1992-2024年全球500米分辨率“类NPP-VIIRS”夜间灯光数据集（第二版）

本数据为全球“类NPP-VIIRS”夜间灯光数据集（第二版），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。该数据集利用重采样后的Landset年尺度EVI数据与经过过饱和校正和时序校正的DMSP-OLS夜间灯光数据生产EANTLI数据，首先利用2013年EANTLI数据训练模型参数；其次将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；最后通过合成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。从像元尺度、市级尺度、省级尺度与DMSP-OLS饱和区域像元验证数据精度，结果表明研究本数据集与真实NPP-VIIRS数据之间有接近1:1的关系，可以直接将其与NPP-VIIRS夜间灯光数据同时使用。数据集储存为TIFF格式，数据集由33个数据文件组成，数据量为310.5 GB。

This is the Global "NPP-VIIRS-like" Nighttime Light Dataset (Version 2). A super-resolution Attention U-Net model with skip connections was designed by improving the standard Attention U-Net, enabling a cross-sensor (DMSP-OLS and NPP-VIIRS) calibration framework for nighttime light data. This dataset generates EANTLI data using resampled annual-scale Landsat EVI data and DMSP-OLS nighttime light data that have undergone over-saturation correction and temporal correction. First, the model parameters were trained using the 2013 EANTLI data; second, the EANTLI data from 1992 to 2012 were fed into the trained model to obtain the Version 2 "NPP-VIIRS-like" data for corresponding years; finally, annual synthetic products for corresponding years were generated by synthesizing the monthly NPP-VIIRS nighttime light data from 2013 to 2024, thereby producing the Version 2 "NPP-VIIRS-like" dataset covering 1992 to the present. The data accuracy was verified at pixel, municipal, and provincial scales, as well as in saturated pixel regions of DMSP-OLS. The results demonstrate that this dataset exhibits an almost 1:1 correlation with real NPP-VIIRS data, and can be directly used concurrently with NPP-VIIRS nighttime light data. The dataset is stored in TIFF format, comprises 33 data files, and has a total size of 310.5 GB.

搜集汇总

背景与挑战

背景概述

该数据集是全球范围的夜间灯光数据，覆盖1992年至2024年，空间分辨率为500米，通过深度学习模型实现跨传感器校准，将DMSP-OLS和NPP-VIIRS数据融合生成年度产品。数据验证显示与真实NPP-VIIRS数据高度一致，适用于连续时间序列分析，总数据量为310.5 GB，以TIFF格式存储。