Introduction: In recent years, land cover has been recognized as a key indicator for assessing climate change, ecosystems, and the management of natural resources. Growing challenges in acquiring up-to-date and accurate data have led to the adoption of modern remote sensing technologies. Among these, the Sentinel-1 Synthetic Aperture Radar (SAR) imagery has emerged as a reliable source for land surface characterization. These images are produced by a SAR system using active microwave technology, enabling effective operation in all weather conditions. Therefore, this technology proves to be an appropriate tool for generating reliable and detailed land cover maps. The aim of this research is to improve the classification of land cover by simultaneously utilizing images acquired from both the ascending and descending orbits of Sentinel-1. In addition, the study investigates the impact of applying radiometric terrain flattening corrections on the overall accuracy of the classification results.
Materials and Methods: This study examined three different regions in Iran: Marand, Sari, and Chadegan, which were selected due to their varied land cover characteristics and the presence of mountainous areas. The data used consisted of Sentinel-1 satellite images with VV and VH polarizations from both ascending and descending passes. Preprocessing steps included applying orbit file, thermal noise removal, border noise removal, calibration, radiometric terrain flattening, speckle filtering, Range-Doppler terrain correction, and conversion to decibels. Additionally, the data were rescaled to a specific range using the min-max normalization method. The Random Forest (RF) algorithm was then employed to classify land cover into five classes: water, soil, vegetation, urban areas, and agriculture. Finally, the results were evaluated using overall accuracy, the kappa coefficient, and class-specific accuracy metrics.
Results and Discussion: The results indicate that the simultaneous use of ascending and descending images without applying radiometric terrain flattening significantly improves classification accuracy across all study areas. For example, in Marand, the overall classification accuracy increased from 65.33% to 79.17%, representing an approximate improvement of 13%. In Sari, the combination of images raised the overall accuracy from 55.67% to 75.41%, while in Chadegan, it resulted in an approximate 12% increase from 56.88% to 68.06%. Regarding class-specific accuracy, in Marand, the vegetation class improved from 43.41% to 69.64%, and in Sari, the soil class accuracy increased from 19.57% to 46.40%. Numerical analysis suggests that combining images from different orbits provides complementary perspectives of the Earth's surface, helping to reduce distortions caused by viewing angles and topography. In addition, the results reveal that while radiometric terrain flattening can enhance the accuracy of certain classes when using a single image, its application in the combination of two images may cause excessive similarity between some classes, ultimately reducing overall performance.
Conclusion: In conclusion, this research highlights the importance of concurrently using Sentinel-1 images from both ascending and descending orbits, particularly when radiometric terrain flattening is not applied, which plays a crucial role in enhancing the accuracy of land cover classification. The observed improvement in overall accuracy, ranging from 13% to 20% across different study areas, underscores the strong potential of this approach for land cover mapping. Moreover, the findings of this study demonstrate that the preprocessing methods employed for Sentinel-1 images have a significant impact on the accuracy and efficiency of classification models. In some cases, applying radiometric terrain flattening can lead to a decrease in both accuracy and efficiency. Therefore, optimally combining Sentinel-1 data from multiple orbital passes can lead to more accurate and reliable land cover maps. The approach presented in this study can thus serve as a valuable reference for future studies in the field of remote sensing, particularly those focused on improving land cover classification for environmental and agricultural applications.