Evaluation of the Usability of Geostationary Satellite Images for Gap Filling of Polar Orbiting Satellite-based NDVI
Sung-Joo Yoon, Han-Sol Ryu, Jinyeong Kim, Tae-Ho Kim *

Ocean Convergence Division, Underwater Survey Technology 21. Inc., Korea
* thkim[at]ust21.co.kr

Abstract

To understand vegetation responses to climate and environmental changes, long-term vegetation index analysis based on satellite imagery is used as a crucial tool for identifying spatiotemporal variability in crop growth and its causes. Polar orbiting optical satellite imagery is widely used for normalized difference vegetation index (NDVI) analysis. However, data gaps caused by clouds often hinder vegetation monitoring. To eliminate data gaps, recent studies have attempted to generate gap-free vegetation indices using synthetic aperture radar (SAR) images or spatial statistics techniques. In this study, we propose a method that utilizes the geostationary ocean color imager (GOCI)-II, capable of producing Earth observation images hourly, to generate gap-free vegetation indices of Sentinel-2. We present a conditional generative adversarial network (cGAN) model based on the U-Net architecture to transform GOCI-II images to match the spatial and spectral features of Sentinel-2 images. Using Sentinel-2 products as ground truth, the model is trained to minimize both pixel-level reconstruction error and adversarial loss, enabling the generation of transformed GOCI-II data with a resolution similar to Sentinel-2 data. The experiment was conducted on agricultural land in South Korea. The results showed that the converted GOCI-II output achieved a significant structural similarity index measure (SSIM) and low root mean square error (RMSE) compared to actual Sentinel-2 observations. Visual inspections also confirmed that the spatial texture and radiometric consistency were maintained. These results suggest that the proposed model can be utilized as an important preprocessing component in a gap-filling framework and effectively generate Sentinel-2 compatible data even under cloudy conditions. In future studies, the converted output will be used to restore cloud-covered NDVI values, enabling continuous vegetation monitoring in frequently cloud-covered regions.

Keywords: GOCI-II, Sentinel-2, gap filling data, vegetation monitoring

Topic: Topic A: General Remote Sensing