Global PM2.5 Dataset: Hybrid Calibration of CAMS and MERRA-2 PM2.5 Reanalysis Products during 2017-2019

By integrating two reanalysis PM_2.5 products (CAMSRA, MERRA-2), a <strong>global daily hybrid-calibrated PM_2.5 concentration dataset</strong> is generated through the proposed CM-HC scheme with ERT model. This products include <strong>1095</strong> global PM_2.5 GeoTIFF files, starting from Jan 01, 2017 to Dec 31, 2019 (about <strong>0.95GB</strong> memory after uncompressing this zip file). To prove the superiority of this product, some experiments are implemented as follows: comparing with 1) two original products; 2) results of two separate calibration schemes with ERT; 3) results of CM-HC with other three ML models (RF, GBDT, XGBoost). Above analyses include two aspects, that is, the accuracy results and mapping effects. More details can be viewed at the thesis. One <strong>GeoTIFF</strong> file includes one-day calibrated PM_2.5 data. It is worth noting that the files contain the data of ocean area, however, our thesis only shows the results of land area in order to visually display the mapping effect before and after calibrating. Also, there is no site on the sea to validate in our study.本研究通过融合两款再分析细颗粒物（PM₂.₅）产品（CAMSRA、MERRA-2），结合极端随机树（ERT）模型与所提出的CM-HC方案，构建而成全球逐日混合校准细颗粒物浓度数据集。该数据集包含1095份全球细颗粒物（PM₂.₅）地理标记图像文件格式（GeoTIFF）文件，时间跨度为2017年1月1日至2019年12月31日；解压该压缩包后，数据集总占用存储空间约0.95GB。 为验证该数据集的性能优势，本研究开展了如下对比实验：1）与两款原始再分析产品进行对标；2）与仅采用极端随机树（ERT）的两种独立校准方案的输出结果进行对比；3）与基于CM-HC方案结合其余三种机器学习（ML）模型（随机森林RF、梯度提升决策树GBDT、极限梯度提升XGBoost）得到的结果进行对比。上述分析从精度指标与空间制图效果两个维度展开，更多研究细节可参见本相关学术论文。 每份GeoTIFF文件存储单日校准后的细颗粒物浓度数据。值得注意的是，数据集文件涵盖海洋区域的细颗粒物数据，但本研究仅展示陆地区域的分析结果，以便更直观地呈现校准前后的空间制图效果差异。此外，本研究未布设海洋站点用于模型验证。