Deep Learning Models to Estimate High Resolution NDVI for Multiple Augmentation Factors

The Normalised Difference Vegetation Index (NDVI) is a highly useful tool for monitoring vegetation, derived from remote sensing satellite data such as Sentinel-2 satellites. However, the 10m resolution provided by these satellites may not always be sufficient. While there are commercial satellites that offer high resolution images, they tend to be prohibitively expensive and thus inaccessible to many users. In response to this challenge, Deep Learning (DL) models have emerged as valuable tools for estimating high resolution images from low resolution inputs. In this context, the present study provides distinct augmentation factor results for High Resolution NDVI estimations. For this purpose, training-testing images are extracted from WorldView satellite-derived products, and the two variants of well-known DL Super-Resolution Residual Network (SRResNet) models are selected. The augmenting factors considered to enhance the NDVI results are 5, 6.25, and 8. This preliminary context aims to bridge the resolution gap between Sentinel-2 images and those provided by commercial satellites like WorldView, laying the groundwork for future improvements in satellite image resolution.