NUAA-CR4L8/9 dataset: A thin cloud removal dataset for Landsat 8 and 9 images

This is a thin cloud removal dataset (NUAA-CR4L8/9) for Landsat 8 and 9 images. If you find this useful, consider citing our work: [1] Li, J., Wang, Y., Sheng, Q., Wu, Z., Wang, B., Ling, X., Liu, X., Du, Y., Gao, F., Camps-valls, G., Molinier, M., 2025. CloudRuler : Rule-based transformer for cloud removal in Landsat images. Remote Sens. Environ. 328, 114913. https://doi.org/10.1016/j.rse.2025.114913 [2] Du, Y., Li, J., Sheng, Q., Zhu, Y., Wang, B., Ling, X., 2024. Dehazing Network: Asymmetric Unet Based on Physical Model. IEEE Trans. Geosci. Remote Sens. 62, 1–12. https://doi.org/10.1109/TGRS.2024.3359217 The Collection 2 Level 1 data served as the source data for the NUAA-CR4L8/9 dataset. There are 20 paired images, consisting of both cloudy and cloud-free scenes, from Landsat 8 and 9, acquired between 2022 and 2024, with an 8-day time interval for the same region in each image pair. In each image pair, if the Landsat 8 or 9 image is cloudy, the cloud-free image is chosen from the other satellite. The ratio of training and testing image pairs is set to 4:1. In this way, 16 image pairs are used for training, and four image pairs are used for testing, respectively. All the images are located in Southeast of USA. Both training and testing datasets contain different types of land cover. This makes the NUAA-CRL8/9 dataset representative.