全球500米分辨率“类NPP-VIIRS”夜间灯光数据集(第二版)(1992-2024年)
收藏资源简介:
该数据为全球“类NPP-VIIRS”夜间灯光数据集(第二版),可用时间序列为1992年至2024年。通过改进Attention U-Net模型设计了一种带有跳跃连接的超分辨率Attention U-Net模型实现了夜间灯光数据跨传感器(DMSP-OLS和NPP-VIIRS)校正方案。模型主要包含编码、解码和注意力门残差连接三个过程,其中编码是将图像下采样获取丰富的细节特征,特征经过解码重建高分辨率影像,编码特征与解码影像结合经过注意门可以减少无用信息。将1992年至2012年的EANTLI输入至训练后的模型中,获得对应年份的第二版“类NPP-VIIRS”数据(NPP-VIIRS-like NTL Data Version 2),之后通过合成2013年至2024年的月度NPP-VIIRS夜间灯光数据获得对应年份的年合成产品,从而生产出1992年至今的第二版“类NPP-VIIRS”数据集。
This dataset is the global "NPP-VIIRS-like" Nighttime Light (NTL) Dataset Version 2, with a valid time series covering 1992 to 2024. An enhanced Super-resolution Attention U-Net model with skip connections was developed to realize a cross-sensor (DMSP-OLS and NPP-VIIRS) correction framework for nighttime light data. The model comprises three core modules: encoder, decoder, and attention-gated residual connections. The encoder downsamples input images to extract rich detailed features; the decoder reconstructs high-resolution imagery from the encoded features; and the fusion of encoder features and decoded outputs via attention gates helps mitigate irrelevant information. Specifically, the Enhanced Annual Nighttime Light Index (EANTLI) data from 1992 to 2012 was fed into the pre-trained model to generate the Version 2 "NPP-VIIRS-like" nighttime light data for each corresponding year. Thereafter, annual composite products for the years 2013 to 2024 were derived by synthesizing monthly NPP-VIIRS nighttime light datasets of these years. Ultimately, the global Version 2 "NPP-VIIRS-like" NTL dataset spanning 1992 to the present was constructed.




