In Press (Spring)
Back to the articles list |
Back to browse issues page
Faculty of Geography, Department of Physical Geography, University of Tehran, Tehran, Iran , goorabi@ut.ac.ir
Abstract: (53 Views)
1. Introduction
Sand dune morphodynamics, shaped by interactions between wind regimes, sediment availability, topography, and sparse vegetation, are key drivers of desert landscape evolution and indicators of desertification processes (Ashrafzadeh et al., 2017; Ehsani & Foroutan, 2014). The Rig-e Yalan region, situated in the hyper-arid Lut Desert of eastern Iran (Figure 1), represents one of the most extreme natural laboratories globally, with surface temperatures exceeding 70 °C (Mildrexler et al., 2011) and diverse dune morphologies, including barchans, linear ridges, and star dunes (Milani et al., 2021). Its enclosed topography, concentrated wind systems, and minimal human disturbance provide an exceptional setting for automated morphodynamic modeling. Traditional field surveys face challenges in this harsh environment, highlighting the importance of remote sensing, cloud-computing, and machine learning approaches (Goorabi & Yamani, 2025). The objectives of this study are: (1) to preprocess and integrate multi-sensor datasets; (2) to derive geomorphometric and spectral indices; (3) to classify homogeneous morphodynamic units using supervised and unsupervised models; and (4) to validate the classification framework for monitoring desertification and informing adaptive land management.
2. Methodology
A multi-sensor workflow was implemented in Google Earth Engine (GEE) using datasets from Sentinel-2 (10 m), Landsat-8 (30 m), Sentinel-1 SAR (10 m), and topographic data from SRTM (30 m) and ALOS PALSAR (12.5 m). Preprocessing included atmospheric correction, cloud/shadow masking (QA60, BQA), geometric and terrain correction. In total, 27 indices were extracted (Table 1), including 13 geomorphometric indices (slope, aspect, Topographic Position Index [TPI], Terrain Ruggedness Index [TRI]), 11 spectral indices (NDVI, NDWI, Bare Soil Index [BSI], etc.), and 3 Sentinel-2 vegetation indices (NDRE). Dimensionality reduction was achieved via Principal Component Analysis (PCA), where PC1–PC3 captured most variance. Two new composite metrics—the Morphodynamic Activity Index (MAI) and Integrated Dune Mobility Score (IDMS)—were developed (Table 2). Classification combined a Random Forest (RF) algorithm with 100 trees and 5-fold cross-validation, alongside K-Means clustering (K=5, Elbow Method). The COSI-Corr algorithm was applied to multi-temporal imagery for sub-pixel dune migration analysis. Validation used Google Earth Pro imagery and synthetic datasets, evaluated via overall accuracy (OA), Kappa coefficient, and confusion matrices.
3. Results
The RF model achieved an OA of 91.14% and a Kappa of 0.88, successfully delineating six morphodynamic units: active dunes (15.5%), semi-active dunes (20.8%), sandy surfaces (36.3%), hard surfaces (12.5%), wind-eroded areas (8.4%), and mixed terrains (6.5%) (Figure 2). High TRI and TPI values in the southern sector indicated stronger aeolian activity. The MAI (0.847) and IDMS (0.923) confirmed Rig-e Yalan’s extreme dune mobility, driven by negligible vegetation (mean NDVI = 0.086), elevated bare soil exposure (mean BSI = 0.180), and rugged terrain (mean TRI = 0.802). Tasseled Cap analysis revealed high brightness (mean = 0.350), near-zero greenness (–0.025), and minimal wetness (0.025), consistent with hyper-arid conditions (Figure 3). The Desertification Risk Index (DRI = 0.756) and Ecological Sensitivity Index (ESI = 0.78) highlighted severe environmental vulnerability. Pearson correlations showed a strong negative NDVI–BSI relationship (r = –0.82) and positive slope–TRI relation (r = 0.73), validating consistency. K-Means clustering corroborated the RF classification (Figure 4). COSI-Corr revealed dune migration, particularly in active dune zones, confirming dynamic morphodynamics (Figure 5).
4. Discussion & Conclusions
This study demonstrates a robust, automated framework for classifying and monitoring dune morphodynamics in hyper-arid landscapes. By integrating multi-sensor datasets and advanced machine learning, it overcomes challenges of spectral similarity and lack of field data. The RF model achieved high accuracy (91.14%, Kappa 0.88), while novel indices (MAI = 0.847, IDMS = 0.923) provide scalable tools for quantifying dune mobility, supporting desertification monitoring (Miszalski et al., 2023; Blumstein & MacManes, 2023). The use of GEE ensured efficient, reproducible analysis across large spatio-temporal datasets. Findings confirm Rig-e Yalan’s exceptional morphodynamic activity and ecological fragility (ESI = 0.78), underscoring its UNESCO World Heritage significance. The framework can be extended to other deserts (e.g., Sahara, Gobi) for global comparative studies. Limitations include absence of in-situ validation and reliance on GEE’s processing infrastructure. Future work should incorporate targeted field campaigns, test deep learning approaches, and establish global dune dynamics databases. These results provide actionable insights for climate adaptation, sustainable land management, and regional policy-making.
Sand dune morphodynamics, shaped by interactions between wind regimes, sediment availability, topography, and sparse vegetation, are key drivers of desert landscape evolution and indicators of desertification processes (Ashrafzadeh et al., 2017; Ehsani & Foroutan, 2014). The Rig-e Yalan region, situated in the hyper-arid Lut Desert of eastern Iran (Figure 1), represents one of the most extreme natural laboratories globally, with surface temperatures exceeding 70 °C (Mildrexler et al., 2011) and diverse dune morphologies, including barchans, linear ridges, and star dunes (Milani et al., 2021). Its enclosed topography, concentrated wind systems, and minimal human disturbance provide an exceptional setting for automated morphodynamic modeling. Traditional field surveys face challenges in this harsh environment, highlighting the importance of remote sensing, cloud-computing, and machine learning approaches (Goorabi & Yamani, 2025). The objectives of this study are: (1) to preprocess and integrate multi-sensor datasets; (2) to derive geomorphometric and spectral indices; (3) to classify homogeneous morphodynamic units using supervised and unsupervised models; and (4) to validate the classification framework for monitoring desertification and informing adaptive land management.
2. Methodology
A multi-sensor workflow was implemented in Google Earth Engine (GEE) using datasets from Sentinel-2 (10 m), Landsat-8 (30 m), Sentinel-1 SAR (10 m), and topographic data from SRTM (30 m) and ALOS PALSAR (12.5 m). Preprocessing included atmospheric correction, cloud/shadow masking (QA60, BQA), geometric and terrain correction. In total, 27 indices were extracted (Table 1), including 13 geomorphometric indices (slope, aspect, Topographic Position Index [TPI], Terrain Ruggedness Index [TRI]), 11 spectral indices (NDVI, NDWI, Bare Soil Index [BSI], etc.), and 3 Sentinel-2 vegetation indices (NDRE). Dimensionality reduction was achieved via Principal Component Analysis (PCA), where PC1–PC3 captured most variance. Two new composite metrics—the Morphodynamic Activity Index (MAI) and Integrated Dune Mobility Score (IDMS)—were developed (Table 2). Classification combined a Random Forest (RF) algorithm with 100 trees and 5-fold cross-validation, alongside K-Means clustering (K=5, Elbow Method). The COSI-Corr algorithm was applied to multi-temporal imagery for sub-pixel dune migration analysis. Validation used Google Earth Pro imagery and synthetic datasets, evaluated via overall accuracy (OA), Kappa coefficient, and confusion matrices.
3. Results
The RF model achieved an OA of 91.14% and a Kappa of 0.88, successfully delineating six morphodynamic units: active dunes (15.5%), semi-active dunes (20.8%), sandy surfaces (36.3%), hard surfaces (12.5%), wind-eroded areas (8.4%), and mixed terrains (6.5%) (Figure 2). High TRI and TPI values in the southern sector indicated stronger aeolian activity. The MAI (0.847) and IDMS (0.923) confirmed Rig-e Yalan’s extreme dune mobility, driven by negligible vegetation (mean NDVI = 0.086), elevated bare soil exposure (mean BSI = 0.180), and rugged terrain (mean TRI = 0.802). Tasseled Cap analysis revealed high brightness (mean = 0.350), near-zero greenness (–0.025), and minimal wetness (0.025), consistent with hyper-arid conditions (Figure 3). The Desertification Risk Index (DRI = 0.756) and Ecological Sensitivity Index (ESI = 0.78) highlighted severe environmental vulnerability. Pearson correlations showed a strong negative NDVI–BSI relationship (r = –0.82) and positive slope–TRI relation (r = 0.73), validating consistency. K-Means clustering corroborated the RF classification (Figure 4). COSI-Corr revealed dune migration, particularly in active dune zones, confirming dynamic morphodynamics (Figure 5).
4. Discussion & Conclusions
This study demonstrates a robust, automated framework for classifying and monitoring dune morphodynamics in hyper-arid landscapes. By integrating multi-sensor datasets and advanced machine learning, it overcomes challenges of spectral similarity and lack of field data. The RF model achieved high accuracy (91.14%, Kappa 0.88), while novel indices (MAI = 0.847, IDMS = 0.923) provide scalable tools for quantifying dune mobility, supporting desertification monitoring (Miszalski et al., 2023; Blumstein & MacManes, 2023). The use of GEE ensured efficient, reproducible analysis across large spatio-temporal datasets. Findings confirm Rig-e Yalan’s exceptional morphodynamic activity and ecological fragility (ESI = 0.78), underscoring its UNESCO World Heritage significance. The framework can be extended to other deserts (e.g., Sahara, Gobi) for global comparative studies. Limitations include absence of in-situ validation and reliance on GEE’s processing infrastructure. Future work should incorporate targeted field campaigns, test deep learning approaches, and establish global dune dynamics databases. These results provide actionable insights for climate adaptation, sustainable land management, and regional policy-making.
Keywords: Aeolian processes, Geomorphometry, Geomorphometric indices, Random Forest, Google Earth Engine
Type of Study: Research |
Received: 2025/08/1
Received: 2025/08/1
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
