Tree Cover Classification in North Korea from Sentinel-2 Time Series (2019–2024)

This dataset provides annual tree cover classification maps of North Korea from 2019 to 2024, derived from Sentinel-2 time-series imagery. The maps were generated using a deep learning framework that integrates a ConvLSTM-based autoencoder for time-series reconstruction with a U-Net segmentation model. This design improves robustness to cloud contamination and missing observations, ensuring consistent classification under challenging conditions. Validation was conducted for the year 2021 using ESA WorldCover 2021 as the reference label, supplemented by independent visual interpretation of high-resolution imagery. Results demonstrated an overall accuracy above 96% for tree cover classification in 2021. The dataset further captures annual tree cover dynamics in North Korea, revealing a general increase in tree cover from 2019 to 2024, consistent with reported afforestation efforts and most previous remote sensing–based observations. This dataset is intended to support research on forest monitoring, ecological restoration, and carbon accounting in regions with limited ground-based observations.