全球PM2.5浓度（1973-2024）

1）数据内容：1973-2024年全球陆地逐月0.25度的PM2.5浓度。 2）数据来源及加工方法：利用地面能见度观测、气象再分析、地理属性和其他辅助数据，基于机器学习方法（LightGBM）建立了一个虚拟的全球PM2.5监测网络，并使用面积加权的地理加权回归生成网格浓度数据集。 3）数据质量描述：训练结果（2005-2020）的相关系数、MAE和RMSE分别为0.97、2.08μg m-3和5.36μg m-3。对历史（2000-2004）和未来（2021-2022）情景的验证的相关系数分别为0.85和0.92。插值模型的相关系数为0.97，RMSE为5.28μg m-3，MAE为3.18μg m-3。误差分析表明1973-2022年期间不确定性较小且稳定，误差为1.95-2.32μg m-3。 4）数据应用成果及前景：补充无观测地区的历史PM2.5数据，提供长期的高时空分辨率的PM2.5浓度数据，为细颗粒物的时空变化和对环境、健康和气候变化的影响的分析提供数据支持。

1) Data Content: Monthly global land-surface PM2.5 concentration data at 0.25° spatial resolution from 1973 to 2022. 2) Data Source and Processing Method: Using ground-based visibility observations, meteorological reanalysis data, geographic attributes, and other auxiliary datasets, a virtual global PM2.5 monitoring network was established based on the machine learning method LightGBM, and a gridded concentration dataset was generated using area-weighted geographically weighted regression. 3) Data Quality Description: The correlation coefficient, MAE, and RMSE of the training results (2005–2020) are 0.97, 2.08 μg m⁻³, and 5.36 μg m⁻³, respectively. The correlation coefficients of validation for historical (2000–2004) and future (2021–2022) scenarios are 0.85 and 0.92, respectively. The interpolation model achieves a correlation coefficient of 0.97, RMSE of 5.28 μg m⁻³, and MAE of 3.18 μg m⁻³. Error analysis indicates that the uncertainty during 1973–2022 is small and stable, with an error range of 1.95–2.32 μg m⁻³. 4) Data Application Achievements and Prospects: This dataset supplements historical PM2.5 data for regions without in-situ observations, provides long-term PM2.5 concentration data with high spatiotemporal resolution, and supports analyses of the spatiotemporal variations of fine particulate matter and their impacts on the environment, human health, and climate change.

搜集汇总

背景与挑战

背景概述

该数据集提供了1973-2022年全球陆地逐月0.25度的PM2.5浓度数据，基于机器学习方法生成，具有高时空分辨率和较小的误差范围，适用于细颗粒物时空变化及其对环境、健康和气候变化影响的分析。