Abstract


Background and Aims: In Synthetic Aperture Radar (SAR) images, the backscatter value is significantly affected in areas with complex topography. This issue leads to a reduction in the radiometric accuracy of the images and causes phenomena such as foreshortening and layover. Such effects degrade the quality and reliability of information extractable from radar imagery, making their accurate correction essential for scientific applications. In this regard, numerous models have been developed to rectify these effects, which are generally classified into three main categories: models based on the local incidence angle, models based on pixel area correction, and hybrid models. Each of these approaches has achieved satisfactory results in backscatter correction according to its structure and performance. The present study aims to develop a novel model with high accuracy for backscatter correction in Sentinel-1 images, particularly in forested and mountainous regions.


Materials and Methods: In this study, a backscatter correction model was developed by combining two different models. First, the RTF (Radiometric Terrain Flattening) model was applied due to its favorable performance in correcting backscatter values in areas with severe topography. However, after applying this model, some areas located behind the elevations, which could not be accurately modeled, were eliminated. Subsequently, the sinusoidal model was applied to the image corrected by the RTF model. Given the suitable performance of this model in correcting backscatter in areas with gentle topography, its application was restricted to pixels with a local incidence angle exceeding 13 degrees. Finally, the pixels eliminated by the application of the RTF model on the descending pass image were replaced with their corresponding values from the ascending pass image. To investigate the generalizability of the proposed model, introduced as the "Improved RTF model", it was implemented not only in the Kheyrudkenar region of Mazandaran province but also in three other areas within the Dalkhani forests. To evaluate the model's performance, two statistical indices variance reduction and the reduction of the regression slope between the local incidence angle and backscatter as well as a tree species classification method were used.


Results and Discussion: In the image of the Kheyrudkenar region, the backscatter variance in VV polarization was reduced by 86.1% after correction with the RTF model and by 91.6% after applying the Improved RTF model, relative to the original image. Furthermore, in VH polarization, the variance decreased by 90% following the RTF model correction and by 93.4% with the Improved RTF model. The slope of the regression line between the local incidence angle and backscatter in VV polarization decreased from -0.00201 for the RTF model to -0.00011 for the improved model, and in VH polarization, it was reduced from -0.00033 to -0.00016, indicating a significant reduction in topographic effects. Additionally, the overall accuracy and Kappa coefficient of the tree species classification increased from 47% and 0.18 with the RTF model to 52% and 0.26 with the Improved RTF model, respectively. To assess the model's generalizability, the two statistical evaluation methods were applied to three other regions in the Dalkhani forests. The obtained results also demonstrated a significant improvement in backscatter correction in these areas.


Conclusion: The Improved RTF model, by combining the advantages of both the RTF and sinusoidal models and simultaneously utilizing ascending and descending Sentinel-1 imagery, has provided a remarkable improvement in backscatter value correction. This new model not only exhibits higher accuracy than the individual models but also significantly enhances the quality of the corrected images by mitigating the negative effects of topography. Therefore, this method can serve as a powerful tool in research related to the analysis of radar imagery, assisting researchers in achieving more accurate results.