LGHAP v2: Global daily 1-km gap-free PM2.5 grids (2014)

A Long-term Gap-free High-resolution Air Pollutants concentration dataset (abbreviated as LGHAP) is of great significance for environmental management and earth system science analysis. In the current release of LGHAP dataset (LGHAP v2), we provide 22-year-long gap free aerosol optical depth (AOD) and near-surface PM2.5 concentrations with daily 1-km resolution covering the global land area from 2000 to 2021. Leveraging an improved big earth data analytic framework with attention-reinforced tensor construction and adaptive background information updating schemes, gap-free AOD grids were firstly derived via an integration of multimodal AODs and air quality measurements acquired from diverse satellites, ground monitors, and numerical models. For better predicting PM2.5 concentration across the globe, a scene-aware ensemble learning graph attention network (SCAGAT) was then developed to account for large modeling bias over regions with limited or even none in situ air quality measurements. These datasets were archived in the NetCDF (nc) format, while data in every year were archived as an individual submission. Python, MATLAB, R, and IDL codes were also provided to help users read and visualize the LGHAP v2 data.长期无间隙高分辨率空气污染物浓度数据集（缩写为LGHAP）对于环境管理与地球系统科学分析具有重要意义。在本次发布的LGHAP数据集（LGHAP v2）中，我们提供了2000年至2021年覆盖全球陆地区域、每日更新、分辨率为1公里的22年无间隙气溶胶光学厚度（aerosol optical depth, AOD）与近地表PM2.5浓度数据。研究团队借助改进且融合了注意力增强张量构建与自适应背景信息更新策略的大型地球数据分析框架，首先通过整合多模态气溶胶光学厚度数据与来自多颗卫星、地面监测站点及数值模式获取的空气质量观测数据，生成了无间隙气溶胶光学厚度格网数据集。为了在全球范围内更精准地预测近地表PM2.5浓度，我们进一步开发了场景感知集成学习图注意力网络（scene-aware ensemble learning graph attention network, SCAGAT），以解决在原位空气质量观测数据匮乏甚至缺失的区域存在的显著模型偏差问题。本数据集以网络通用数据格式（NetCDF, nc）存储，且按年份将每年的数据单独归档为独立文件。同时还提供了Python、MATLAB、R及IDL代码，以帮助用户读取并可视化LGHAP v2数据集。