Super-resolution is essential for improving images in computer vision and image processing. When using super-resolution for images of natural landscapes, some challenges are encountered, related to factors such as degradation present in the real world with different levels of illumination and the presence of small details and noise that make the process of applying and reconstructing super-resolution more difficult. With the advances in Deep Learning techniques, Real-ESRGAN has stood out in the transformation of images from low to high resolution, and can be used to elucidate the challenges encountered in the super-resolution of natural landscapes. In this sense, this research applies Real-ESRGAN to images of natural landscapes with realistic degradation, with the aim of generating super-resolution images of real scenarios. Using the DIV2K, Landscape Pictures and Landscape Classification datasets, four training sessions were carried out, varying the iterations and adjusting hyperparameters. The inclusion of datasets focused on landscapes, in addition to DIV2K, and enriched the database, optimizing the model. A quantitative analysis was carried out, using MSE, PSNR, SSIM and NIQE to evaluate performance. The best experimental results achieved high-quality images with an MSE of 0.029, an NIQE of 2.5566, a PSNR of 22.43 and an SSIM of 0.525, preserving original details and structures. A qualitative analysis was also carried out to assess the visual characteristics of the images, confirming that the results generated achieved an improvement in visual quality. The results indicate that the Real-ESRGAN methodology, based on landscape-oriented datasets, is effective in improving image quality in a consistent and robust manner.