A methodology has been developed aiming to classify the vegetation patches in the Antarctic Peninsula (AP) using multiple pixel scales. These scales range from the cm scale of the unmanned aerial vehicles (UAV) images collected in the field, the 2 m scale of very high-resolution images from platforms such as Worldview and Quickbird to the 10 m and 30 m scale of the high-resolution images of the Sentinel and Landsat sensors, respectively. This multiscale methodology was developed and validated on Barton peninsula (king George Island) for a period where all pixel scale imagery is available. The detection of vegetation on contemporary images of a test area allowed for a better choice of classifiers and quantification of total area of vegetation detected as function of the scale factor. The classifiers chosen were grouped into three categories: For images with pixel sizes far lower than the average vegetation patch dimension, an object based support vector machine was selected; for the images with pixel sizes of the same order of magnitude as the average vegetation patch size, a pixel based deep learning classifier with a Resnet backbone was selected; lastly, for images with pixel sizes larger than the average vegetation patch, a linear spectral unmixing was chosen. Now, the proposed task is to use the classifiers and parameters achieved in the methodology development phase to quantify the vegetation cover for different areas across the AP throughout the historic coverage of satellite images. For this, the Landsat collection is the one that offers the largest coverage, spanning more than three decades. The areas were chosen for having at least one UAV image collected in the field for validation. The temporal comparison of vegetation cover in these areas will help determine the effect of climate on such a pristine environment, using vegetation as a proxy for climate change.