跳到正文
Environmental Science & Technology· Shang·· 6 天前AI 评分64

机器学习框架重建中国逐小时 5 km 无缝气溶胶光学厚度

Near-Real-Time Mapping of Seamless All-Hour Aerosol Optical Depth in China

AI 导读

一项发表于 Environmental Science & Technology 的研究构建自动化机器学习框架,融合网格化能见度数据、预报 AOD 与时空协变量,在中国近实时重建逐小时 5 km 分辨率的气溶胶光学厚度(AOD)。

AI 深度导读基于原文整理的分节解读
背景
气溶胶光学厚度(AOD)是评估气溶胶气候强迫与空气质量的核心参数,但持续云覆盖和夜间卫星反演的缺失,使全天候、无缝隙的AOD监测难以实现。
做法
作者构建了一个自动化机器学习框架,把新近可用的格点化能见度数据集与预报AOD及时空协变量融合,在中国区域以小时、5 km分辨率近实时重建AOD。模型以静止卫星观测训练,并用MODIS与AERONET夜间地基观测做独立验证。
结论
交叉验证中模型表现稳健,Pearson相关系数R=0.95,优于再分析和预报产品,并能填补云遮挡与黑夜造成的缺口。独立验证中MODIS R=0.61、AERONET夜间R=0.76。重建结果呈现出午间最低、清晨最高的日变化,并记录了2020至2022年逐年下降;在一次破纪录沙尘暴中,框架追踪了云遮挡与夜间阶段的气溶胶演变。
意义
该工作提供了高分辨率、低延迟的昼夜气溶胶连续记录,可服务于空气质量响应、资料同化与日变化气溶胶—气候研究。其精度在独立验证中低于训练验证,实际应用中的不确定性仍需结合具体场景评估。

原文摘要

出版方登记 · Crossref

Aerosol optical depth (AOD) is central to assessing aerosol climate forcing and air quality, yet persistent cloud cover and the absence of nighttime satellite retrievals preclude seamless, all-hour monitoring. Here, we develop an automated machine-learning framework that fuses a newly available gridded visibility data set with forecast AOD and spatiotemporal covariates to reconstruct AOD over China at an hourly 5 km resolution in near-real-time. Trained on geostationary satellite observations, the model achieves robust performance (Pearson correlation coefficient R = 0.95 in cross-validation) and outperforms reanalysis and forecast products while filling cloud- and darkness-induced gaps. Independent validation against MODIS and ground-based nighttime Aerosol Robotic Network (AERONET) measurements yields consistent accuracy (MODIS R = 0.61 and AERONET R = 0.76), confirming physical generalizability. The resulting seamless record exposes a pronounced diurnal cycle with a midday minimum and an early morning maximum, and documents a consistent year-on-year decline from 2020 to 2022 that aligns with China’s clean-air actions and COVID-19 lockdowns. During a record-breaking dust storm, the framework tracks aerosol dynamics during cloud-obscured and nocturnal phases, capturing an uninterrupted process-level account of the event. By delivering the day–night aerosol continuum at high resolution and low latency, this work bridges a critical observational gap in air-quality response, data assimilation, and diurnal aerosol-climate research.

页面未能读取全文,以上为出版方登记的摘要;方法与图表见原文

本页是 AI 精读,不转载论文原文;完整论文(含方法、图表与参考文献)见原文

来源:Environmental Science & Technology · doi.org