Development of a Satellite Imagery Dataset Based on KOMPSAT-3 and Sentinel-2 for Amazon Deforestation Monitoring

Abstract

Deforestation in the Amazon Rainforest remains a critical environmental threat that accelerates global warming and ecological destruction, necessitating advanced monitoring solutions. While satellite-based remote sensing has become an essential tool for tracking these changes, the accuracy of detection models often depends on the spatial resolution and quality of the training data. This study presents a curated multi-resolution, multi-modal satellite imagery patch dataset (IDR/1434) designed to support robust deforestation monitoring. The dataset integrates high-resolution KOMPSAT-3 imagery with multi-spectral Sentinel-2 data and labels sourced from the MultiEarth Workshop. The 10 m Sentinel-2 binary labels were resampled to the KOMPSAT-3 patch dimensions for array-level comparison; this step harmonizes image size but does not create new high-resolution label information. The dataset is systematically organized using a comprehensive file-naming convention that includes sensor types, timestamps, and spatio-temporal metadata to facilitate data matching. Quantitative evaluation using image-level metrics, such as structural similarity and deep feature similarity, indicates moderate structural and semantic consistency, while the interpretation is limited by the inherent resolution and spectral gaps between sensors. By providing manually refined KOMPSAT-3 high-resolution labels and native/ resampled Sentinel-2 label products, this dataset serves as a valuable benchmark for research in satellite image super-resolution and cross-domain adaptation for environ - mental monitoring.