Scholarlink

<p>In regions lacking visibility monitoring sites (non-site), the timely detection and early warning of long-term changes in visibility are significantly hindered, posing considerable risks to human safety and infrastructure development. Accurate and timely seasonal warnings are essential to reduce visibility hazards. Here, we develop a seasonal visibility forecast model that takes into account the geographical distance (GDAI). The model is based on an encoder-decoder architecture, integrates geographic distance information, and makes full use of future and historical visibility information and meteorological data to achieve forecasts up to 12&#xa0;months in advance. The GDAI model consistently outperformed the basic AI model (BAI) across all lead times, demonstrating superior forecasting capabilities. Specifically, GDAI achieved an average increase of 23.06% in R<sup>2</sup>, while the RMSE and MAE were reduced by 22.63% and 40.11%, respectively. GDAI achieves more accurate 12-month visibility forecasts in non-site areas. The average forecast error is reduced by 0.61–1.10&#xa0;km. GDAI is more reliable than the forecast accuracy of BAI. Verification results in different regions and climate zones show that the accuracy of GDAI is better than that of BAI at more than 70% of the sites. This means that the AI model considering geographic distance can provide earlier visibility seasonal warnings in non-site areas, highlighting the key role of geographic distance in seasonal forecasts. This research not only provides scientific reference for human travel, but also opens up new methods and perspectives for the accurate and seasonal forecasting of major disasters such as global fog seasons and sandstorms.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Global Non-site Regional Visibility Seasonal Forecast Integrating Geographical Distance

  • 期刊论文

  • 发表日期：

    2025年8月25日

查看全文
  • Shuaiwen Wu,
  • Siyu Chen,
  • Yaping Shao,
  • Yue Zhang,
  • Wenhan Li,
  • Shasha Qiao

摘要

In regions lacking visibility monitoring sites (non-site), the timely detection and early warning of long-term changes in visibility are significantly hindered, posing considerable risks to human safety and infrastructure development. Accurate and timely seasonal warnings are essential to reduce visibility hazards. Here, we develop a seasonal visibility forecast model that takes into account the geographical distance (GDAI). The model is based on an encoder-decoder architecture, integrates geographic distance information, and makes full use of future and historical visibility information and meteorological data to achieve forecasts up to 12 months in advance. The GDAI model consistently outperformed the basic AI model (BAI) across all lead times, demonstrating superior forecasting capabilities. Specifically, GDAI achieved an average increase of 23.06% in R2, while the RMSE and MAE were reduced by 22.63% and 40.11%, respectively. GDAI achieves more accurate 12-month visibility forecasts in non-site areas. The average forecast error is reduced by 0.61–1.10 km. GDAI is more reliable than the forecast accuracy of BAI. Verification results in different regions and climate zones show that the accuracy of GDAI is better than that of BAI at more than 70% of the sites. This means that the AI model considering geographic distance can provide earlier visibility seasonal warnings in non-site areas, highlighting the key role of geographic distance in seasonal forecasts. This research not only provides scientific reference for human travel, but also opens up new methods and perspectives for the accurate and seasonal forecasting of major disasters such as global fog seasons and sandstorms.

Graphical abstract