Document Type : Research Paper

Authors

• Masoud Safari ¹
• Marzban Faramarzi ²
• REza Omidpour ³
• Hassan Fathizad ⁴

¹ MSc in Desert Management and Control, Department of Rangeland and Watershed Management, Faculty of Agriculture, Ilam University, Ilam, Iran

² Associate Professor, Department of Rangeland and Watershed Management, Faculty of Agriculture, Ilam University, Ilam, Iran

³ Assistant Professor, Department of Rangeland and Watershed Management, Faculty of Agriculture, Ilam University, Ilam, Iran

⁴ PhD in Combating Desertification, Department of Management Arid and Desert Regions, College of Natural Resources and Desert, Yazd University, Iran

Abstract

Background and Objective
Vegetation cover and aboveground biomass are critical components of arid and semi-arid ecosystems, playing a vital role in environmental sustainability, carbon sequestration, biodiversity conservation, and providing forage for both domestic and wild herbivores. Due to their high sensitivity to environmental changes, these regions require accurate and continuous monitoring. In recent decades, the use of vegetation indices derived from remote sensing data has emerged as a promising approach for modeling vegetation cover and biomass in natural resource management. However, these indices have been less thoroughly evaluated in arid and semi-arid areas. This study aims to assess the efficiency of Landsat 8 and Sentinel-2 satellite imagery and various vegetation indices in modeling vegetation cover and aboveground biomass in the arid rangelands of western Iran, using linear regression methods.
Materials and Methods
This study was conducted in the arid rangelands of Ilam Province, western Iran. Field data were collected using a random-systematic sampling method across 50 plots (30 × 30 meters) during the peak growing season and prior to livestock grazing. In each plot, the percentage of live vegetation cover, litter, bare soil, and gravel was recorded. Aboveground biomass was measured by clipping vegetation at ground level and weighing the oven-dried samples. Satellite images from Landsat 8 and Sentinel-2, corresponding to the same period as the field sampling, were acquired. Both slope-based vegetation indices, such as NDVI and EVI, and distance-based indices, such as MSAVI and SAVI, were extracted using ENVI and ArcGIS software. Linear regression analysis was employed to model vegetation cover and aboveground biomass. The performance of the selected indices was evaluated using the coefficient of determination (R²) and root mean square error (RMSE).
Results
The results indicated that distance-based vegetation indices, such as MSAVI, outperformed slope-based indices in modeling vegetation cover and aboveground biomass. The MSAVI index derived from Landsat imagery provided the highest accuracy, with an R² of 0.19 for vegetation cover and 0.21 for biomass. Linear regression modeling showed that vegetation cover in the study area ranged from 5% to 81%. Additionally, the estimated aboveground biomass ranged from 6.8 to 58.4 grams per square meter (gm²). According to the results, when using slope-based indices such as NDVI, Sentinel-2 imagery yielded more accurate vegetation cover modeling compared to Landsat imagery. However, when distance-based indices were applied, Landsat imagery delivered significantly higher accuracy, while the accuracy of Sentinel-2 images decreased relative to their performance with slope-based indices.
Conclusion
The results of this study indicated that the accuracy of vegetation cover modeling is influenced by both the types of indices used and the spatial resolution of satellite data. For slope-based indices like the Normalized Difference Vegetation Index (NDVI), Sentinel imagery demonstrated higher accuracy than Landsat for estimating vegetation cover. Conversely, for distance-based indices, Landsat imagery provided better accuracy, while the precision of Sentinel imagery decreased under these conditions. These findings emphasize the importance of selecting appropriate spectral indices tailored to specific regional characteristics, as this choice can significantly impact modeling accuracy. Furthermore, the results highlight that in arid regions, distance-based indices may be more effective, as they mitigate the impact of bare soil reflectance and provide more reliable information about vegetation cover. This research offers a foundation for optimizing remote sensing methodologies in future studies.
 

Keywords

• linear regression
• Sentinel
• vegetation Index
• Landsat
• modelling