VegGAN: A Generative Adversarial Network for Downscaling JPSS/VIIRS Vegetation Indices

Vegetation plays a crucial role in various fields, including meteorology and agriculture. Satellite sensors are widely used for vegetation monitoring, and numerous vegetation index (VI) products have been developed based on satellite observations. Currently, the Visible Infrared Imaging Radiometer Suite (VIIRS) aboard the National Oceanic and Atmospheric Administration’s (NOAA) Joint Polar Satellite System (JPSS) provides global coverage through three satellites (Suomi-NPP, NOAA-20, and NOAA-21), each with a 16-day revisit time. Despite its widespread use, accurately monitoring VI using VIIRS data is challenging due to its limited spatial resolution. In this article, we propose a deep-learning-based approach to downscale VIIRS VI products using a super-resolution generative adversarial network (GAN) to enhance their spatial resolution. Specifically, the proposed vegetation GAN (VegGAN) consists of a generator and a discriminator, employing adversarial training to improve performance, where the low-resolution VI data from VIIRS serve as input to the VegGAN, and the high-resolution VI data from Landsat satellites are used as target labels to optimize the network’s parameters. The comprehensive experimental results demonstrate that the proposed VegGAN significantly improves the spatial resolution of the VI products derived from VIIRS in the selected study domain. In addition, the framework provides a synthetic VI product with both high temporal and spatial resolutions, demonstrating its potential to improve satellite-based vegetation monitoring.