferability.
Leave-One-Region-Out validation experiments showed that the ReliabilityGate framework reduced mean cross-regional error by approximately five percent. In the Ravar region, the reported coefficient of determination increased from 0.689 to 0.776 after applying the mixture framework.
Scientific and Practical Applications
The resulting High-Resolution Terrain Ruggedness Mapping framework can support several environmental and geoscientific applications, including:
- Soil erosion assessment
- Hydrological response modeling
- Watershed prioritization
- Geomorphological susceptibility analysis
- Environmental and land management
- Infrastructure planning
- National-scale geospatial decision-making
The final products include continuous K-coefficient maps for Zagros, Ravar and Moghan, as well as standardized physical-range maps with K values between 0.01 and 0.65.
These standardized outputs facilitate consistent interpretation and comparison of terrain conditions across regions with different environmental characteristics.
Report Information
Title: Development and Validation of a Hybrid Remote Sensing and Machine Learning Framework for High-Resolution Terrain Ruggedness (K-Coefficient) Mapping Across Heterogeneous Geomorphological Regions
Authors: Hamid Nazari, Behrad AkbarPour, Golchehreh Farahani, Jalal Karami, Saeed Arefipour and Ramin Abbasi
Employer: Research Institute for Earth Sciences
In cooperation with: UNESCO Chair on Coastal Geo-Hazard Analysis
Chairholder: Hamid Nazari
Head of the Executive Council: Morteza Talebian
Publisher: Khazeh Publication
First Edition: 2026
Number of Pages: 94
ISBN: 978-622-1582-07-5
Download the Full Scientific Report